Analysis of deep learning techniques for the identification and classification
of crops
Análisis de técnicas de deep learning para la identificación y clasificación de cultivos
Hover Hernán Torres Gutama.
Student, Universidad Católica de Cuenca, San
Pablo de La Troncal campus.
hover.torres@est.ucacue.edu.ec
https://orcid.org/ 0000-0002-8590-8207
Luis Stalin Jara Obregón
Engineer, Universidad Católica de Cuenca campus
San Pablo de La Troncal lsjaraob@ucacue.edu.ec
https://orcid.org/0000-0003-4958-5698
Abstract
The purpose of the study is to make known the importance of the use of computational techniques in
agriculture, the use of deep learning allows artificial neural networks to perform data analysis through
the use of preposition logic, this makes computers capable of to identify images or make predictions.
The research methods used in the study were exploratory, bibliographical and descriptive research, they
were very helpful to analyze these modern computational techniques. In the investigation it was
determined that these methods are currently widely used in agriculture, being the convolutional neural
network algorithm the most used for this objective. The use of these techniques is shown as one of the
most important keys for the growth of precision agriculture.
Key words: Deep learning techniques, identification, crop classification, neural networks, agriculture.
Resumen
El estudio tiene como finalidad dar a conocer la importancia del uso de técnicas computacionales en la
agricultura, el uso del deep learning permite que las redes neuronales artificiales puedan realizar análisis
de datos mediante el uso de la lógica preposición, esto hace que las computadoras sean capaces de
identificar imágenes o realizar predicciones. Los métodos de investigación empleados en el estudio
fueron la investigación exploratoria, bibliográfica y la descriptiva, fueron de mucha ayuda para analizar
estas técnicas computacionales modernas. En la investigación se determinó que estos métodos son muy
utilizados actualmente en la agricultura, siendo el algoritmo de redes neuronales convolucionales el más
utilizado para este objetivo. El empleo de estas técnicas se muestra como una de las claves más
importantes para el crecimiento de la agricultura de precisión.
hover.torres@est.ucacue.edu.ec
http://centrosuragraria.com/index.php/revista, Published by: Edwards Deming Institute,
Quito - Ecuador, July - September vol. 1. Num. 14. 2022, This work is licensed under a
Creative Commons License, Attribution-NonCommercial-ShareAlike 4.0 International.
https://creativecommons.org/licenses/by-nc-sa/4.0/deed.es
Received January 09, 2022
Approved: June 09, 2022
22
Palabras clave: Técnicas de deep learning, identificación, clasificación de cultivos, redes neuronales,
agricultura.
Introduction
After the sanitary emergency declared by the covid-19 disease at the beginning of 2020, the
world has turned its attention to an important sector for our subsistence such as agriculture,
because we have been used to everything going well in our daily lives, such as turning on the
water faucet of our sink, turning on the electric light switch, going to the supermarket to buy
vegetables and fresh fruit, without thinking about all the processes behind each result that
seems so normal to us.
Agriculture, like many sectors today, has undergone profound changes such as the development
of electronics and information and communication technologies that favor a term in diffusion
such as precision agriculture. This technological advance has reached a level that allows the
farmer to measure, analyze, and manage the variability within the plots and extensions of land
of their crops. (Garcia, Martinez, & Garcia, 2018).
An important topic in precision agriculture is the identification and classification of crops
through satellite images and images captured by unmanned aerial vehicles (UAV), for which
different techniques and methods have been developed. In our study we will focus on Deep
Learning, which is a computational model that allows artificial neural networks to perform data
analysis through the use of preposition logic, this makes computers able to identify images or
make predictions.
In the field of agriculture, Herrara (2016) states that Deep Learning is able to identify pests,
fungi and other plant diseases. According to the data provided by this computer system,
appropriate treatments can be given according to environmental factors such as weather, wind,
floods, droughts that directly affect crop yields. Therefore, the technology allows the farmer to
have accurate information which allows anticipating adverse conditions to achieve crops with
high yields and consequent profitability.
The study on the analysis of deep learning techniques for crop identification and classification
is important because it allows an efficient management of land extensions, for which we will
divide this paper into 3 sections. Section 1 will introduce the reader to the main concepts and
definitions involved in deep learning. In section 2 we will analyze the state of the art on the
topic of study. In section 3 we will show the most widely applied deep learning techniques and
algorithms for crop identification and classification. In section 4 we will present the
conclusions and recommendations of the study.
Torres and Jara 2022
July - September vol. 1. Num. 14 2022
Materials and methods
Machine learning is an evolving branch of computational algorithms that are designed to
emulate human intelligence by learning from the surrounding environment. They are
considered the workhorse in the new era of so-called big data. Machine learning-based
techniques have been successfully applied in diverse fields ranging from pattern recognition,
computer vision, spacecraft engineering, finance, entertainment and computational biology to
biomedical and medical applications. (El Naqa, 2015)
In general, existing AI algorithms make use of computational power to solve certain types of
problems, but they do so in a specific way. Through some input data, an algorithm learns to
classify information or to make predictions through concrete patterns. It is something that gives
the figure of being extraordinary, but what it does is to use that input data to refine its
identification or prediction, being this prediction of the algorithm, which, instead of being
static, is a dynamic learning process that varies as new data enters.
What differentiates Artificial Intelligence from other computer programs is that it does not have
to be programmed specifically for each scenario. We can teach it things (Machine Learning),
but it can also learn by itself (Deep Learning). While there are multiple variants of each, they
can be broadly defined as follows. (Alonso, 2021):
• AI (Artificial Intelligence): a machine that is capable of imitating human reasoning.
• ML (Machine Learning): a subset of Artificial Intelligence where people "train"
machines to recognize patterns based on data and make their predictions.
• DL (Deep Learning): a subset of ML in which the machine is able to reason and draw its
own conclusions, learning by itself.
Deep learning is an artificial intelligence algorithm that seeks the recognition of images to
identify complex patterns. One of its main characteristics is that it presents automatic learning,
i.e. unsupervised. (Herrera, 2016). In this sense, Deep Learning artificial intelligence is based
on neural networks inspired by human networks, which allow to make predictions according
to the needs. All this allows the system to record the data obtained through images, taking into
account variables such as climate, temperature and humidity of the crops, which can affect their
production performance.
By means of Deep Learning, artificial neural networks can be designed, using an infinite
amount of data for their training. It is important to note that this system can be implemented
through computer programs to create artificial neurons and then use them to simulate the
functioning of a biological neural network. (Parraga, Alcivar, Riascos, & Becerra, 2020).
Neural networks are responsible for processing images through a computational model, which
can identify the characteristics of plants, such as leaves, spots of different colors, which will
24
be very useful to identify or classify images and detect patterns or objects that will allow the
identification and classification of crops.
Deep Learning has a wide use in agriculture since it allows performing many processes that
are nowadays automated. With the help of this system it is possible to detect any type of
diseases in crops, which is essential to obtain a good production and save money and labor.
(Herrera, 2016).
Among the main advantages offered by Deep Learning in agriculture are:
• Improving crop yields
• Predicting crop production
• Knowing in advance about climate change, whether cold or heat waves. (Marengo, et al.,
2020)
Deep Learning in agriculture, therefore, offers many advantages as it can control large
agricultural plantations and improve crop yields and, of course, get the most out of the land.
Deep Learning techniques for the identification and classification of crops.
The Deep learning techniques most commonly used in agriculture are:
• Random forest
It is a supervised learning algorithm, it creates a forest and randomly takes several trees, which
allows it to accurately predict the state of the crop.
• Support Vector Systems (SVM)
This vector system is used in classification and regression tasks, because it allows to obtain
artificial vision images. One of the advantages of this system is that its error level is low and
the results are easy to interpret.
• Artificial neural networks
This system is based on the structure of the human brain, which allows it to transmit
information. In other words, artificial neural networks can recognize and send images and
voice.
Torres and Jara 2022
July - September vol. 1. Num. 14 2022
• Convolutional networks.
These networks are specialized for processing data to locate patterns in images, objects and
scenes. The images can be classified and eliminate those that are not necessary to be more
accurate with the data being searched.
Similarly through Deep Learning techniques, the farmer can predict crop yields accurately,
because this system can provide important data about the crop, such as the quality of the
product and the possible price in the market when the product is already harvested. This data
allows the farmer to know in advance which products he can plant and whether or not they will
generate profit when they are already cultivated. (Big Data Site, 2019).
The study on the analysis of Deep Learning techniques for the identification and classification
of crops is qualitative in approach because its purpose is to investigate and interpret the topic
of study through bibliographic sources, since it seeks to analyze how the technology can be
used to identify and classify crops, through which the farmer can make timely decisions to
avoid economic losses. Authors such as Hernández, Fernández and Baptista (2014)mentioned
that exploratory research allows to observe from the same actors in the research process,
which has allowed to know closely the most important facts about the application of
convolutional neural network technology to identify crops through satellite images such as
Landsat-8, Sentinel-1 and Sentinel-2.
Through bibliographic research it was possible to obtain information from several authors in
order to find solutions to the problems posed by means of a two-way analysis, in which it has
been related to existing data from different sources. That is to say that bibliographic sources
allow the researcher to know the coherences that exist between the theory and the data
referring to the study, so it is necessary for the researcher to use reliable sources and be able
to carefully compare with the data of similar authors.
The study is descriptive, since it will allow a detailed description of the analysis of Deep
Learning techniques for the identification and classification of crops. In this regard, Gutiérrez
and Rosas (2014)point out that descriptive research makes it possible to analyze or describe
the phenomena under study, in order to learn more about the problem.
According to Guevara, Verdesoto, and Castro (2020)through descriptive research, the
researcher can know in depth the subject under study, since he can explain step by step the
research process in order to analyze the how and why of the problem under study. Such is the
case of the study on the analysis of Deep Learning techniques for the identification and
classification of crops.
In this regard, Hernández and Mendoza (2019)mention that in a descriptive study several
concepts and variables can be selected, to measure them independently, with the purpose of
describing the causes and consequences that originate the phenomena. According to the
26
aforementioned, descriptive studies offer the possibility of making predictions, to obtain an
estimate of how the phenomena will affect in the future.
The methods used in the study on the analysis of Deep Learning techniques for the
identification and classification of crops were the deductive method, since it allows the analysis
of postulates, theorems, laws, principles, among others. Taking into account that this method
is based on the universal application of facts or solutions to apply them to particular solutions.
While the deductive method was used to elaborate conclusions by means of particular sections,
which are accepted as valid.
Result
According to Baruffaldi's results (2019)Deep Learning techniques allow the farmer to observe
the state of the crops through a video. For example, he can identify where weeding is required
in the planting, as well as whether the crop growth is adequate. The study shows that Deep
Learning techniques offer many advantages for agriculture because they can identify climatic
changes, whether hot or cold. Another important fact presented in this study is that these
techniques can predict the quality of crops, so the farmer can take advantage of these benefits
to optimize the production of different crops.
In this research it was possible to show that Deep Learning, allows farmers to yield much more
production of their crops and significantly lowering the economic losses that can generate crop
failures. In fact, by using Deep Learning, the farmer can supervise the entire planting process
up to the moment of harvesting.
In this regard, Jácome (2021)states that the most widely used crop prediction systems are those
that learn based on image and text data, in order to guarantee adequate learning.
Deep Learning systems allow crop predictions to be made in 25%, based on data obtained
through image and text, 21% is done by means of audio and video, 17% is performed by means
of images and optical flows, 10% is performed based on soft tissue studies also known as MRI,
PET. 9% is performed by gene expression, 5% is performed by medical imaging, 4% is
performed using multispectral imaging and 2% is performed by gene expression.
According to the study presented by Banchón Fajardo and Soriano Navas (2021)they point out
that by using Deep Learning in crops, the farmer can know in advance the weather predictions
such as winter or summer and which product may be the most suitable for planting, to achieve
greater profitability in both crop and economic terms. In this sense, Bonilla et al.
(2021)mentions that the machine learning system can be applied in various types of crops such
as sugar cane plantations, vegetable crops, tubers and other agricultural products.
Torres and Jara 2022
July - September vol. 1. Num. 14 2022
According to these data it can be seen that the application of the Deep Learning system in sugar
cane cultivation is represented by a total of 41%, while in the cultivation of vegetables is used
14%, in what has to do with the cultivation of tubers this system is used in 12%, while in other
crops is used in 33%. These data show that Deep Learning is a technological tool that brings
many advantages for agriculture, because through videos and digital photographs the farmer
can maximize crop yields, achieving greater economic profitability. Similarly, he can control
crop diseases and act in time to reduce losses. For this and many other reasons, Deep Learning
has become indispensable in agriculture.
Therefore, Deep Learning is widely used in agriculture around the world, because it can predict
with a high degree of accuracy on the yield and needs of crops, through visual data and text
that can improve the conditions and profitability of crops.
The literature is largely based on the processing of satellite images, but currently we have a
growing number of studies applying UGV (Unmanned Aerial Vehicles) images as a remote
sensing platform and classical machine learning algorithms with classification, clustering and
detection logics.
The most widely used technique for this type of detection and classification are convolutional
neural networks, which show better results than a traditional multilayer feedforward network.
The main drawback to improve the results with images from unmanned aerial vehicles is that
these devices have little memory and storage capacity, studies could be conducted to send data
to cloud platforms, but there would be a drawback in the transmission of data from the device
to the database in the cloud. As a recommendation for future work, studies could be conducted
to implement Deep Learning Techniques in agriculture in order to obtain benefits in crops,
because in Ecuador this technique is rarely used.
Conclusions
The use of Deep Learnig techniques allows the farmer to obtain better yields in agriculture and
achieve greater profitability. In addition, the techniques can be applied to make weed control
more efficient, especially in large areas, which simplifies costs and reduces the use of
herbicides. Automatic crop classification using new deep learning techniques is one of the most
important keys to the growth of precision agriculture. It allows reducing the number of workers,
the amount of chemicals applied, among others.
References
AgemPapa (2021). Technical Information: How to predict yield in potato crop with precision
agriculture? Retrieved from https://www.argenpapa.com.ar/noticia/10197-informacion-
tecnica-como-predecir-el-rendimiento-en-el-cultivo-de-papa-con-agricultura-de-
precision
28
Andrade, J. A. (2018). Mobile application for detection and treatment of crop damage in the
Taura parish of Durán canton, using image analysis software based on Machine Learning
techniques. Retrieved from http://repositorio.ucsg.edu.ec/bitstream/3317/11769/1/T-
UCSG-PRE-ART-IPM-150.pdf
Banchón, D. A., & Soriano, G. K. (2021). Design of technological platform based on machine
learning to control the factors involved in aeroponic crop production. Retrieved from
http://repositorio.ug.edu.ec/bitstream/redug/52322/1/B-CINT-PTG-
N.660%20Banch%c3%b3n%20Fajardo%20David%20Alberto%20.%20Soriano%20Na
vas%20Geomara%20Katherine.pdf
Baruffaldi, J. M. (2019). Adversarial neural networks for weed recognition. Retrieved from
http://www.clei2017-46jaiio.sadio.org.ar/sites/default/files/Mem/EST/est-05.pdf
Bonilla, J. S., Dávila, F. A., & Villa, M. W. (2021). Study of the use of artificial intelligence
techniques applied to soil analysis for the agricultural sector. Retrieved from
https://recimundo.com/index.php/es/article/view/978/1672
Carderón, A., & Hurtado, H. (2019). Micro-services based architecture for web applications.
TIA (Technology and Academia Research) 7(2), 55-61.
Garcia, J. J., Zambrano, J. J., Alcivar, R., & Zambrano, W. D. (2020). Agricultural crop yield
prediction using machine learning. Retrieved from
https://dialnet.unirioja.es/descarga/articulo/7672167.pdf
Guevara, G. P., Verdesoto, A. E., & Castro, N. E. (2020). Educational research methodologies
(descriptive, experimental, participatory, and action research). Scientific journal World
of Research and Knowledge. Retrieved from
https://dialnet.unirioja.es/descarga/articulo/7591592.pdf
Gutiérrez Ríos, Y., & Rosas de Martínez, A. I. (2014). The place of orality in school: initial
explorations on teachers' conceptions. Caldas: Universidad Distrital Francisco José de
Caldas.
Hernández, R., & Mendoza, C. (2019). Research methodology. Quantitative, qualitative and
mixed routes. Retrieved from https://virtual.cuautitlan.unam.mx/rudics/?p=2612
Hernandez, R., Fernandez, C., & Baptista, P. (2014). Metodología de la Investigación. Mexico
D.F: Mc Graw Hill.
Herrera, C. A. (2016). Implementation of a statistical and predictive analysis module for
agriculture using bigdata and machine learning, integrated to the Iotmach system.
Torres and Jara 2022
July - September vol. 1. Num. 14 2022
Jácome, R. (2021). Crop yield prediction using multimodal deep learning. Retrieved from
http://refbase.cidis.espol.edu.ec/files/robertojacomegalarza/2021/150_RobertoJacomeG
alarza2021.pdf
Jara Estupiñan, J. G. (2016). Implementation of support vector machine (SVM)-based
algorithms for electrical systems: topical review. Tecnura Journal, 20(48), 149-170.
Laurentin, H. (2020). Importance of yield prediction in sugarcane in a context of digital
transformation. SofOS.
Marengo, J., Chou, S., Torres, R., Giarolla, A., Alves, L., & Lyra, A. (2020). Climate change
in Central and South America: Recent trends, future projections, and impacts on regional
agriculture.
Ochoa, C. (2015). Implementation of information and communication technologies ICT for the
improvement of accounting and financial management in Fundimetales Company.
Retrieved from https://repositorio.uptc.edu.co/bitstream/001/1551/1/TGT-287.pdf
Parraga, J., Alcivar, R., Riascos, J., & Becerra, M. (2020). Aphids Detection on Lemons Leaf
Image Using Convolutional Neural Networks.
Big Data Site (2019). Machine learning in agriculture: applications and techniques. Retrieved
from https://sitiobigdata.com/2019/12/24/aprendizaje-automatico-en-agricultura/#
Zanelli, E. A. (2019). Development of an agricultural crop yield estimation method using
optical satellite imagery in Buenos Aires province. Obtenido de
https://rdu.unc.edu.ar/bitstream/handle/11086/16505/Alvarez%20Zanelli%2C%20E.%2
0Desarrollo%20de%20un%20m%C3%A9todo%20de%20estimaci%C3%B3n%20de%
20rendimiento%20de%20cultivos..%20.pdf?sequence=1&isAllowed=y
