Intelligence driven Agricultural Ecosystem : A Review
Department of Computer Science and
Engineering Kathmandu University
Department of Computer Science and
Engineering Kathmandu University
Department of Computer Science and
Engineering Kathmandu University
Department of Computer Science and
Engineering Kathmandu University
—ICT combined with recent surge in big data technologies and high-performance computing has created incredible
potential to upgrade the current conventional way of Farming in Nepalese Domain. We present a comprehensive review on
application of Information Technology with a focus on Data Analytics and Intelligent systems in optimizing agricultural
production. We hypothesized agriculture as a linear process involving :1) Plantation Phase 2) Production Phase 3) Distribution
phase , on which we studied the potential impact of ICT. The Plantation phase reviews supervised learning in predicting suitability
of the crop for a given location acknowledging the attributes of soil like salinity , sodicity, pH and alkalinity along with average
temperature in the region and data driven yield prediction of crops. The second phase is concerned with monitoring of crop
quality during the production with advanced Image processing techniques (Deep Convolutional neural network) to detect ripening
in crops and disease with state of the art accuracy. Distribution phase studies tremendous impact of online platform for linking
right crop sellers (most likely farmers) to potential buyers thus eliminating agents from the scene. These technologies are a must to
revolutionize the traditional agricultural trend in Nepal by a smarter data driven approach.
Keywords—Soil Analytics, Precision Agriculture, Crop Management, Machine Intelligence
More than half the population of Nepal is involved in farming. To be approximate the demography shows the presence is
66% of the total population. The participation of the population is high, nevertheless the production rate is comparatively low.
The project hypothesizes agriculture as a linear process involving, three major phases.
The first being, “Plantation Phase”, which is performed through analytical outlook. Data science acts as the backbone of the
project. The second being, “Right Land Right Person”, with tourist farming. Tourist farming is the new way of farming where
people travel to farm to a new place choosing their land of interest based on the characteristics and the potential the land
bears. The Internet acts as the method of connection for the tourist. The third being, “Elevating from the infertility”, where the
land is used not only based on the characteristic it holds, but the market, trade, demand and supply, it can be addressed as
well, which sequentially allows the land to be suitable for other purposes, contributing to the economy. “Right Food Right
Person”, is a web based platform residing along, “Right Land Right Person”, which allows to provide the market to the
produced crop, finding the right person for the produced food.
1.1 Artificial Intelligence
1.1.1 Neural Networks
A Neural Network is a system planned to operate like a human brain. Human information processing takes place through
the communication of many billions of neurons connected sending signals to other neurons. Likewise, a Neural Network is a
network of artificial neurons, as found in human brains, for solving artificial intelligence problems such as image
identification, image recognition, natural language processing, topic modelling etc. Neural networks is the replication of the
human brain and the connection of neurons constituted within, but unlike human brain it is not so powerful, however with
feeding of a lot of data neural networks can even outperform human brain in a particular niche. Specifical use of Artificial
Neural Networks (ANN) can also be done for prediction of amount of crops that can be yielded. Ranjeet and Armstrong, here
 have explained it taking into account the prospect of Crops in Nepal.
1.1.2 Computer vision
Computer vision, or CV for short, is an academic term that describes the ability of a machine to receive and analyze
visual data on its own, and then make decisions about it. That can include photos and videos, but more broadly might include
“images” from thermal, or infrared sensor, detectors and other sources.
CV lies at the heart of our system, predominantly for yield prediction and detecting diseases from the plant. Images of various
stages of plant is taken and trained with the ResNet or VGG architecture, looking at the behaviour of the images. Rastogi et al.
here , has explained how leaf disease detection and grading can be done through Computer Vision.
1.2 Data analytics
Data analytics is a branch of computer science that deals with analysis of raw data to draw conclusions from the
provided information. Data analytics techniques are widely used to study the trends and measures which are hidden in a huge
mass of information. We can effectively use the collected information from data analytics to increase the efficiency of a
system. Through the analysis of data in agriculture we are able to find the primary factors that come into play for the higher
production of crops.
Machine Learning systems are data hungry, and systems like these follow the concept of Garbage-In-Garbage-Out very
strictly. So, analysis of data which empowers us to extract useful information from chunks of dataset, and figure out the useful
conclusion out of it needs to be done.
1.3 World wide web and empowered connectivity
Internet with World Wide Web, has made the world smaller than ever by dramatically impacting communication, the
major aspect of our life. Internet today connects more than 3 Billion people around the globe. Actually, the internet was
invented for the same purpose i.e. for communication. However, with the rise of 738 million users to 3 Billion people in
timespan of 18 years, the purpose of the internet has moved way ahead from just being a means of simple communication to a
source of strong and empowered connectivity.
Fig1 : Development Pipeline for Data driven Agriculture
2.1 Smart data driven plantation
2.1.1 Machine learning based potential crop suggestor
We used regression on a manually collected dataset with feature set comprising the potential crops that could be grown
given a particular location. With careful interview with domain experts on the validity of dataset as well as the feasibility of
integrating a mathematical model in crop prediction to help farmers in real life scenario, we build a DNN based
recommendation model for suggesting possible crops that could be grown on a land given feature sets :
We have coined the term “Right Land Right Crop” for this potential crop suggestion system. This DNN based
recommendation model assigns right crop to the right land, based upon the above set of features.
2.1.2 Yield Prediction
One of the most important aspects on Machine Learning use case that’s still not leveraged to the fullest in developing
countries like Nepal. Yield prediction is one of notable applications include those in the works of ; an efficient,low-cost,
and non-destructive method that automatically counted coffee fruits on a branch. The method calculates the coffee fruits in
three categories: harvestable, not harvestable, and fruits with disregarded maturation stage. The works in  used nonlinear
predictive methodology based on ANN using feature set of comprised from sensor data of soil physiological parameters
(8798 feature vectors) to associate high resolution data on soil and crop with isofrequency classes of wheat yield productivity.
For yield prediction we need to train our machine learning model particular to the plant. The training data would be the image
collected from the sensors of different stages of a particular plant. Different stages of that plant would be labeled as per their
particular stage, on a monthly basis. After data collection and labelling is accomplished, we look forward to actually training
For training the image data, ResNets and VGG are two fine architecture that can be implemented. Nevertheless in the
limited conditions like ours,VGG performs finest in comparison to ResNets.
Finally the yield prediction system understands the early behavior of the plant, through the various stages of the plant, the first
month, the second month and so on. Outlining of the perfect growth i.e harvestable or the imperfect growth i.e non
harvestable can be known in the early stages.
2.1.3 Disease Detection
With advancement of SOTA results in Image processing algorithms, more specifically, Convolutional Neural networks
(CNN), a domain of neural network focused around image dataset, there is potential to develop crop health classifier based on
leaf images to entire farm lands health over a large area based on ariel images. The author of  presented a method for
detection and screening of Bakanae disease in rice seedlings.
2.1.4 Elevating from Infertility
With hundreds of different data points, lands which are unused the reason being its infertility, can be used in some way
that fits particularly to that land. The data for a particular land, consists of not only its soil features, but also the market related
data about it. This allows to know the market exposure of that land, which powers to design the kind of farming be it
livestock, Poultry, Beekeeping, Fish farming or the kind of Agriculture that has not much of the link with the fertility of the
soil. The graph below shows the number of livestock and meat production in Nepal. Elevating from infertility
predominantly aims to increase the livestock production and take it to optimal level, sustainable for the market.
Fig2: Livestock and Meat Production in Nepal
2.2 ICT to connect Commercial farming with potential people
ICT stands for “Information and Communication Technologies”. In the past few decades, information and communication
technologies have provided society with a vast array of new communication capabilities. Using the application of one of those
capabilities, we have designed a web-based system that acknowledges two of the main components in agriculture, the
producer end and the consumer end.
2.2.1 Right Land Right Person
This module is an Internet based section, where the available land suited highly for particular crop can be assigned to
the right person for that crop. Figuring out the right person for the right land could be a challenge, nevertheless the data based
on the user i.e. provided by the user about his/her field of interest and the data attributes of the land can be compared with a
machine learning model and a unique profile about a person can be created. Making a reference to that unique profile, crops
that can interest the user can be figured out, and the land that holds the potential to produce those crops can be retrieved from
the database and catered to the user. The establishment of the connection between the available land, that is the
resource in this case and the available right person, can result in the maximum production from that land.
Right Land Right Person, can leverage the amount of crop produced from the land, as this system provides an additional
prospect of the related attributes would be registered in the Internet based communication channel, and through that channel
people can show their interest in the land, based on their experience with the attributes of the land.
2.2.2 Right Food Right Person
This section prospects to establish a supply chain between the farmer and the market of the crop cultivated by the farmer.
In simple term, the right person is assigned with the right food, through an analytical outlook of his/her preference based on
their activity over our Internet based channels. Our internet based channel basically means the internet based service provided
by the system where goods cultivated by the farmers can be sold. Based on the machine learning method of collaborative
filtering , we can assign the right kind of food to the right person, who are connected with the system over the Internet
based channel. This phase is built up with the data analysis in an economic approach, i.e. the economy of particular land needs
to be analyzed under this section.
3. PROCEDURES AND METHODS
3.1 Prerequisites for development
The development activities goes through the series of operation. These are inter-related activities, where one activity has
influence in the possibility of the other. The main activities for development are listed below.
3.1.1 Data/ Information Gathering
As this is a data driven project, the data acts as the fuel which accelerates it. The data gathering needs to be performed on
a variety of sections. The standard or benchmark data collection about some of the famous crops across the globe will act as
the training data for the algorithm of the system. Certain methodologies of gathering data/ information gathering is
implemented here. The nature of data to be collected also needs to be different. Crop suggestion system needs data which is
statistically correct, data which is presented in the form of numbers. Collection of that nature of data should be done
accordingly. While for Yield Prediction and diseases detection systems the data to be collected should be images. As both of
these systems work on image based data. So, the data/information gathering acts as the primary component for the overall
system. Both the numericals and the imagery data are the subject of interests.
In the figure below, we map districts into a two dimensional space and cluster (left) them based on similarity of their annual
temperature dataset (right). Just using the annual temperature as a feature vector we were able to develop a prototype model
that could cluster regions based on their temperature similarity. This is a small potential of how Data driven algorithms help
us exploit information that seems so obvious. If we try to optimize our model adding additional parameters like soil
properties, Wind speed and rainfall dataset, we could obtain a clustering model of reasonable accuracy. This information can
be exploited by government when developing agricultural agenda for a district. By analyzing similar other zones,
simultaneous actions could be taken, thus accelerating the work.
Fig3: Temperature based cluster of districts of Nepal
3.1.2 Developing the Internet Based Channel
The sections “Right Land Right Person” and “Right Food Right Person” has direct link with this Internet based channels.
Internet based channels primarily means those platforms connected over internet designed for registration, profiling and
communication of the end users. End users here essentially means the consumer of the crop and the producer.
Database design and development, communication channel establishment, chat channel and the telephone channel are some of
the activities which make up this section.
3.1.3 Training the number of Machine Learning Model
There is no value for data unless we extract some patterns or the meaning out of it. The collected data needs to be
imposed to certain labels and some recursive pattern needs to be extracted from the similar kind of data, that differentiates it
among the other set of data. So, training a model allows the system to recognize the recursive pattern over similar kind of data
and eventually recognize the uniqueness. We use ResNet and VGG architectures for training the dataset.
Device with Nvidia GTX 1050i can be the essential hardware for training the dataset. Using CUDA - a parallel computing
platform and application programming interface model created by Nvidia, tensorflow-GPU was able to use CUDA-enabled
graphics processing unit for the processing.
The above mentioned activities can be built upon certain procedures and methods. The completion of the possibility of the
technology exists following these procedures.
The term “data flows”, means data is very vibrant, it moves, it transfers from one place to the other. The explicit
vision of the technology makes the data collection much explicit. The collection of data can For this, the organization
such as research gate, act as the major source. The standard data can be collected from the research gate, where
number of scientists come together and provide with authentic datasets. Second, the land data collection, this activity
is chained up with number of other activities. For this section the soil needs to be collected from the particular land,
that needs to be tested, the soil collection can be done in the similar approach, that is in practice now. That particular
soil, then needs to pass through the test that divides it into various attributes, those attributes which will be
predefined. The out-coming attributes value will act as the data for the system. Also, opendatanepal  has several
datasets on Estimated Production of various cash crops in 75 districts of nepal. According statical analysis from
Ministry of agricultural department, Nepal 1625 metric ton maize produced on year 2072/73. Also Nada(National
Data Archive) provides census and demographic data of various years 
A composite soil quality rating system using a weighted ranking procedure for soil textural class, soil organic matter,
pH and major nutrients may be a feasible semi-quantitative approach for assessing the quality of soil in relation to
productivity and susceptibility to erosion. The rating index, however, requires additional testing and validation
systematically across the wide range of agro ecological zones of the region.
●Communication in agriculture
Internet is one of the biggest connectivity platforms humans have ever created. To make use of this connection a
platform is a must. The system’s website is the place where the publishing of the result after the data manipulation
needs to be performed. The part of chatting system, communication channel, user support and user guidelines are
achieved within the web based channel sections.
According to nepal telecom There are more than 60% of people with internet access .
This increases the possibility that more number of farmers can access the platform for the help and suggestions. The
platform must integrate following basic services.
1. A simple interface where farmers can input the data and get suggestion.The simple interface should be
stressed as much as possible because our users are not tech savvy people. So it is of utmost importance that
anyone should be able to use the system regardless of his educational background.
2. A discussion form to interact with other people and share their experiences and opinion
3. A chat platform where farmers can connect with government officials or experts on agriculture.
4. A feedback system where farmers can report the success of the crops suggested by the system. It will help
generate more data which can be used to further improve system.
This acts as the face of the whole system. The system needs to express itself, for it to be used and the expression is
possible under this section. After the output from the data manipulation section comes with a proper order, the output
needs to be published to the world. Proper user interface design with explicit explanation about factors such as the
potential outcome, income and also elements such as the weather, environment of any publishing land is mentioned
under this section of the web portal.
The new technology remains unknown and unexplored until it's among the audience. To make it involved among
them it should be able to provide the actual need and should change what user thinks is not required. The feedback
section is done after web publishing, as it provides further suggestion about the system.
4. CONTRIBUTION TO OTHER TECHNOLOGIES
4.1 Data Ecosystem
Data acts as the soul for this technology. The whole system has a direct relationship with the data ecosystem. Activities of
data processing and data explanation is crucial of the system. So, the whole data related activity can contribute in the technical
improvement of data mining and data manipulation. There is still a long way to go in the field of data ecosystem in our
country and with coming census of 2021, the data related technology used in this system can contribute to a fine extent.
4.2 Artificial Intelligence
Data attains its full potential when its calculated, elaborated and concluded, and successive activities to accomplish, we
need an intelligence. The system is based on the intelligence generated by the computer, thus the second field it can contribute
in is Artificial Intelligence (AI) . There is a whole wave flowing across the globe, related to this kind of intelligence, the
reason it being faster, cheaper, accurate and diligent. The technology we are building has predominant use of artificial
intelligence in the field of agriculture. Farming can be benefited enormously with AI. Malnutrition in crops, satellite image
processing, drip irrigation, Internet of Things (IOT) are some of the application in this field.
4.3 Soil technology
The section of the system, “Right Land Right Crop”, is designed to provide a new horizon to the soil technology. Talking
about the present technology for soil in Nepal, when a particular soil is tested it is divided into handful number of attributes,
some of them being amount of Nitrogen, Phosphorous, and Potassium in the soil, along with its PH value. While this
technology defines the soil into some additional attributes as well. This can be a push to the existing soil technology as it lacks
behind, testing of some of the important attributes of the soil which has the potential to define, how the outcome from the soil
looks like. Further, the computation of soil data in an artificial intelligence approach is one of the vital technical lifts that the
system can provide.
In a nutshell, the field the researched technology tries to converge is the blend of technology, in agriculture. The correct
statement would be, “the technology for agriculture using data analytics”, this is the target point of the whole system.
5. EXPECTED OUTCOMES FROM THE SYSTEM
The researched technology moves through the path which addresses data and agriculture. In the developed nations these two
sectors go side by side for sustainable results, nevertheless in the South Asian nations the approach is still orthodox. This
technology can indeed supply a push to these two pillars of prosperity.
Highly anticipated outcomes can be better known with the division, based on the sectors.
5.1 Probabilistic agriculture
In the field of agriculture the researched technology produces the mathematical probability of a crop on a particular land,
which in turn helps to make strategies taking in priority those crops with a differentiated approach.
5.2 Data driven change agents
Every sector must pass through changes overtime for the elevated growth, and there must exist those factors for every sector
which catalyze its growth and people related with it. This researched technology can identify the change agent, for the land
across the country with the help of analytical outlook. Further, the expected outcome will be so decentralized, that it holds the
caliber to find the changing agent on land by land basis and mobilize it. Yes, the point is not only about finding the changing
agent it is also about the mobilization of the changing agent.
5.3 Data aware literacy
In the 21st century, there still exist a number of people who have been through their formal education. In a developing
nation like ours, the number is significantly higher. Nevertheless in this silicon age, the data driven farming can provide data
aware literacy to these people, making them bear the potentiality to perform the data powered agriculture. As, the farming is
to be done on a specific crop, the knowledge of that specific crop can be provided through the data.
While Data is more important than ever before, minimal effort is seen from government site to actually carry out mass data
collection program suited to applying Machine learning or for mere basic analytics.
●Farmers to trust on a website that advocates them to substitute their practices and plant something new in their land
or use some pesticide in their land?
●How many farmers have access to internet from that 60% internet users in nepal? Maybe a web platform is not the
best solution right now in the context of nepal.
●Testing the system itself is a challenging task because a large number of days is required for production of any crop?
 Ranjeet, T., Armstrong, L.(n.d.). An Artificial Neural Network for Predicting Crops Yield in Nepal.
 Rastogi, A. (n.d.). Leaf disease detection and grading using computer vision technology & fuzzy logic.
 Amatya, S.; Karkee, M.; Gongal, A.; Zhang, Q.; Whiting, M.D. Detection of cherry tree branches with full
foliage in planar architecture for automated sweet-cherry harvesting. Biosyst. Eng. 2015, 146, 3–15.
 Pantazi, X.E.; Moshou D.; Alexandridis T.; Whetton R.L.; Mouazen A.M., Wheat yield prediction using machine
learning and advanced sensing techniques. Computer and Electronics in Agriculture
 Chung, C.L.; Huang, K.J.; Chen, S.Y.; Lai, M.H.; Chen, Y.C.; Kuo, Y.F. Detecting Bakanae disease in rice seedlings
by machine vision. Comput. Electron.
 NepalMap profile: Nepal. (n.d.). Retrieved from https://nepalmap.org/profiles/country-NP-nepal/
 Schafer, B., Frankowski, D. (n.d.). Collaborative Filtering Recommender Systems.
 (n.d.). Retrieved from
 Bajracharya, R. M. (n.d.). Soil quality in the Nepalese context - An analytical review.
 Internet users in Nepal increases rapidly, penetration reaches 63 percent. (2019, March 12). Retrieved from