Anda di halaman 1dari 29

Rey Fasha Syihab Ulwan

H0223136

Resume Jurnal Internasional

A Survey on Smart Farming Data, Applications, and Techniques

Tujuan dari survei ini adalah untuk meninjau secara komprehensif mengenai big data di dalam sistem
smart farming, pengaplikasian sistem smart farming dengan spesialisasi pada area penghasil pangan,
peninjauan pembuatan dan inovasi mesin, dan diskusi dari potensi dan tantangan smart farming sebagai
penelitian masa depan. Metodologi pengambilan data dari jurnal ini meliputi dua metode. Metode pertama
adalah menelaah artikel secara komprehensif (journal papers, conference papers, dan situs) dan dicari secara
detail melalui laman seperti Web of Science, Scopus, dan Google Scholar. Metode kedua adalah dengan
pertanyaan ilmiah sebagai kriteria utama dan dijabarkan sebagai berikut.

1. Apa saja jenis data yang dihasilkan dari smart farming?


2. Apa saja aplikasi big data dalam smart farming?
3. Apa saja teknik yang digunakan dalam analisis big data smart farming?
4. Apa saja tantangan yang masih ada dan solusi atau jalur sebagai jawaban dari tantangan dalam hal ini?

Sistem dan mekanisme smart farming ini masih baru dan memiliki banyak kekurangan dalam
pengumpulan big data dan pengaplikasian data-data tersebut dalam smart farming. Ultilisasi mesin belajar
atau Artificial Intelligence (A.I.) telah menarik perhatian dari ilmuwan-ilmuwan dalam bidang ini. Salah satu
poin penting sebagai intisari adalah pengambilan dan interpretasi data. Pembaruan dan pengembangan secara
cepat dalam pengaplikasian big data masih perlu membahas dan menjawab beberapa tantangan untuk masa
depan. Smart farming perlu mengumpulkan berbagai jenis data, seperti data iklim, data terkait media
pertumbuhan, data persiapan konservasi hayati, data tanaman dan hewan, dan data eksternal. Di samping itu,
ada data lain yang perlu diambil dan dilakukan pendekatan seperti, data manajemen tani, data penginderaan
jauh (remote sensing), data terkait pengairan, dan data genomik.

Pengaplikasian big data meliputi tiga hal berikut, 1) Prediksi hasil pertanian, 2) Analisis pertumbuhan,
dan 3) Pengendalian kualitas. Prediksi hasil pertanian mengadopsi beberapa metode atau teknik seperti,
metode pemilihan fitur, tuning mode, teknik ANN (Artificial Neural Networkl), teknik MLR (Multiple Linear
Regression), Deep Gaussian Model Process, Convolutional Neural Network (CNN), Supervised Kohonen
Networks (satellite images), and SVM (Support Vectot Machine). Analisis pertumbuhan meliputi analisis
fundamental yaitu dengan mengukur tinggi tanaman sebagai data awal. Selanjutnya, lebar tanaman dan jumlah
daun per tanaman adalah data yang diperlukan dalam domain ini. Pengendalian kualitas merupakan proses
meminimalkan bahaya dari hama dan penyakit. Pengendalian kualitas mempunyai 4 aspek utama yaitu
penilaian dan penyortiran kualitas, deteksi penyakit, deteksi dan pengendalian hama, dan deteksi gulma.
Sebagai fokus utama adalah penyiapan pertanian cerdas dengan pengaplikasian teknologi yang canggih untuk
mencapai hasil produksi yang maksimal dengan meminimalkan segala hambatan dan tantangan serta
berkelanjutan terintegrasi.

Tantangan-tantangan smart farming terbagi menjadi dua, yaitu tantangan teknis dan tantangan non-
teknis (Bacco et al., 2019). Data, jaringan, dan informasi diklasifikasikan sebagai tantangan teknis, sementara
kondisi ekonomi dan keterampilan perspektif sebagai tantangan non-teknis. Pengumpulan data merupakan
langkah awal dan masalah pertama yang menghambat perkembangan IoT (Internet of Things) pertanian.
Ketidakseimbangan data dan algoritma yang tidak cocok dari berbagai sumber data yang relevan semakin
menghambat pengumpulan data. Terlebih masalah minor seperti volume, kecepatan, variasi, dan kebenaran
merupakan masalah terpadu dalam pengumpulan data. Pengumpulan data secara real-time untuk pengambilan
keputusan cepat merupakan masalah individual petani yang bertugas. Masalah lain yang muncul dan
mendesak dengan pertanian ialah peternakan atau penanganan hewan yang dikenal sebagai livestock ranching.
Perkiraan hasil panen secara real-time yang termasuk sebagai keakuratan data cepat merupakan tantangan
penting perlu dihadapi. Kualitas data perlu dipertahankan untuk menyokong konsep berkelanjutan, baik dalam
hal pertanian itu sendiri ataupun lahan sekitar.

Tantangan lain yang menjadi konsiderasi para ilmuwan dalam pengumpulan yaitu data iklim yang
tidak menentu dan sukar untuk di prediksi. Perubahan kondisi iklim sangat berpengaruh dalam smart farming
karena bisa membuat terjadinya fluktuasi data yang seharusnya konstan, menjadi fluktuatif. Terjadinya musim
pancaroba, kekeringan dan musim penghujan yang tidak menentu atau bahkan penurunan agen penyerbuk
seperti lebah. Tantangan status sosial-ekonomi antara lain seperti kepemilikan data. Pemrosesan dan
pengelolaan kumpulan data berdimensi tinggi dan multi-modal berhubungan dengan status sosial-ekonomi
seseorang. Pencarian tenaga kerja terampil (skilled labour) juga merupakan salah satu masalah yang akan
dialami oleh pertanian cerdas, para pekerja haruslah memiliki berbagai keterampilan terkait pengolahan data
dan bisa berpikir terbuka untuk menerima inovasi atau teknologi baru.

Solusi dari berbagai masalah diatas bisa diatasi dengan melakukan penelitian tentang domain
permasalahan terkait. Praktik industri dalam menghasilkan dan mengembangkan teknologi seharusnya bekerja
sama dengan sektor pertanian. Kebijakan-kebijakan terkait smart farming sudah sewajarnya memperhatikan
nilai-nilai dari budaya, politik, dan ekologi. Perlunya melakukan pemanenan tepat waktu untuk efisiensi pupuk
yang digunakan dan meminimalisir pemborosan pasca panen sehingga memperole keuntungan yang optimal.
Analisis gambar bisa membantu untuk melakukan pemanenan tepat waktu karena menjadi patokan apakah
produk pertanian tersebut sudah siap panen atau belum. Gambaran analisis smart data diatas merupakan
esensis dari IoT dalam sektor pertanian dan sebagai langkah awal para petani bisa memulai dengan penerapan
teknologi ke dalam sektor pertanian dan para ilmuwan bisa mulai melakukan riset mendalam dalam domain
terkait.
A survey on smart farming data, applications and techniques
Sandya De Alwisa,⁎, Ziwei Houa, Yishuo Zhanga, Myung Hwan Nab, Bahadorreza Ofoghia,

Atul Sajjanhara
a
School of Information Technology, Deakin University, Burwood, Victoria 3125, Australia

b
Department of Mathematics and Statistics, Chonnam National University, Gwangju, South Korea

A R ticl E INfo A bst RACt

Article history: The Internet of Things (IoT) and the relevant technologies have had a significant impact on smart farming as a
major sub-domain within the field of agriculture. Modern technology supports data collection from IoT devices
Received 4 July 2021
through several farming processes. The extensive amount of collected smart farming data can be utilized for daily
Received in revised form 23 January 2022 decision making and analysis such as yield prediction, growth analysis, quality main- tenance, animal and
Accepted 25 January 2022 aquaculture, as well as farm management. This survey focuses on three major aspects of contemporary smart
farming. First, it highlights various types of big data generated through smart farming and makes a broad
Available online 13 February 2022 categorization of such data. Second, this paper discusses a comprehensive set of typical applications of big data in
smart farming. Third, it identifies and introduces the principal big data and machine learning techniques that
are utilized in smart farming data analysis. In doing so, this survey also identifies some of the major, current
Keywords:
challenges in smart farming big data analysis.This paper provides a discussion on potential pathways toward
Smart farming
Data analysis more effective smart farming through relevant analytics-guided decision making.
Big data
© 2022 Elsevier B.V. All rights reserved.
Machine learning
Digital farming
Predictive farming
Farming industry

Contents

1. Introduction.......................................................................................................................................................................................................................................2
2. Methodology.......................................................................................................................................................................................................................................2
3. Types of smart farming data............................................................................................................................................................................................................3
3.1. Climatic data..........................................................................................................................................................................................................................3
3.2. Growth media related data...................................................................................................................................................................................................3
3.3. Bio-security measures...........................................................................................................................................................................................................3
3.4. Crop and animal measures...................................................................................................................................................................................................4
3.5. Other smart farm-related data.............................................................................................................................................................................................4
3.6. External data..........................................................................................................................................................................................................................5
3.7. Smart farming data resources..............................................................................................................................................................................................6
4. Big data applications and machine learning models in smart farming.......................................................................................................................................6
4.1. Big Data Applications............................................................................................................................................................................................................6
4.1.1. Yield Prediction and Growth Analysis..................................................................................................................................................................6
4.1.2. Quality maintenance...............................................................................................................................................................................................7
4.1.3. Animal husbandry and aquaculture......................................................................................................................................................................7
4.1.4. Farm management..................................................................................................................................................................................................8
4.2. Machine learning models......................................................................................................................................................................................................8
4.2.1. Regression................................................................................................................................................................................................................9
4.2.2. Support vector machines........................................................................................................................................................................................9
4.2.3. Deep learning techniques.....................................................................................................................................................................................10
5. Discussion........................................................................................................................................................................................................................................10


Corresponding author.
E-mail address: sdealwi@deakin.edu.au (S.D. Alwis).

https://doi.org/10.1016/j.compind.2022.103624 0166-
3615/© 2022 Elsevier B.V. All rights reserved.
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

5.1. Challenges..............................................................................................................................................................................................................................11
5.2. Opportunities.........................................................................................................................................................................................................................11
6. Conclusions........................................................................................................................................................................................................................................12
Declaration of Competing Interest....................................................................................................................................................................................................12

Acknowledgement.............................................................................................................................................................................................................................12

References.........................................................................................................................................................................................................................................12

plant classification to separate the weed from the crop. Recently,


Kamilaris and Prenafeta-Boldú (2018b) surveyed deep learning (DL)

1.Introduction

Farming has been important in all ages of human history,not only


to feed the increasing population but also to serve other purposes
such as the production of medicine, fibre and fuel. With the new
advances in science, technology and equipment,agrochemicals and
genetically modified food have been introduced to agriculture with
an aim to achieve a high yield while minimizing the labour cost
(Ryu et al., 2015). In recent decades, Smart Farming has emerged
as an application of the Internet of Things (IoT) with the
development of information and communication technology
(Wolfert et al., 2017). Smart farming usually refers to the utilization
of digital technologies, IoT, cloud computing, robotics, sensors,
location systems and artifi- cial intelligence on the farm (Regan,
2019; Wolfert et al., 2017; Smart Farming, 2022). Hansen et al.
(2018b) defined smart farming as the “application of innovations
in engineering and technology to offer positive interventions in
farming activities.” Across different con- texts,smart farming is also
known as digital farming, data-driven farming, Agriculture 4.0 and
sometimes Agriculture 5.0. The infra- structure of the farm
consists of sensors for environmental data and surveillance
cameras.Through the gateways,those data are con- nected to the
service platforms accessible to farmers.Similar to other agriculture
technologies, smart farming aims to control and improve food
manufacture (Suebsombut et al., 2017).
In the past decade,smart farming has been developed into var-
ious domains such as large-scale open environment farming(field
crops), control environment (plant houses)and dairy and poultry
farming. Compared to traditional farming, smart farming has more
advantages with new technology, facilities and data gathered
during the farming processes. The quality of products can be
enhanced by proper monitoring. For example, any pest intrusion
or disease could be detected in advance from data collected.
Previously recorded information could be used for farm labour
force support, efficient resources management (Suebsombut et al.,
2017), limiting the use of non-essential pesticides and fertilizers
(Moon et al., 2017). This type of novel farming could even result in
sharing domain expert knowledge (Ryu et al., 2015).
To harvest qualitative products,improve the profit and make in-
formed decisions, there is a need for farmers to take full advantage
of relevant data sources, so they can benefit from the discovered
regularity, patterns, or knowledge embedded in accumulated data. It
is well known that big data in smart farming provide such an op-
portunity. Currently, agricultural practices are leveraging huge data
analysis methods to utilize smart farming data sensibly and cost-
effectively. Typical smart-farming applications are discussed here to
show the versatility of big data in smart farming analytics.
Although a large number of data mining-related studies have
been published, only a few literature reviews have been done on
smart farming. Wolfert et al. (2017) provided an overview of big data
in smart farming, with a focus on socio-economic-related issues. Jin
et al. (2018) surveyed the data assimilation of remote sensing and
crop models. Bharate and Shirdhonkar (2017) reviewed works on
plant disease detection using image processing techniques such as
Artificial Neural Networks (ANNs) and Support Vector Machines
(SVMs). Hamuda et al. (2016) conducted a survey on image-based

2
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

applications in agriculture. Glaroudis et al. (2020) also focused on


the research challenges in the utilization of IoT protocols in smart
farming and Farooq et al. (2019) studied the “Role of IoT in
agriculture for the implementation of smart farming”. Bacco et al.
(2019) survey, “Research activities on smart farming” specialised in
digitization agriculture. Friha et al. (2021) have presented smart
agriculture- based technologies for the IoT. They have talked about
wireless technologies, aerial vehicles, cloud computing and
platforms. More recently Idoje et al. (2021) have recorded
observations regarding issues and challenges of smart farming
technologies.
Existing surveys have mainly focused on smart farming techni-
ques while falling short of a detailed review of the big data techni-
ques involved. In several current smart farming
scenarios,researchers and industry practitioners exhibit a focus on
understanding the big data-related techniques. Such information
could provide significant knowledge for practitioners who intend to
implement and utilize big data in their smart applications. The main
goal of this survey is therefore to fill in the existing gap as well as
provide a review of the state-of-the-art big-data-focused smart
farming data analysis techniques and applications. The contributions
of the survey therefore include the following:

• A comprehensive overview of smart farming big-data with a


comprehensive categorization over the existing data in the
domain.
• A discussion on smart farming applications with a specific focus
on the food producing areas of farming systems, such as yield
prediction, animal detection and farm management.
• A review of the most widely used machine learning techniques in the
above applications, such as Linear Regression, Support Vector
Machines and Artificial Neural Networks.
• A discussion on smart farming data challenges and potential paths
for future research in the domain of smart farming.

Our survey on the above topics will shed more light on several
aspects of the contemporary smart farming practice as related to the
utilization of big data and relevant analytical technologies. As
such, we summarize the key elements of smart farming data in
Fig. 1.

2.Methodology

This survey is based on our literature search and analysis for


published work in the domain between 2015 and May 2021, also
including a few major works before 2015. There were two steps
involved in the literature search. In the first step, full articles (journal
papers, conference papers and websites) were searched through the
key online bibliographical sites such as Web of Science, Scopus and
Google Scholar. Although there are several agriculture-related areas,
such as timber plantations, forests, bio-fuel plantations and florists,
our focus was on food-producing farming systems. Food-producing
farming systems are field crops such as wheat, barley, horticultural
crops such as fruits and vegetables, vineyard, animal husbandry and
fisheries. The phrases smart farming, digital farming, data-driven
farming, Agriculture 4.0, Agriculture 5.0, big data and the IoT were
among the keywords used to search for published works. We iden-
tified 54 journal articles, 6 conference papers and 2 websites. In
the

3
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Fig. 1. Key elements of smart farming data. Examples are given for each key element.

second step, the following research questions were considered as the in farming. Since sensors were introduced to farms, micro-climatic
main criteria for the selection of research papers:

• What are the various types of datagenerated by smart farms?


• What are the favoured big data applications in smart farming?
• What are the lead techniques used for smart farming big data
analysis?
• What are the remaining challenges and possible pathways in the
domain?

In the following sections, we will discuss our findings with re-


spect to the above research questions.

3.Types of smart farming data

Researchers and industry practitioners collect smart farming


data as such data provide valuable information such as an
optimum condition for plant growth, yield prediction, pests and
disease de- tection. With the availability of modern data collection
devices, smart farming data is becoming unwieldy. The need for
proper ca- tegorization of data arises at this point to organize these
inordinate amounts of data. There are different ways in which data
can be ca- tegorized. For instance, an action-oriented data
categorization was introduced by Hartmann et al. (2016) for big
data in the business domain. Cimino et al. (2021) have broadly
classified Digital Twins, as part of which, they also provide a crude
classification of data into those from the physical world of
machines and plants versus syn- thetic data. According to the
nature of the existing data sources and also based on the existing
literature, we categorise smart farming data into six categories,
including climate, growth media, bio-se- curity, crop or animal
measures, external and other smart-farm-re- lated data as
illustrated in Fig. 2.

3.1. Climatic data

Climatic data can be categorised into macro-climatic and micro-


climatic. Most macro-climatic data is not generated directly through
the farm but obtained from organizations such as the Bureau of
Meteorology. Hence, we refer to this type of data as external data in
Fig. 2. Early farming studies used these macro-climatic data, espe-
cially for yield prediction (Ferreira Bocca et al., 2016; Matsumura
et al., 2015). Temperature, rainfall, relative humidity, solar radiation,
cloud cover, wind speed and direction are main measures considered

4
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624
data was gathered through sensors not only for the controlled en-
vironment, like greenhouses and plant houses but also in large fields.
Apart from the temperature, the humidity of the micro-environ-
ment, carbon dioxide concentration and transpiration of the crop are
also common measures used. Most macro-climatic data are collected
as satellite images and are typically unstructured whereas sensor-
generated micro-climatic data, such as daily temperature and
rain- fall, are usually well structured.

3.2. Growth media related data

Naturally, plants are grown on the soil and early farmers followed
the natural process. They have chosen different crops to grow in
various soil conditions. For example, muddy soil is for rice and
regular moist soil for fruits and vegetables. Soil profile, salinity,
soil moisture, organic and nutrient content, pH level and water-
log conditions were primary recorded data. However, with
technological development, the soil has been replaced by
alternative growth media and methods. Coir dust and coal are
some materials used for some plant growth later on.
Hydroponics (Fig. 3) is a relatively re- cent invention to replace
soil. There are some advantages to those growth media: one is
that all crop requirements are supplied through the media.
Farmers could add nutrients and control water with a
hydroponics system (Cambra et al., 2018). Fig. 3 shows the
architecture of hydroponic farming. All growth media and green-
house conditions data can be collected through sensors and the
system will provide feedback to auto-calibrate for optimum
condi- tions.

3.3. Bio-security measures

To maintain the high quality for products, bio-security measures


are essential. Pests, diseases and weeds are crop-damaging factors
other than natural disasters. Most of the pests are insects in the
field of agriculture. Pest identification, in other words, insect type,
life stage of the insect and the numbers at each stage are the main
data that could be collected regarding pest control. Traditionally
these data were taken manually. However, nowadays, automated
systems were developed to gather the aforementioned pest
associated data (Goldshtein et al., 2017).
Plant diseases can be identified by detecting specific patterns in
plants. Various techniques like spectroscopic and imaging technology
are available to obtain patterns (Bharate and Shirdhonkar, 2017).
Weed classification or weed type and amount of weed spread

5
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Fig. 2. An overview of the main categories of smart farming related data. Examples of each data category are given below the major categories.

in the field are the appropriate measures to take. As with pests and beha- viours indicate the physiological and physical states of the
disease, automated weed classification systems were developed (Sa et animal.
al., 2018). These types of bio-security measures can be collected from
images, videos and sensors. Climatic data such as temperature,
precipitation and wind speed together with leaf wetness have been
used to forecast diseases (Sarri et al., 2017). For example, if the leaf
wetness is too high, there is a high possibility the disease will spread.
Estimating the amount of weed on the field, pest or disease damage to
the crop is a non-trivial task because of noisy data.

3.4. Crop and animal measures

Crop and animal measurements could be obtained as phenolo-


gical data, morphological data and physiological data. To obtain a
high yield, choosing the correct variety of crop for farming is im-
portant Marko et al. (2016). To decide the correct crop variety,
phenological data such as the past growth patterns and morpholo-
gical data like the harvest of a particular plant variety can be used.
Plant height, leaf area index, number of flowers or fruit are im-
portant structured data that have been used to forecast future
yield Alwis et al. (2019). Phenological data and morphological data
are useful for comparative analysis of growth. These types of data can
be recognized as one of the most significant big data sources,
which includes a variety of data, structured, semi-structured, or
un- structured, such as images, field notes, genomic sequencing
and growth information.
Body measurements are important in livestock animals.
Particularly the weight of the animal is useful when evaluating the
growth performance and reproductive efficiency of the animal. As an
example, weight loss indicates poor health. Also, this kind of data is
used to determine the amount of food needed by the animal to
avoid underfeeding or overfeeding (Hansen et al., 2018b). Animal

6
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624
For instance, the daily activities of ruminant animals are resting,
eating, walking and ruminating. Valuable information was
obtained from animal behaviour for efficient farm management
and in- dividual animal health status (Barwick et al., 2018).
Similar to other smart farming systems, aquarium farming also
uses bio-sensors that monitor the physiology and behaviour of
sentinel animals. This provides information on animal well-being.
Nowadays, smart- farming technology has been used for
beekeeping as well (Khan et al., 2021).

3.5. Other smart farm-related data

In addition to the aforementioned data types in smart farming,


there are other types of data that are utilised for more effective and
efficient smart farming.

Farm management: Within a farm, water (Freebairn et al., 2018;


Culibrina and Elmer, 2015), farm machinery (Fountas et al., 2015)
and fertilizer (Asefpour Vakilian and Massah, 2017) management
are the main functions of farm operations that could involve the
utilization and analysis of large amounts of data. Farm ex-
penditure (e.g., salaries and transportation), marketing data (e.g.,
price and sales), as well as post-harvest wastage are other farm-
related measures. These types of structured data are collected by
humans (Wolfert et al., 2017). In the domain of farm manage-
ment, communication and IoT are among the most popular re-
search areas that have a focus on the improvement of data
collection systems (Moon et al., 2017; Ryu et al., 2015).
Research on irrigation systems has improved water supply
methods (dela Cruz et al., 2017a). Sensor-based monitoring
systems have been used to collect animal farm environmental
data promptly and react accordingly to maintain optimum levels
of environment.

7
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Fig. 3. Network architecture of a hydroponic system. All data related to the growth media and greenhouse conditions can be collected through sensors and the system will provide feedback to
auto-calibrate for optimum conditions.

To achieve these optimum levels, they have developed a low-cost,


Genomic data: Genome sequence, gene models, functional an-
sensor-based computer system to monitor animal welfare.
notation, and polymorphic loci are related data types that can
Remote sensing data: Remote sensing is the technique using
be obtained along with genomic data. Various additional
aircraft-based and satellite-based sensors to gain information
informa- tion is provided for variation data, including
about an object, including aerial and satellite images and Global
population struc- ture, individual genotypes, linkage and
Positioning System (GPS) data. Many researchers have utilized
phenotype data. Bolser et al. (2016) introduced a data source
remote sensing data in smart farming development, for example,
platform. It gains the ability to visualize, mine and analyse plant
estimating canopy state variables and soil properties (Jin et al.,
genomic data.
2018), yield prediction (You et al., 2017; Johnson, 2014; Johnson
Image data: Color and texture are the primary features of
et al., 2016), livestock detection and to monitor animal behaviour
images. In Story et al. (2010) research on deficiency detection,
(Barwick et al., 2018).
red-green-blue (RGB) and hue-saturation-luminance (HSL) va-
Beverages related data: In everyday life, drinks have a significant
lues are considered as colour features and entropy, energy, con-
role in providing essential hydration and nutrients and they are
trast, and homogeneity as textural features. Automatic weed
part of many social gatherings. The production of such
detection using textural features (Hamuda et al., 2016;
beverages is directly related to agriculture. Milk, tea, coffee, hot
Bakhshipour et al., 2017) and strawberry harvesting using colour
chocolate and other various fruits for fruit juice are among these
features are applications related to these data types.
beverages. Alcohol-related drinks such as wine, beer and liquor
are other beverages with large industries in the world. Grapes,
cereal grains, and rice are some of the commonly used crops for
3.6. External data
the production of alcohol. Similar to other agricultural fields,
IoT technologies are applied to manage crops and produce
We refer to the data that are collected from outside a farm as
drinks (Aquino et al., 2017). In the wine industry, top varieties
external data. These types of data can be obtained from different
as well as the best regions and yield (tonnes per hectare) are
sources, such as government departments and research institutes.
the most va- luable data (Anderson et al., 2010). Some of the
Weather data, for example, temperature, humidity and also
wine-related research works are as follows; Mucherino and
satellite images, are some examples of external data. The Bureau of
Ruß (2011) have studied wine fermentation and predicting
Meteorology1 is the institute usually responsible for storing such
yield production. Aquino et al. (2017) estimated the grapevine
data over long periods of time and we discussed some of those in
berry number per cluster using image analysis and Aquino et al.
Sections 3.1 and 3.5 under remote sensing data. There are several
(2018) also made use of image analysis techniques for early
yield prediction in vineyards.

1
For instance, see the Australian climate data at http://www.bom.gov.au/ climate/data

8
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Table 1

Examples of the available open data in the smart farming domain.

Dataset Source

The Food and Agriculture Organization http://www.fao.org/statistics/en/

NASA Earth https://earthdata.nasa.gov/earth-observation-data/

Australian Bureau of Agricultural and Resource Economics and Sciences http://www.agriculture.gov.au/abares/dataThe


Crop/Weed Field Image https://github.com/cwfid/dataset

Plant Village https://www.tensorflow.org/datasets/catalog/plant_village

Maize Image https://osf.io/p67rz/

Rice Leaf Diseases https://archive.ics.uci.edu/ML/datasets/rice+leaf+diseases

other types of external data, such as post-harvest and market-re- farming


lated data, that are not discussed in this survey.
Data mining techniques can help each role in the farming in-
3.7. Smart farming data resources dustry. Farmers would like to know the future yield of their crop.
Researchers may wish to find out information such as growth
In the context of smart farming, several open data exist, some of
which are listed in Table 1. These open data sets are owned by
government institutes that collect and generate the data. The main
purpose of publishing agriculture data is to inspire research in areas
such as allowing farmers to maximize farming practice and helping
the agriculture value chain. For example, Johnson (2014) have used
freely available satellite data sets for forecasting corn and soybean
yields.

Government Institutes Databases These datasets are collected


by government institutes. They contain data related to farm ac-
tivities including biosecurity, climate, fisheries, forests, GPS,
productivity and soils.
The Crop/Weed Field Image Dataset (CWFID) The dataset is
collected by the autonomous field robot Bonirob in an organic
carrot farm. As the benchmark dataset for precision agriculture, it
comprises 60 images of carrots in the early true leaf growth
stage with annotations. The dataset contains information such
as field images, vegetation segmentation masks, and crop/weed
plant type annotations.
PlantVillage Dataset The dataset is collected at experimental
research stations associated with Land Grant Universities in the
USA. It contains 54,303 images of healthy and unhealthy leaves
from 14 species which can be further divided into 38 categories.
Although the original source of the dataset is not available, the
data is still accessible from TensorFlow.
Maize Image Dataset The dataset is the largest publicly available
image set, annotated for single plant disease affecting maize. It
contains 18,222 images of maize leaves which are taken in
three ways: by a hand-held camera, a camera mounted on a
boom, and a drone. All leaf images are annotated with 10,5705
northern leaf blight (NLB) lesions by experts.
Rice Leaf Diseases Dataset The dataset is collected by a digital
camera from a rice field. It contains 120 rice leaf images from
three classes including bacterial leaf blight, brown spot and leaf
smut.

In addition to the above-mentioned government resources,


non- commercial and non-governmental smart farming data sets
are available, such as semi-structured interviews and field notes
from observations in (Carolan, 2020) which contributed to the
develop- ment of knowledge in digital agriculture platforms,
precision farming techniques, farm equipment and robotics.

4. Big data applications and machine learning models in smart

9
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624
patterns of the plant/animal, optimum conditions for growth, best
control surroundings for pests and disease. Dairy farmers may wish
to know the best conditions for cows for more yield. There are some
efforts in the applications of big data techniques in smart farming.
For some of these applications, data from internal sources (on-farm)
and external sources are integrated. For example, some yield pre-
dictions used internal data on yield as well as external data, such as
macro climatic and satellite images (Lobell, 2013; Matsumura et al.,
2015). When predicting future yield, one can make use of previous
yield data. However, this is not sufficient given other influencing
factors that can result in varying harvest. In the case of a drought
situation, for instance, yield can be less. On the other hand, with
sufficient rain, higher yield can be expected. With the
incorporation of data, such as climatic data, soil data and crop data,
into the models, more accurate predictions can be made ( Ferreira
Bocca et al., 2016).

4.1. Big Data Applications

4.1.1. Yield Prediction and Growth Analysis


According to past and present situations, future yield can be
predicted. Not only farmers are getting the benefit of yield predic-
tion, but also policymakers, food marketing agencies and govern-
ments for food security are benefited (Johnson et al., 2016). Yield
prediction approaches differ from each other with the crop, input
data categories and the prediction model.

Yield Prediction Researches were tending to predict yield with


available data. Ferreira Bocca et al. (2016) focused on feature
selection and model tuning using Random Forest in sugar cane
dataset including macro climatic data. Matsumura et al. (2015)
have used both ANN and Multiple Linear Regression (MLR)
techniques to forecast maize yield against climate changes with
climatic, crop, soil and fertilizer data as inputs. You et al. (2017)
have used remote sensing data for the Deep Gaussian Process
model and Convolutional Neural Network (CNN) to predict
yield in soybean. Soil data, satellite images were used for
Supervised Kohonen Networks (SKNs) for wheat yield prediction
(Pantazi et al., 2016). Lobell (2013) has studied crop yield gap
analysis using remote sensing data (satellite data) with linear
regression. Micro-climatic data, hydroponic data and crop data
gathered from smart farms are used for tomato yield prediction
and factor interpretation by Alwis et al. (2019) for the deep
learning model. Akbar et al. (2018) have proposed ANN models
for optimization and prediction of turmeric oil yield using
macroclimate, soil, crop and oil yield data. Berry number per
cluster in grape was esti- mated by Aquino et al. (2017) using
vineyards images and the ANN and SVM techniques.
Growth Analysis Growth can be measured using different
para- meters. Plant height is a fundamental measurement.
Other than plant height, plant width, number of leaves per
plant are some regularly used dimensions. Yield is the final
result of farming. Without proper growth of the crop, a rich
harvest cannot be expected. Kang and Wang (2017) developed
a Knowledge and Data-Driven Model (KDDM) to serve as a
descriptive model for

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Table 2
Applications of yield prediction and growth analysis with relevant data categories.

Application Data category ML Algorithm Ref.


Direct Yield Prediction Macro Climate, Soil, Crop Data Random Forest Ferreira Bocca et al., 2016
Direct Yield Prediction Macro Climate, Soil(fertilizer), Crop Data ANN and MLR Matsumura et al., 2015
Direct Yield Prediction Remote Sensing Data CNN with Deep Gaussian Process You et al., 2017
Crop yield gap analysis Remote Sensing Data (Satellite data) Linear Regression Lobell, 2013
Oil yield prediction Macro Climate, Soil, Crop Data, Other (Oil yield) ANN Akbar et al., 2018
Yield Prediction with Factor Interpretation Micro Climatic, Hydroponics, Crop Data DEEP CNN Alwis et al., 2019
Estimate berry number per cluster Crop Data (Vineyards Images) ANN and SVM Aquino et al., 2017
Yield prediction with growth analysis Soil, Remote Sensing Data (Satellite images) SKN Pantazi et al., 2016
Descriptive model for plants Climatic, growth media and plant growth measures KDDM Kang and Wang, 2017

plants. They made use of climatic data, growth media data such
vehicle (MAV). The authors evaluated features for images using
as fertilizer, plant growth measures, specifically biomass pro-
dense semantic weed classification.
duction and fruit setting data. Major applications of yield pre-
Pest management Pest management is another critical
diction and growth analysis are listed in Table 2.
element in farming. Pest, weed and disease are bonded with
each other to damage the crop. Weeds grant shelter to pests
4.1.2. Quality maintenance
while establishing the crop plants. Pests damage the plant
A better quality product can be obtained only by minimizing harm
directly and damage in- directly by spreading diseases.
from pests and diseases. Therefore, early detection of pests, diseases,
Goldshtein et al. (2017) developed an auto-monitoring trap for
and weeds could improve the quality of the product. Pests and
Mediterranean fruit fly. Linear rank regression analysis was
diseases directly damage the crop (Sun et al., 2018). Weeds absorb
used to evaluate the performance.
much of the water and fertilizer given to the crop. At the same time,
weeds provide shelter to pests and act as a host for diseases.
4.1.3. Animal husbandry and aquaculture
Therefore, numerous studies were carried out to get rid of these
Smart farming helped not only monitor animals but also provides
harmful bugs, weeds and diseases. Detecting ripened fruits and
facilities such as feeding animals and milking. IoT applies to the area
harvest at the right time are some of the key factors to main- taining
of animal husbandry as well as the area of aquaculture in animal
the value of fruit and minimize wastage. The applications of quality
farming. The recent heightened awareness and improvement in
maintenance are listed in Table 3.
animal welfare, has motivated researchers to bring new technologies
to animal farming (Caria et al., 2017). Animal monitoring via sensors
Quality Assessment Grading and sorting are essential to main-
helps to improve the quality of the animal’s life. Recognition of an-
tain the quality of the product. Further research is striving to
imals using face and body conditions are some other applications
keep the uniformity of the products. Quality of malting barley
in this area. The applications of quality maintenance are listed in
was assessed by Ramirez-Paredes and Hernandez-Belmonte
Table 4.
(2020) using color, shape and texture descriptors. They have used
the image data for the Support Vector Machines-Radial Basis
Animal Husbandry In animal husbandry, individual animal re-
Function(SVM-RBF) classifier for their solutions.
cognition is essential for both animal monitoring and treatments.
Disease Detection A large number of experiments have been
Hansen et al. (2018a) worked on pig face recognition using facial
carried out for disease detection, mainly with images using dif-
images of pigs with CNN. Lee et al. (2019) supported large-scale
ferent techniques. Bharate and Shirdhonkar (2017) have reviewed
smart farms by developing a functional monitoring system. They
plant disease detection using image analysis. They have ad-
have used video data of pigs for Mask-R-CNN, Deep learning, and
dressed a series of diseases in a range of crops such as apple rot,
Gaussian Mixture Model. Linear Discrimination Analysis(LDA)
grape powdery mildew, Aspergillus fruit rot, using ANN and SVM
and Quadratic Discriminant Analysis(QDA) were used by Barwick
techniques. SVM classified images as infected or non-infected.
et al. (2018) to predict sheep behaviour.
Identification and classification of damaged corn kernels using
Aquaculture Aquaculture is one of the fastest-growing primary
SVM are studied by Sun et al. (2018). Images of corn kernels were
industries in the world. To improve all areas of this industry,
used as inputs.
together with animals provided by the mini bio-sensors, en-
Weed detection Weed detection is another main area of smart
vironmental sensors can also used. Farm management determi-
farming practices. Bakhshipour et al. (2017) proposed an ANN
nations, animal welfare, social consciousness and consequently,
technique to overcome the difficulties brought by leaf occlusion
sustainable productivity can be increased using real-time
and overlapping. Sa et al. (2018) have used the dense (pixel-wise)
animal and environment monitoring. Two telemetry-based
CNN technique to address the issue with weed. They have col-
systems to reflect feeding behaviours in farmed Atlantic salmon
lected multi-spectral images of crops and weeds by a micro aerial
were de- veloped by Føre et al. (2011). They have used video
footage and field test data for time series analysis.

Table 3

Applications of quality maintenance and relevant data categories.

Application Data category ML Algorithm Ref.


Quality assessment Color, shape and texture (images) SVM-RBF Ramirez-Paredes and Hernandez-Belmonte, 2020
Quality maintenance Crop data (Corn kernels images) SVM Sun et al., 2018
Disease detection Disease (images) ANN and SVM Bharate and Shirdhonkar, 2017
Disease observation in real field conditions Color and textural (images) SVM Zhou et al., 2015
Weed detection Weed (images) ANN Bakhshipour et al., 2017
Weed detection Weed (images) CNN Sa et al., 2018

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Table 2
Auto-monitoring fruit fly trap Remote sensing data, insect count Linear rank regression Goldshtein et al., 2017

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Table 4
Applications of animal husbandry and aquaculture and relevant data categories.

Application Data category ML Algorithm Ref.


Animal recognition Animal data (images of face) CNN Hansen et al., 2018a
Animal monitoring Video data Mask-R-CNN, DL, Gaussian Mixture Model Lee et al., 2019
Automatic system for monitoring of cows Animal 3D images Rolling ball algorithm Hansen et al., 2018b
Sheep monitoring Ear acceleration signals from sheep LDA and QDA Barwick et al., 2018
Feeding behaviours in Atlantic salmon Videos footage and field test data Time series analysis Føre et al., 2011

Table 5

Applications of farm management and relevant data categories.

Application Data category ML Algorithm Ref.


Service platform to control greenhouses Farm management related data SDI model Barmpounakis et al., 2015
Farm machinery management information system (FMMIS) Machinery data PSM and SSM Fountas et al., 2015
Fertilizer practice Soil (fertilizer), crop data (yield) Linear Regression Muhammed et al., 2017
Workforce reduction Images CNN Quiroz and Germá n, 2020
Water usage optimization Farm management ANN dela Cruz et al., 2017a
Water monitoring System Water level varying data FDSS dela Cruz et al., 2017b
Predicting on water and electricity consumption Farm management ANN and SVM Shine et al., 2018
Water stress index in wine grape Macro, micro climatic (canopy temperature) ANN King and Krista, 2018
Optimally watering Farm data Linear regression Muangprathub et al., 2019
Tea leaves plucking robot Stiffness of the branches Gaussian distribution Motokura et al., 2020
Fruit harvesting robots Color of fruit CNN Xiong et al., 2020
Harvesting robots Color and shape CNN Mao et al., 2020
Harvesting robots Color and shape (RGB-D images) SVM Li et al., 2020
Strawberry harvesting robots Color and shape (RGB images) DEEP CNN Ge et al., 2019

4.1.4. Farm management fruits as possible within the right ripening period (Li et al., 2019).
Managing the farm day today is not an easy task. Making the
correct decision for maximum profit and optimizing operational
work such as monitoring crops and animals, water and food supply,
fertilization and labour management, are some of the daily demands
on a farm. The applications of farm management are listed in Table 5.

General Farm Management Systems The smart farming system


helps farmers significantly by providing a range of applications.
Monitoring field in real-time, managing crop environment ac-
cording to weather change, feeding animals based on their body
conditions, are some of them. All these ventures involve big
data applications. Fountas et al. (2015) designed a new “Farm
Ma- chinery Management Information System (FMMIS)” by
applying Problem Structuring Methods (PSM) and Soft Systems
Metho- dology (SSM). This system helps farmers utilise farm
machinery management. The data-driven from farm machinery
were used as input data. Muhammed et al. (2017) have
investigated fertilizer practice with soil and fertilizer data, crop
data and yield data applying linear regression. Quiroz and
Germá n (2020) have proposed an image recognition approach
based on a CNN with blueberry plants.
Water Management Systems Much and varied research has
been conducted to reduce wastage and optimize usage. dela
Cruz et al. (2017a) have proposed the Smart Farm Automated
Irrigation System (SFAIS) using Neural Network to optimize the
water usage in the smart farms. dela Cruz et al. (2017b)
designed a Fuzzy-based Decision Support System (FDSS) for
water mon- itoring, using water level difference data by
applying Fuzzy logic. On-farm direct water and electricity
consumption was predicted using ANN and SVM algorithms by
Shine et al. (2018). The water stress index in wine grapes was
King and Shellie’s King and Krista (2018) approach using macro
climatic data and canopy tem- perature model with a neural
network. Muangprathub et al. (2019) have proposed optimally
watering using farm data. They have used linear regression to
analyze those data.
Harvesting Robots Harvesting is the most critical stage of
farming. Farmers need to pay attention to ripening states, any
damage to the fruit while harvesting and collection of as many

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Table 4
Therefore, manual fruit harvesting is a highly labour-intensive
process. Harvesting robots were introduced to the farming in-
dustry as the latest invention to overcome this difficulty. These
robots are programed to pick fruits through the use of early
images/data collected from the farm and utilizing advanced
machine learning techniques. Various types of robots are in the
field for harvesting purposes that make use of different
farming data. Tea plucking robot was Motokura et al. (2020)
invention, which makes use of the stiffness of the branches as
input data with a Gaussian distribution mechanism. Mao et al.
(2020) used the CNN algorithm and fruit colour and shape as
input data to form a strawberry picking robot and cucumber
detecting robots. RGB images are used in Li et al. (2020) with
SVM and Ge et al. (2019) with deep CNN for their harvesting
robots.

4.2. Machine learning models

Most of today’s farming data are analyzed using machine


learning (ML) algorithms; thus, in this section, some of the most
frequently used ML models in the domain will be discussed. These
ML techniques have been widely used to analyze the aforemen-
tioned smart farming data.
The ML algorithms that have been used in smart farming data
analysis can be summarized into three broad categories:
traditional regression, tree or kernel-based models and the ANN-
based models such as deep learning models. We group the main
functions of smart farming applications into three major
categories, i.e., prediction, detection and optimization. Table 6
summarizes all the techniques and applications, from which some
observations can be made.
While there are numerous machine learning techniques, we
se- lected some of the most commonly used techniques for a more
detailed review, i.e., linear regression, SVMs and ANNs. Linear re-
gression models have been mostly used for farm management
practice, such as controlling fertilizers (Muhammed et al., 2017) and
optimizing water usage (Muangprathub et al., 2019). The majority
of quality maintenance applications used SVMs, e.g.,
(Bakhshipour et al., 2017; Sa et al., 2018). The most widely
utilised technique, the ANNs and their variants, have been used for a
variety of applications, including yield prediction (Akbar et al.,
2018; Matsumura et al., 2015; Alwis et al., 2019), quality
maintenance (Bharate and

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Table 6

Machine learning models and big data applications in smart farming.

Algorithm Application Ref.


Linear Regression Detection/recognition-animal Goldshtein et al., 2017
Linear Regression Other-minimization fertilizer/water Muhammed et al., 2017; Muangprathub et al., 2019
Linear Regression Prediction-yield Lobell, 2013
SVM Detection-damages; disease Sun et al., 2018; Ramirez-Paredes and Hernandez-Belmonte, 2020
SVM Detection-fruit Li et al., 2020
ANN Other-water stress index King and Krista, 2018
ANN Other-optimization water usage dela Cruz et al., 2017a
ANN Prediction-yield Akbar et al., 2018
CNN Detection/recognition-animal Hansen et al., 2018a
CNN Detection-fruit Mao et al., 2020; Ge et al., 2019
CNN Other-minimization workforce Quiroz and Germá n, 2020
ANN and MLR Prediction-yield Matsumura et al., 2015
ANN and SVM Prediction-consumption Shine et al., 2018
ANN and SVM Detection-disease Bharate and Shirdhonkar, 2017
ANN;CNN Detection-weed Bakhshipour et al., 2017; Sa et al., 2018
CNN with Deep Gaussian Proces Prediction-yield You et al., 2017
DEEP CNN Prediction-yield Alwis et al., 2019
Gaussian Detection-leaves Motokura et al., 2020
KDDM Prediction-growth Kang and Wang, 2017
LDA and QDA Detection/recognition-animal Barwick et al., 2018
Mask-R-CNN, DL, Gaussian Detection/recognition-animal Lee et al., 2019
ANN and SVM Prediction-yield Aquino et al., 2017
PSM and SSM Other-farm manage-FMMIS Fountas et al., 2015
Random Forest Prediction-yield Ferreira Bocca et al., 2016
SKN Prediction-yield Pantazi et al., 2016
Time series Detection/recognition-animal Føre et al., 2011

Shirdhonkar, 2017; Bakhshipour et al., 2017) and with harvesting


4.2.1. Regression
robots (Xiong et al., 2020; Mao et al., 2020).
In terms of general regression forecasting models in smart
Most of the predictive and detective applications used the ANN
farming, the most widely used approach is the time series-based
models. The ANNs and most recent developments in the specific
regression forecasting on yield. The time series-based regression
domain of deep learning have resulted in much more accurate pre-
forecasting on yield makes use of historical yield value at each time
dictions on smart farming data including time series farming data.
point, namely, lag value, to forecast yield value in future time
While the identification and prediction of disease, insects, and har-
points (lags).
vesting time-frames were once challenging tasks, they have been ˆ ˆ
ˆY = + X + ˆ (1)

facilitated by current developments in the ANN and deep learning


domain. This has helped to overcome data complexity challenges. farming domain.
However, ANNs are prone to overfitting when the size of the data is
limited (Tetko et al., 1995). Linear regression, on the other hand, has
provided solutions for detection and optimization applications too.
Although regression is a fundamental technique in predictive mod-
eling, fewer smart farming applications have made use of regression
techniques for prediction purposes. However, most (linear) regres-
sion models fail to capture complex and non-linear relationships
among several smart farming related data attributes and as such,
they lack high effectiveness measures in practice. SVMs have been
used mostly for detection purposes and have been shown to be a
significant and accurate ML technique for anomaly detection and for
cases where there are large numbers of predictor features that are
non-linearly related. The latter is mostly due to the fact that SVM
parameters are not dependent on the number of input predictor
features. These characteristics make SVMs a well-suited technique
for the detection of weeds or damages (similar to anomalies) in the
farming context without being concerned with the input feature size
and complexity. SVMs are capable solutions even in the presence of
non-linearity in data. However, most recent research work in smart
farming utilizes deep learning techniques because of their superior
performances. In some of the latest research works (e.g., Ge et al.,
2019), emphasis has been put on image data. Thus, CNNs have been
utilised for such analyses since they are more effective tools for
image analysis and recognition.
A more detailed description of selected ML techniques, i.e., re-
gression, SVMs and ANNs, are given below. As seen in Table 6,
these techniques form the main body of ML analyses in the smart

1
S.D.
i Alwis,
0 Z. 1
Hou,
i Y. Zhang
i et al. Computers in Industry 138 (2022) 103624

At the heart of a regression model is the relationship between the


dependent and independent variables. For instance, in Eq. (1),
sup- pose we want to forecast the yield Yˆi for farm and we have
concluded that the farm’s environmental variables are Xi and the
yield values
are Yi. The yield values Yi we are forecasting would be the
dependent variable because their value “depend” on the values of
environ- mental variables Xi and the environmental variables
would be the
independent variables. The coefficient ˆ1 reflects the linear re-
lationship between Xi and Yi. One of the examples for yield fore-
casting is (Lobell, 2013). Goldshtein et al. (2017) used regression
to estimate trapping efficiency of the medfly. Muhammed et al.
(2017) and Muangprathub et al. (2019) utilized regression
techniques to find a way to minimize water and fertilizer usage.
The main advantage of using the regression forecasting is the
simplicity of the technique, especially if the dependent and in-
dependent variables are linearly related to each other, such as en-
vironment and yield as shown in Lobell (2013). However, the
assumption on linearity is not always accurate. Also, with the
large amounts of (big) data on smart farming, the prediction
power of regression analysis is challenged by the more
sophisticated ML blue techniques such as SVMs and deep
learning.

4.2.2. Support vector machines


SVM analysis is a popular machine learning tool for classification
and regression (Sun et al., 2018). Some regression problems cannot
adequately be described using a linear model as the relationship
between dependent and independent variables show significant non-
linearity. In such cases, the Lagrange dual formulation allows the

previously-described technique to be extended to non-linear


functions. Particularly in fruit grading, (non-linear) kernel
functions

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

can help to better solve the non-linear regression problems with an


better interpretation capability on how the features from farm could
accurate anomaly detection performance. Identification and classi-
help the tomato yield. Comparing to (Alwis et al., 2019), the work in
fication of damaged corn kernels using SVMs is another application
(You et al., 2017) could be more accurate on the forecasting of yield
studied by Sun et al. (2018). Ramirez-Paredes and Hernandez-
with the usage of the remote sensing data on the CNNs with the
Belmonte (2020) utilized SVMs for detection of damaged fruit. Fruit
Gaussian process. CNNs were also used by Sa et al. (2018) for weed
detection and harvesting robot in Li et al. (2020) is another appli-
detection and by Hansen et al. (2018a) for animal recognition. Quiroz
cation of SVMs.
and Germá n (2020) used CNNs for minimizing workforce. However,
The SVMs technique predicts that the positive class is present
the usage of CNNs on detection tasks are mainly focused on mining
when Eq. (2) is positive. Likewise, it predicts that the negative class
the features of the data, the accuracy and performance is still open
is present when Eq. (3) is negative (Goodfellow et al., 2016).
for further research. Beside CNNs, Matsumura et al. (2015) used a
3wTx + b = 1 non-linear ANN model with one hidden layer to forecast maize
(2)
T yield against climate change.
4w x + b = 1 (3) ˜

The SVMs technique tries to solve the dual problem by using y =


j
w˜ jtanh ( i
wji xi + bj ) +b (6)

Lagrangian dual (Goodfellow et al., 2016).


In the above Eq. (6), the function tanh will transform the input to

maximize f (c …c ) = 1 the non-linearity space and could confirm the complexity that is
n c
n n
1 n yi ci (xi xj)yj cj needed from the ANN model and conduct the better forecasting
i=1
n
i
2 i=1 j=1 results. Akbar et al. (2018) also used ANNs for yield prediction while
subjectto cy =0 Bakhshipour et al. (2017) used ANN models for weed detection
i i
i=1
and
1
0 ci dela Cruz et al. (2017a) utilized them for optimization of water
foralli 2n (4) usage. As discussed before, most deep learning techniques are easy
In the specific domain of yield forecasting, the SVMs kernel Alwis et al. (2019); You et al. (2017) have used deep CNNs for yield
functions can help to better solve the non-linear regression pro- prediction and Ge et al. (2019); Mao et al. (2020) utilized CNNs for fruit
blems with an accurate forecasting performance. However, the detection. The work in Alwis et al. (2019) has the spatial and also
SVM regression technique sometimes tends to fit the training data temporal attentions on top of the CNNs, which could have a
too well and explicitly, which will then cause the so-called over-
fitting problem. In such cases, the forecasting performance is
worse on testing data than that on training data. Another challenge
with SVMs is that the SVM regression is much more difficult to be
interpreted by farming domain experts especially when compared
with the more interpretable techniques such as linear or logistic
regression.

4.2.3. Deep learning techniques


Another technology that has attracted much attention in yield
forecasting is the ANN techniques. Compared to the regression and
SVMs regression, deep learning techniques can handle significantly
large amounts of data due to their complex computational and non-
linearity power and characteristics. Several recent works have used
deep learning techniques for yield forecasting to obtain state-of-the-
art forecasting performances. Deep learning techniques are built
using neural network building blocks (neurons and activation
functions) and use training data in several epochs to learn the re-
lationship between the dependent and independent variables. The
non-linearity challenge and the big data issue are both solved using
deep learning techniques (Goodfellow et al., 2016) although in many
cases, the learning process may take longer than other simpler
techniques such as SVMs. Convolutional Neural Networks (CNNs)
form a subclass of deep neural networks that are mainly applied in
the analysis of image or video data. In the context of smart farming,
CNNs are widely used in yield prediction. Convolution, pooling and
fully connected layers are the three main elements in CNNs. The
convolutional layer shares weights across the two dimensions of the
neurons (or inputs) in a previous layer (2D convolution). Therefore, it
significantly decreases the number of parameters. For instance,
having x as the input image with a linear filter, adding a bias b term,
then applying a non-linear function f(⋅) which is usually Rectified
Linear Unit (ReLU), where p(⋅) is the pooling function whose
filters are determined by the weights W and bias b, and after going
through the pooling layer, output O will be as follows:
O = p (f (W*x + b)) (5)

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624
to fall into overfitting problems and having a trade-off between
the training accuracy and model generalization is still open in this
line of research.

5. Discussion

Proper categorization of smart farming data is essential as it


provides an overall picture of all big data in the domain. redWhile
Wolfert et al. (2017) provided an overview of big data in smart
farming, they only focused on socio-economic-related issues. We
provided a wider understanding of data sources and the major
types including climatic, growth media related, bio-security,
crop/animal measures, external and other relevant data. The
climatic and growth media related data have been extensively
used for growth analysis while bio-security measures can be
analysed and assist with a comprehensive control and
maintenance of high quality products alleviating the negative
effects of bio hazards such as the effect of pests. Other types of
smart farming data, such as crop and animal data have been
utilised for yield prediction purposes. Our findings show that it is
therefore a collection of several different types of data that need to
be systematically collected and analysed in the context of
successful multi-pronged smart farming practices.
In previous studies, Bharate and Shirdhonkar (2017) reviewed
works on plant disease detection, Hamuda et al. (2016) surveyed
image-based plant classification and Kamilaris and Prenafeta-Boldú
(2018b) surveyed deep learning applications in agriculture. While
these studies are limited to one application area relevant to smart
farming, we discussed a broad range of applications such as yield
prediction, growth analysis, quality maintenance, animal and aqua-
culture and farm management. In addition, we described the type of
big data that are collected and analysed as well as the techniques
utilised for the same in terms of those applications. Such data can be
used by farmers to make effective and efficient operations, e.g.,
timely harvesting. Policymakers also may make use of the different
data types to make more informed decisions regarding relevant
operations such as import or export according to yield prediction
results.
There are several tools and techniques that can be utilised for
prediction purposes. Bharate and Shirdhonkar (2017) reviewed
image processing techniques, ANNs and SVMs for plant disease de-
tection and Kamilaris and Prenafeta-Boldú (2018b) viewed deep
learning applications in agriculture. However, such existing studies
did not give cover to the wide range of machine learning techniques
used in smart farming applications.

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

5.1. Challenges
consistency and completeness are the data quality dimensions that
need to be maintained (Taleb et al., 2018). In addition, the reliability
While there are constant developments in the relevant smart
of commercial sensors (Neethirajan, 2020) and cyber security as-
farming technology and applications, challenges also exist. Food and
pects, including social engineering, ransomware and relevant se-
Agriculture Organization of the United Nations predicts that the
curity and privacy issues (Barreto and Amaral, 2018), are among
world’s population will increase to 9 billion by 2050 (Food and
those factors that require further attention.
Agriculture Organization of the United Nations, 2018b). This brings
Among other types of challenges in smart farming are climatic
challenges about how to produce enough food to feed the increasing
conditions, socio-economic status and a skilled workforce. Climatic
population sufficiently, not to mention that there are still 821 million
conditions and change is a highly influential factor resulting in un-
people suffering from hunger (Food and Agriculture Organization of
expected rain/drought, declining pollinator population, e.g., bees.
the United Nations, 2018a). As one of the most efficient ways to
Wolfert et al. (2017) discussed socio-economic challenges,
overcome this challenge, smart farming presents the ability to solve
including data ownership. Geospatial applications need
the problem (Kamilaris and Prenafeta-Boldú , 2018a).
provisioning for real- time processing and suitable infrastructures
Bacco et al. (2019) divided all various types of smart farming
for data storage. In- adequate skills to handle data, processing and
challenges into two different groups, namely, technical challenges
management of high- dimensional and multi-modal data sets are
and non-technical challenges. Data, network and information are the
other issues related to geospatial applications (Kamilaris et al.,
challenges they have classified as technical and economic and skills
2017).
perspectives as non-technical. Security, stack and IoT agricultural
The working atmosphere of smart farming is different from
challenges have been mentioned in Farooq et al. (2019) study.
conventional agriculture. Developed technology in these fields
Moreover, IoT agricultural challenges are classified into hardware,
must be handled by skilled workers including up-to-date farm
network, platforms and other classifications such as interference,
managers (Fountas et al., 2015). Finding highly skilled and
reliability and scalability. Glaroudis et al. (2020) have addressed
experienced labor, with the ability to operate in extremely specific
challenges around systems and applications. Contrary to the existing
work conditions, has also been identified as a relevant challenge in
studies in the domain, we focus our attention on the big data-related
the domain (Johnson, 2014). Apart from the human factor, the
challenges and relevant opportunity scenarios.
challenges on data also need to be acknowledged, including storage,
As the foundation of economic development, agriculture is an
processing, analysis, quality assessment and quality management,
evolving area for many countries. Agriculture creates enormous
as mentioned in Taleb et al. (2018). Lioutas et al. (2019) have also
value. With the developments of agriculture and the IoT, smart
raised fifteen fundamental questions in big data related to farming
farming gains relevant advantages through accessing new technol-
and tried to find solutions through the activity theory approach.
ogies, facilities and data collected during agricultural processes.
Farmers’ inadequate capability to deal with complex data and
With the advent of big data, various types of data-related chal-
double-acting as producers and users of big data are some of the
lenges have also emerged. Knowledge discovery is the focal
points raised. Solutions for those issues require future research
purpose of data through analytics. Over-fitting, data imbalance and
and industrial practice. Some of the other issues are related to
finding suitable algorithms for specific data sets are some of the
cultural, political and ecological limitations or biases of data sets
examples of data challenges in the real world and in the smart
(Bronson and Knezevic, 2016). Trust, equity, dis- tribution of
farming domain. Several challenges also exist in relation to the
benefits and access to data (Fleming et al., 2018), as well as
several dimensions of big data, i.e., volume, velocity, variety and
consumer rejection of technologies (Regan, 2019), are the chal-
veracity.
lenges among those to be further studied.
It is a continuing challenge to service a constantly increasing
world population. It is imperative that smart farm data be used to
5.2. Opportunities
maximum efficiency to improve production as less land becomes
available (Suebsombut et al., 2017). The different data types we
Smart farming opportunities vary in many ways. Unnecessary
summarized in Section 3 could be substantial. This makes it neces-
pesticide, weedicide and fertilizer usage must be addressed. This will
sary to have access to the technology that can be used to capture,
in turn reduce the carbon footprint. That is a significant opportunity
store and deliver such large volumes of data and analyse them ac-
for environmental protection. Within smart farms, water and energy
curately.
usage can be optimized resulting in valuable saving of precious re-
Collecting real-time data and using it for quick decision making
sources (dela Cruz et al., 2017a). Labour efficiency can be achieved by
could also be one of the main challenges in smart farming. In the
smart platforms resulting in a smaller labour force needed for smart
farming industry, dealing with living things, such as plants and an-
farming (JiHye et al., 2017). Considering all these factors, overall
imals introduces an element of urgency. It is essential to react
farming costs might be reduced. Increased number of sensors create
promptly to protect crop and livestock, not only from severe weather
more opportunity to collect more data. An increase in smart farming
conditions and disease but also to supply adequate water for high
data provides invaluable opportunities for researchers to achieve
yield. Estimating near real-time crop yield, especially in food-ex-
progress in many areas of smart farming. Popular applications of
porting countries, ensures global food security, a crucial real-world
data and future improvement will be discussed next.
challenge.
Currently, most of the yield prediction approaches use at least
In Section 3, we discussed a variety of data sources and under-
one if not several attributes such as climate or soil profiles for their
standably, the data has to be integrated to obtain useful information.
predictions. Genomic data and crop data such as leaf area index are
To ensure accurate and timely analysis, this will constitute a huge
hardly used. By utilizing the attributes as much as possible, yield
challenge. Also, the heterogeneous unstructured data needs to be
prediction could be more accurate. For example, Alwis et al. (2019)
analysed and interpreted with the help of both data scientists and
show a more accurate yield prediction for tomatoes using the most
farming experts (Wolfert et al., 2017), which also highlights the
number of attributes. Site-specific prediction is also important be-
challenge of bridging the gap between the two knowledge domains.
cause crop yield depends on all the parameters as well as in-situ
It is essential to maintain the veracity of data during collection,
conditions (Bauer and Aschenbruck, 2018).
storage and transportation to get the maximum benefit (Moon et al.,
At the same time, it is essential to harvest on time to reduce
2017). However, during data collection, especially on the manual
post- harvest wastage and maximize profit. Some fruits may over
plant measures, human error can occur. This may result in inaccurate
ripen and some are under-ripe and not suitable for the market. To
data and consequently in wrong information. Accuracy, timeliness,

1
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624
avoid this wastage, image analysis could help with proper
monitoring and decide the best time for harvest. Further
inventions could lead to a

2
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

machine designed to harvest only fruit ready for the picking. In some to influence the work reported in this paper.
scenarios, the fruit may have a good appearance ready to harvest.
However, the taste might be unsatisfactory. Consumers are clearly
interested in taste as well as appearance. Therefore, Brix value of the
fruit could be measured before the harvest. This is a good oppor-
tunity to create a new harvesting tool by combining both appear-
ances and Brix value.
Smart farming sensors collect all the environmental data every
minute, resulting in a range of data with time. At the same time,
biological data were measured weekly. Therefore, the following
question could be answered with smart farming data. Which
week/s environmental factors affect the yield? What are the most
promi- nent factors that affect crops at different stages of growth?
What are the best strategies to maximize the overall yield?
Recently, Alwis et al. (2019) have worked on the tomato data set and
addressed most of the above questions. Research of a similar
nature, could be done for other crops as well.
In this survey, we addressed the gaps identified in smart
farming particularly, food-producing related farming. Proper
categorization of smart farming data was discussed in Section 3
with a focus on the source of the data. We have identified and
categorized the relevant smart farming applications and also
discussed some major, com- monly used data mining techniques in
this domain. In summary, we put emphasis on giving an overall
picture of smart farming data, applications and analytical
techniques in the domain. We have also discussed some big data
challenges and future opportunities for more effective and efficient
smart farming practices. We believe these challenges and
opportunities are crucial for researchers and industry
practitioners who intend to employ smart farming tech- nologies.
Some relevant aspects of using and analysing big data in smart
farms, such as socio-economic and cyber security issues are to be
further studied in future work.

6.Conclusions

Our intention in this survey was to provide an overview of


smart farming big data analysis with a focus on smart farming data
man- agement strategies, analytic techniques and existing
challenges and opportunities in the domain. We focused on the
importance of uti- lizing machine learning techniques in smart
farming, to both new- comers and experts in agriculture, data
mining and knowledge discovery studies. Smart farming is a
relatively new field and utili- zation of machine learning in the area
has attracted increasing at- tention recently. The most popular
data mining techniques applied in smart farming problems were
therefore a keynote for this paper. We recognize the many possible
developments in the domain ac- cording to the recent literature.
Yet, we believe that data mining in smart farming is a new and fast-
growing area that provides nu- merous solutions to farmers in a
variety of applications in the farming industry. From a smart
farming big data viewpoint, data mining is rapidly becoming an
essential element for decision- making in everyday farm
management practices. Yield prediction, harvest time, crop
information and best variety to plant during next season are
among the valuable insights that can be derived. Although there
are still several big-data related challenges in smart farming
analytics to be more thoroughly addressed, several oppor- tunities
for more efficient and effective farming practices exist and
outweigh the cost and burden of overcoming those challenges.

Declaration of Competing Interest

The authors declare that they have no known competing fi-


nancial interests or personal relationships that could have appeared

2
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624
Acknowledgement Cimino, Chiara, Ferretti, Gianni, Leva, Alberto, 2021. Harmonising and integrating the digital
twins multiverse: a paradigmand a toolset proposal. Comput. Ind. 132, 103501.

The authors would like to thank A/Prof Gang Li for providing the dela Cruz, John R., Baldovino, Renann G., Bandala, Argel A., Dadios, Elmer P., 2017a. Water
valuable feedback. The authors would also like to thank the anon- usage optimization of smart farm automated irrigation system using ar- tificial neural
ymous referees for their valuable comments and helpful suggestions. network, in: Proceedings of the Information and Communication Technology (ICoIC7),
2017 5th International Conference on, IEEE, pp. 1–5.
We gratefully acknowledge support from Australian Government
Research Training Program Scholarship. This work was supported by dela Cruz, John R., Baldovino, Renann G., Culibrina, Francisco B., Bandala, Argel A., Dadios,
Korea Institute of Planning and Evaluation for Technology in Food, Elmer P., 2017b. Fuzzy-based decision support system for smart farm water tank
Agriculture, Forestry(IPET) through Smart Plant Farming Industry monitoring and control, in: Proceedings of the 2017 5th International Conference on
Information and Communication Technology (ICoIC7), IEEE, pp. 1–4.
Technology Development Program, funded by Ministry of
Agriculture, Food and Rural Affairs(MAFRA)(421017-04). Culibrina, Francisco B., Dadios, Elmer P., 2015. Smart farm using wireless sensor network
for data acquisition and power control distribution, in: Humanoid, Nanotechnology,
References Information Technology, Communication and Control,

Akbar, Abdul, Kuanar, Ananya, Patnaik, Jeetendranath, Mishra, Antaryami, Nayak,


Sanghamitra, 2018. Application of artificial neural network modeling for optimi-
zationand prediction of essential oil yield in turmeric (Curcuma longa l.). Comput.
Electron. Agric. 148, 160–178.

Alwis, Sandya De, Zhang, Yishuo, Na, Myung, Li, Gang, 2019. Duo attention with deep
learning on tomato yield prediction and factor interpretation. Pacific Rim
International Conference on Artificial Intelligence. Springer, pp. 704–715.

Anderson, Kym, Nelgen, Signe, Valenzuela, Ernesto, Wittwer, Glyn Martin, 2010. Economic
contributions and characteristics of grapes and wine in Australia’s wine regions.

Aquino, Arturo, P Diago, Maria, Millán, Borja, Tardáguila, Javier, 2017. A new metho-
dology for estimating the grapevine-berry number percluster using image ana- lysis.
Biosyst. Eng. 156, 80–95.

Aquino, Arturo, Millan, Borja, Diago, Maria-Paz, Tardaguila, Javier, 2018. Automated early
yield prediction in vineyards from on-the-go imageacquisition. Comput. Electron.
Agric. 144, 26–36.

Asefpour Vakilian, Keyvan, Massah, Jafar, 2017. A farmer-assistant robot for nitrogen
fertilizing management ofgreenhouse crops. Comput. Electron. Agric. 139, 153–163.

Bacco, Manlio, Barsocchi, Paolo, Ferro, Erina, Gotta, Alberto, Ruggeri, Massimiliano, 2019.
The digitisation of agriculture: a survey of research activities onsmart farming. Array
3, 100009.

Bakhshipour, Adel, Jafari, Abdolabbas, Nassiri, Seyed Mehdi, Zare, Dariush, 2017. Weed
segmentation using texture features extracted from waveletsub-images. Biosyst. Eng.
157, 1–12.

Barmpounakis, Sokratis, Kaloxylos, Alexandros, Groumas, Aggelos, Katsikas, Lampros,


Sarris, Vasileios, Dimtsa, Konstantina, Fournier, Fabiana, Antoniou, Eleni, Alonistioti,
Nancy, Wolfert, Sjaak, 2015. Management and control applications in agriculture
domain via afuture internet business-to-business platform. Inf. Process. Agric. 2 (1),
51–63.

Barreto, Luís, Amaral, António, 2018. Smart farming: cyber security challenges, in:
Proceedings of the 2018 International Conference on Intelligent Systems (IS), IEEE,
pp. 870–876.

Barwick, Jamie, W Lamb, David, Dobos, Robin, Welch, Mitchell, Trotter, Mark, 2018.
Categorising sheep activity using a tri-axial accelerometer. Comput. Electron. Agric.
145, 289–297.

Bauer, Jan, Aschenbruck, Nils, 2018. esign and implementation of an agricultural


monitoring system forsmart farming. IoT Vertical and Topical Summit on Agriculture-
Tuscany (IOT Tuscany). 2018. IEEE, pp. 1–6.

Bharate, Anil A., Shirdhonkar, M.S., 2017. A review on plant disease detection using image
processing, in: Proceedings of the 2017 International Conference on Intelligent
Sustainable Systems (ICISS), IEEE, pp. 103–109.

Bronson, Kelly, Knezevic, Irena, 2016. Big Data in food and agriculture. Big Data Soc. 3 (1),
1–5. https://doi.org/10.1177/2053951716648174. (ISSN 20539517).

Cambra, Carlos, Sendra, Sandra, Lloret, Jaime, Lacuesta, Raquel, 2018. Smart system for
bicarbonate control in irrigation for hydroponicprecision farming. Sensors 18 (5),
1333.

Caria, Marcel, Schudrowitz, Jasmin, Jukan, Admela, Kemper, Nicole, 2017. Smart farm
computing systems for animal welfare monitoring, in: Information and
Communication Technology, Electronics and Microelectronics (MIPRO), 2017 40th
International Convention on, IEEE, pp. 152–157.

Carolan, Michael, 2020. Acting like an algorithm: digital farming platforms and the-
trajectories they (need not) lock-in. Agric. Hum. Values 37 (4), 1041–1053.

2
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Environment and Management (HNICEM), 2015 International Conference on, IEEE, Jin, Xiuliang, Kumar, Lalit, Li, Zhenhai, Feng, Haikuan, Xu, Xingang, Yang, Guijun, Wang,
pp. 1–6. Jihua, 2018. A review of data assimilation of remote sensing and crop models. Eur. J.
Agron. 92, 141–152.
Bolser, Dan, Staines, Daniel M., Pritchard, Emily, Kersey, Paul, 2016. Ensembl plants:
integrating tools for visualizing, mining, andanalyzing plant genomics data. Plant Johnson, David M., 2014. An assessment of pre-and within-season remotely sensed variables
Bioinformatics. Springer, pp. 115–140. forforecasting corn and soybean yields in the united states. Remote Sens. Environ. 141,
116–128.
Føre, Martin, Alfredsen, Jo Arve, Gronningsater, Aage, 2011. Development of two
telemetry-based systems for monitoring the feedingbehaviour of atlantic salmon (Salmo Johnson, Michael D., Hsieh, William W., Cannon, Alex J., Davidson, Andrew, Bédard, Frédéric,
salar l.) in aquaculture sea-cages. Comput. Electron. Agric. 76 (2), 240–251. Farooq, 2016. Crop yield forecasting on the canadian prairies by remotely sen- sedvegetation
Muhammad Shoaib, Riaz, Shamyla, Abid, Adnan, Abid, Kamran, Naeem, Muhammad indices and machine learning methods. Agric. For. Meteorol. 218, 74–84.
Azhar, 2019. A survey on the role of iot in agriculture for the im-
Kamilaris, A., Prenafeta-Boldú, FX, 2018a. A review of the use of convolutional neural
plementation ofsmart farming. IEEE Access 7, 156237–156271. networks in agriculture. J. Agric. Sci. 156 (3), 312–322.

Ferreira Bocca, Felipe, Henrique A, Luiz, Rodrigues, A., 2016. The effect of tuning, feature Kamilaris, Andreas, Prenafeta-Boldú, Francesc X., 2018b. Deep learning in agriculture: a
engineering, and feature selection indata mining applied to rainfed su- garcane yield survey. Comput. Electron. Agric. 147 (February), 70–90.
modelling. Comput. Electron. Agric. 128, 67–76.
Kamilaris, Andreas, Kartakoullis, Andreas, Prenafeta-Boldú, Francesc X., 2017. A re- view on
Fleming, Aysha, Jakku, Emma, Lim-Camacho, Lilly, Taylor, Bruce, Thorburn, Peter, 2018. Is the practice of big data analysis in agriculture. Comput. Electron. Agric. 143
big data for big farming or for everyone? Perceptions in theAustralian grains industry. (September), 23–37. https://doi.org/10.1016/j.compag.2017.09.037. (ISSN
Agron. Sustain. Dev. 38 (3). https://doi.org/10.1007/s13593-018- 0501-y. (ISSN
17730155). 01681699).

Food and Agriculture Organization of the United Nations, 2018a. Help eliminate hunger, Kang, Mengzhen, Wang, Fei-Yue, 2017. From parallel plants to smart plants: intelligent
food insecurity and malnutrition.〈http://www.fao.org/about/what-we- do/so1/en/〉. control andmanagement for plant growth. IEEE/CAA J. Autom. Sin. 4 (2), 161–166.

Food and Agriculture Organization of the United Nations, 2018b. Make agriculture, Forestry Khan, Shahbaz, Tufail, Muhammad, Khan, Muhammad Tahir, Khan, Zubair Ahmad, Anwar,
and Fisheries more productive andsustainable.〈http://www.fao.org/ about/what- Shahzad, 2021. Deep learning-based identification system of weeds and crops
we-do/so2/en/〉. instrawberry and pea fields for a precision agriculture sprayer. Precis. Agric. 1–17.

Fountas, Spyros, Carli, Giacomo, Grøn Sørensen, Claus, Tsiropoulos, Z., Cavalaris, Christos, King, Brad A., Shellie, Krista C., 2018. Wine grape cultivar influence on the perfor- mance of
Vatsanidou, A., Liakos, B., Canavari, Maurizio, Wiebensohn, Jens, Tisserye, B., 2015. models thatpredict the lower threshold canopy temperature of a water stress index.
Farm management information systems: current situation and futureperspectives. Comput. Electron. Agric. 145, 122–129.
Comput. Electron. Agric. 115, 40–50.
Lee, Sungju, Ahn, Hanse, Seo, Jihyun, Chung, Yongwha, Park, Daihee, Pan, Sungbum, 2019.
Freebairn, DM, Ghahramani, A., Robinson, JB, McClymont, DJ, 2018. A tool for mon- itoring Practical monitoring of undergrown pigs for iot-based large-scalesmart farm. IEEE
soil water using modelling, on-farm data, andmobile technology. Environ. Model. Access 7, 173796–173810.
Softw. 104, 55–63.
Li, Jinhui, Tang, Yunchao, Zou, Xiangjun, Lin, Guichao, Wang, Hongjun, 2020. Detection of
Friha, Othmane, Ferrag, Mohamed Amine, Shu, Lei, Maglaras, Leandros, Wang, Xiaochan, fruit-bearing branches and localization of litchiclusters for vision-based har- vesting
2021. Internet of things for the future of smart agriculture: a com- prehensive survey robots. IEEE Access.
of emerging technologies. IEEE/CAA J. Autom. Sin. 8 (4), 718–752 ISSN 23299274.
Li, Zhiguo, Miao, Fengli, Yang, Zhibo, Wang, Hongqiang, 2019. An anthropometric study
Ge, Yuanyue, Xiong, Ya, Tenorio, Gabriel Lins, From, PÅlJohan, 2019. Fruit localization and for the anthropomorphic design oftomato-harvesting robots. Comput. Electron. Agric.
environment perception for strawberryharvesting robots. IEEE Access 7, 147642– 163, 104881.
147652.
Lioutas, Evagelos D., Charatsari, Chrysanthi, LaRocca, Giuseppe, De Rosa, Marcello, 2019.
Glaroudis, Dimitrios, Iossifides, Athanasios, Chatzimisios, Periklis, 2020. Survey, comparison Key questions on the use of big data in farming: an activity theoryapproach. NJAS
and research challenges of iot applicationprotocols for smart farming. Comput. Netw. Wagening. J. Life Sci. 90–91 (October 2018), 100297. https://doi.org/10.1016/
168, 107037. j.njas.2019.04.003. (ISSN 22121307).

Goldshtein, E., Cohen, Y., Hetzroni, Amots, Gazit, Y., Timar, D., Rosenfeld, L., Grinshpon, Y., Lobell, David B., 2013. The use of satellite data for crop yield gap analysis. Field Crops Res.
Hoffman, A., Mizrach, A., 2017. Development of an automatic monitoring trap for 143, 56–64.
mediterranean fruitfly (Ceratitis capitata) to optimize control applications frequency.
Comput. Electron. Agric. 139, 115–125. Mao, Shihan, Li, Yuhua, Ma, You, Zhang, Baohua, Zhou, Jun, Wang, Kai, 2020. Automatic
cucumber recognition algorithm for harvesting robots in thenatural environment using
Goodfellow, Ian, Bengio, Yoshua, Courville, Aaron, Bengio, Yoshua, 2016. Deep Learning 1 deep learning and multi-feature fusion. Comput. Electron. Agric. 170, 105254.
MIT Press, Cambridge.
Marko, Oskar, Brdar, Sanja, Panic, Marko, Lugonja, Predrag, Crnojevic, Vladimir, 2016.
Hamuda, Esmael, Glavin, Martin, Jones, Edward, 2016. A survey of image processing Soybean varieties portfolio optimisation based on yield prediction. Comput. Electron.
techniques for plant extraction andsegmentation in the field. Comput. Electron. Agric. Agric. 127, 467–474.
125, 184–199. https://doi.org/10.1016/j.compag.2016.04.024
Matsumura, Kanichiro, Gaitan, Carlos F., Sugimoto, Kenji, Cannon, Alex J., Hsieh, William
Hansen, Mark F., Smith, Melvyn L., Smith, Lyndon N., Salter, Michael G., Baxter, Emma M., W., 2015. Maize yield forecasting by linear regression and artificial neuralnetworks in
Farish, Marianne, Grieve, Bruce, 2018a. Towards on-farm pig face recognition using Jilin, China. J. Agric. Sci. 153 (3), 399–410.
convolutional neuralnetworks. Comput. Ind. 98, 145–152.
Moon, Aekyeung, Kim, Jaeyoung, Zhang, Jialing, Liu, Hang, Son, Seung Woo, 2017.
Hansen, MF, Smith, ML, Smith, LN, Jabbar, K.Abdul, Forbes, Duncan, 2018b. Automated Understanding the impact of lossy compressions on iot smart farm analytics, in: Big
monitoring of dairy cow body condition, mobility and weightusing a single 3d video Data (Big Data), 2017 IEEE International Conference on, IEEE, pp. 4602–4611.
capture device. Comput. Ind. 98, 14–22.
Motokura, Kurena, Takahashi, Masaki, Ewerton, Marco, Peters, Jan, 2020. Plucking motions
Hartmann, PhilippMax, Zaki, Mohamed, Feldmann, Niels, Neely, Andy, 2016. Capturing for tea harvesting robots using probabilisticmovement primitives. IEEE Robot. Autom.
value from big data-a taxonomy of data-driven businessmodels used by start-up firms. Lett. 5 (2), 3275–3282.
Int. J. Op. Prod. Manag.
Muangprathub, Jirapond, Boonnam, Nathaphon, Kajornkasirat, Siriwan, Lekbangpong,
Idoje, Godwin, Dagiuklas, Tasos, Iqbal, Muddesar, 2021. Survey for smart farming Narongsak, Wanichsombat, Apirat, Nillaor, Pichetwut, 2019. Iot and agriculture data
technologies: challenges and issues. Comput. Electr. Eng. 92 (January), 107104. analysis for smart farm. Comput. Electron. Agric. 156, 467–474.
https://doi.org/10.1016/j.compeleceng.2021.107104. (ISSN 00457906).
Mucherino, Antonio, Ruß, Georg, 2011. Recent developments in data mining and
JiHye, O., Noh, Dong-Hee, Sohn, Young-Ho, 2017. Empirical test of wi-fi environment agriculture, in: Industrial Conference on Data Mining-Workshops, pp. 90–98.
stability for smart farm platform, in: Proceedings of the 2017 4th International
Conference on Computer Applications and Information Processing Technology (CAIPT), Muhammed, SE, Marchant, BP, Webster, R., Whitmore, Andrew P., Dailey, G., E Milne, Alice,
IEEE, pp. 1–5. 2017. Assessing sampling designs for determining fertilizer practice fro- myield data.
Comput. Electron. Agric. 135, 163–174.

2
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624
Neethirajan, Suresh, 2020. The role of sensors, big data and machine learning in modern
animalfarming. Sens. Bio-Sens. Res. 29 (July), 100367. https://doi.org/10.
1016/j.sbsr.2020.100367. (ISSN 22141804).

Pantazi, Xanthoula Eirini, Moshou, Dimitrios, Alexandridis, Thomas, Whetton, RL,


Mouazen, Abdul Mounem, 2016. Wheat yield prediction using machine learning and
advanced sensingtechniques. Comput. Electron. Agric. 121, 57–65.

Quiroz, Ignacio A., Alférez, Germán H., 2020. Image recognition of legacy blueberries in a
chilean smart farmthrough deep learning. Comput. Electron. Agric. 168, 105044.

Ramirez-Paredes, Juan-Pablo, Hernandez-Belmonte, Uriel-Haile, 2020. Visual quality


assessment of malting barley using color, shape andtexture descriptors. Comput.
Electron. Agric. 168, 105110.

Regan, Áine, 2019. ’Smart farming in Ireland’: a risk perception study with key-
governance actors. NJAS Wagening. J. Life Sci. 90–91 (January), 100292. https://
doi.org/10.1016/j.njas.2019.02.003. (ISSN 22121307).

Ryu, Minwoo, Yun, Jaeseok, Miao, Ting, Ahn, Il-Yeup, Choi, Sung-Chan, Kim, Jaeho, 2015.
Design and implementation of a connected farm for smart farmingsystem. Sensors 1–
4 (2015 IEEE) (IEEE).

Sa, Inkyu, Chen, Zetao, Popović, Marija, Khanna, Raghav, Liebisch, Frank, Nieto, Juan,
Siegwart, Roland, 2018. weednet: Dense semantic weed classification using
multispectralimages and mav for smart farming. IEEE Robot. Autom. Lett. 3 (1), 588–
595.

Sarri, Daniele, Martelloni, Luisa, Vieri, Marco, 2017. Development of a prototype of


telemetry system for monitoring thespraying operation in vineyards. Comput.
Electron. Agric. 142, 248–259.

Shine, P., Murphy, M.D., Upton, J., Scully, T., 2018. Machine-learning algorithms for
predicting on-farm direct water andelectricity consumption on pasture based dairy
farms. Comput. Electron. Agric. 150, 74–87.

@MISC {no author_smart_nodate, {Smart {Farming, 2022}},{https://encyclopedia.pub/


3339}, {2021-09-18} {Accessed: 2021-09-18}.

Story, David, Kacira, Murat, Kubota, Chieri, Akoglu, Ali, An, Lingling, 2010. Lettuce calcium
deficiency detection with machine vision computedplant features in controlled
environments. Comput. Electron. Agric. 74 (2), 238–243.

Suebsombut, Paweena, Sekhari, Aicha, Sureepong, Pradorn, Ueasangkomsate, Pittawat,


Bouras, Abdelaziz, 2017. The using of bibliometric analysis to classify trends and
future directions on ”smart farm”, in: Digital Arts, Media and Technology (ICDAMT),
International Conference on, IEEE, pp. 136–141.

Sun, Xuehua, Guo, Min, Ma, Miao, Mankin, Richard W., 2018. Identification and clas-
sification of damaged corn kernels with impactacoustics multi-domain patterns.
Comput. Electron. Agric. 150, 152–161.

2
S.D. Alwis, Z. Hou, Y. Zhang et al. Computers in Industry 138 (2022) 103624

Taleb, Ikbal, Serhani, Mohamed Adel, Dssouli, Rachida, 2018. Big data quality: a survey, in: Proceedings of the 2018 IEEE International Congress on Big Data,
BigData Congress 2018 - Part of the 2018 IEEE World Congress on Services, pp. 166–173. doi: 10.1109/BigDataCongress.2018.00029.

Tetko, Igor V., Livingstone, David J., Luik, Alexander I., 1995. Neural network studies. 1. comparison of overfitting andovertraining. J. Chem. Inf. Comput. Sci. 35
(5), 826–833.

Wolfert, Sjaak, Ge, Lan, Verdouw, Cor, Bogaardt, Marc-Jeroen, 2017. Big data in smart farming - a review. Agric. Syst. 153, 69–80.

Xiong, Ya, Ge, Yuanyue, Johan From, Pål, 2020. Push and drag: An active obstacle separation method for fruit harvesting robots, arXiv preprint
arXiv:2004.09233.

You, Jiaxuan, Li, Xiaocheng, Low, Melvin, Lobell, David, Ermon, Stefano, 2017. Deep Gaussian Process for Crop Yield Prediction based on Remotesensing Data.
AAAI, pp. 4559–4566.

Zhou, Rong, Kaneko, Shun’ichi, Tanaka, Fumio, Kayamori, Miyuki, Shimizu, Motoshige, 2015. Image-based field monitoring of cercospora leaf spot in sugar beet
byrobust template matching and pattern recognition. Comput. Electron. Agric. 116, 65–79.

Sandya De Alwis is a Ph.D. candidate in the School of Information Technology, Faculty of Science, Engineering and Built
Environment, Deakin University, Australia. She obtained her B.Sc. and M. Phil in Plant Science from the University of Peradeniya,
Sri Lanka. Her current research is smart farming data mining. Her research interests in- clude machine learning, data mining, and
pattern re- cognition.

Ziwei Hou received the B.S. (Hons) degree of Information Technology from Deakin University, Australia, in 2018. She is currently a
Ph.D. student at School of Information Technology, Deakin University. Her research interests include Machine Learning, and
Natural Language Processing.

Yishuo Zhang is the PHD student in Deakin University School of IT. His research interests are data mining, AI industrial application
and also tourism demand fore- casting.

Anda mungkin juga menyukai