AGROVISION : YOUR AI COMPANION FOR SMART FARMING

Komal Jadhav Jaswant Singh

Artificial Intelligence Artificial Intelligence
G.H Raisoni College of Engineering & Management, G.H Raisoni College of Engineering & Management, Pune
Pune jaswantsingh0292@gmail.com
jadhavkomal56@gmail.com
Akash Oad
Shanuj Tiwari Artificial Intelligence
Artificial Intelligence G.H Raisoni College of Engineering & Management, Pune
G.H Raisoni College of Engineering & Management, akashoad2121@gmail.com
Pune
shanujtiwari2003@gmail.com Dr. Sonali Sonavane
Artificial Intelligence
Bhagyashri Wankar G.H Raisoni College of Engineering & Management, Pune
Artificial Intelligence sonali.sonavane@gmail.com
G.H Raisoni College of Engineering & Management,
Pune
bhagyashri.wankar@gmail.com

Abstract: The agriculture sector plays a very important role in and early disease detection. AgroVision is designed to enhance
the Indian economy, and nearly 48% of the country's population decision-making by providing real-time, actionable insights,
depends on it for their livelihood. It contributes around 17% to enabling farmers to optimize agricultural, reduce risks, and
its GDP. The study here uses machine learning algorithms promote sustainable farming practices.
designed to enhance agricultural practices by soil quality 1. Crop Recommendation System: This particular
analysis, suggesting crops and fertilizers used, and detection of characteristic allows producers to decide what crops are most
diseases in plants. The system, through advanced classification suitable for planting given any specific conditions of the soils
models and image recognition techniques like Random Forest and weather. It considers variables such as soil fertility,
and Convolutional Neural Networks, is able to predict the best moisture content, and patterns of local climate. Using advanced
crop choices, offer fertilizer recommendations for improving machine learning algorithms, AgroVision can provide
soil health, and identify diseased plants with treatment advice. recommendations tailored to the unique circumstances of each
This web platform allows data-driven decision-making among farmer. This means that farmers can choose crops that are not
the farmers to shift from an intuitive approach to farming based only suitable for their soil but also predicted to perform well in
on scientific methods. That eventually leads to the practice of the prevailing weather conditions, aiming at increased yields
sustainable agriculture, increases crop yields, and provides and better productivity.
greater economic resilience to farmers across India, ultimately 2. Fertilizer Recommendation System: The fertilizer
benefiting food security and rural development. recommendation module advice can be developed in the type
and amount of fertilizers. Considering the nutrient content of
Keywords: Agrovision, Agriculture, Machine Learning, the soil and the demand requirements of the crops, this module
Disease detection, Recommendation system, Smart Farming ensures that the right fertilizers are used at the right time by
farmers. This is how targeted application improves soil health
I. INTRODUCTION
and supports the optimal growth of crops. Since real-time data
Agriculture is one of the vital sectors of India's economy, as it integration is possible, farmers could consider changing their
provides employment to almost half of its population and fertilization strategies over fluctuating conditions to make
significantly contributes to the country's GDP. However, farming more efficient and sustainable.
traditional farming has several hurdles that reduce its 3. Plant Disease Detection: Advanced image recognition
productivity not only through resource wastage but also affect technology is used to identify plant diseases early on within this
the environment negatively. It involves inefficient crop module of AgroVision . Analysis of the images of plant
selection, poor fertilizer management, and delayed plant leaves helps differentiate between healthy and diseased plants.
diseases. With the advancement of modern technologies, Early detection enables farmers to respond appropriately, thus
particularly ML and DL, there is great potential for finding preventing the spread of diseases and loss in crops. Therefore,
solutions to these problems present in agricultural fields with not only does it keep crops healthy, but it also reduces the
precision farming solutions overuse of pesticides, making farming more sustainable
AgroVision is an innovative project that seeks to harness the
power of ML and DL to offer farmers intelligent, data-driven
tools that improve crop management, soil health optimization,
 II. LITERATURE SURVEY yields. The AgroVision project seeks to tackle these challenges
by utilizing machine learning and deep learning technologies. It
This emerging use of ML and DL in agriculture has proven to provides real-time data to farmers, enabling them to make
be the driving force behind many studies in crop informed decisions on crop selection, fertilizer application, and
recommendations, optimization of fertilizer, and detection of early disease detection, promoting sustainable farming and
plant diseases. These studies thus lay the basis for the improving overall agricultural productivity.
Agrovision project, which converges data-driven technologies
to offer smart farming solutions. Below are summaries of some IV. METHODOLOGY
of the key research papers applicable to the development of
Agrovision. The methodology of the AgroVision project is organized into
five main steps, each essential for providing accurate and real-
[1]
In the article, the authors looked into how it's possible for time insights to farmers. By integrating machine learning (ML),
deep learning (DL), and a user-friendly web interface,
CNNs to identify plant diseases from leaf images. Taking into
AgroVision creates a comprehensive decision-making tool that
consideration image processing and ML, the application was addresses the unique challenges faced by farmers in modern
able to obtain a high precision in disease identification. This agriculture.
paper lays the groundwork for Agrovision’s plant disease
detection module, which helps farmers identify crop diseases 1.Data Collection: The first step in AgroVision involves
early and take preventive measures. gathering a variety of secondary datasets crucial for informed
agricultural decisions. This includes detailed soil composition
[2] data—such as pH levels, moisture content, and nutrient profiles
This paper discusses the integration of machine learning with
including nitrogen (N), phosphorus (P), and potassium (K)—
diverse datasets, including soil composition, weather patterns, which are essential for understanding soil fertility. Additionally,
and historical crop yields, to provide accurate crop and fertilizer weather information covering temperature, humidity, and
recommendations. The research highlights the importance of rainfall patterns is collected to provide context for crop growth
data-driven agriculture, which aligns with Agrovision’s goal to conditions. [20], [21]
offer precise crop and fertilizer suggestions, enabling farmers to
optimize resource use and improve sustainability.

[3]
This research used ensemble learning methods, such as
Random Forest and Stacking, to improve crop
recommendations' robustness and reliability. The finding from
the results of the paper directly impacted the selection of
algorithms for selecting algorithms for the crop
recommendation system by Agrovision.

[4]
This paper examines the role of weather forecasting in
optimizing crop selection. By using Long Short-Term Memory
Fig 1. Dataset
(LSTM) networks to analyze historical weather data, the authors
successfully predicted weather patterns and their impact on crop
The datasets also encompass various crop types grown in
growth. Agrovision incorporates these insights into its crop
different regions, which helps tailor recommendations to
recommendation system, providing farmers with
specific local conditions. Furthermore, a large collection of
recommendations that factor in future weather conditions.
plant disease images is prepared, featuring labeled images of
[5] both healthy and diseased crops. These images are vital for
The authors of this paper used image processing and ML
training deep learning models focused on detecting plant
algorithms to classify plant diseases based on leaf images. They
diseases accurately. The collected data forms the foundation for
applied techniques such as segmentation and feature extraction
training the ML and DL models, enabling AgroVision to
to isolate diseased regions. The research supports the image
provide precise crop and fertilizer recommendations as well as
recognition capabilities of Agrovision’s plant disease detection
early disease detection [22], [23].
tool, which uses ML models to accurately diagnose crop
diseases and suggest treatments.
2. Data Preprocessing: After collecting the data, it undergoes
[6] preprocessing to ensure it is clean and ready for analysis. This
This study investigates the application of big data analytics in
involves several critical steps, including data cleaning to
precision agriculture, focusing on weather forecasting. By
remove inconsistencies, missing values, or noise that could
analyzing large datasets, the authors improved weather
affect data quality and model performance. Normalization is
predictions, helping farmers make informed decisions about
then performed to scale the data so that all features are
planting and harvesting. Agrovision draws from this research to
comparable; this aids machine learning algorithms in
incorporate weather-based insights, enhancing the accuracy of
interpreting it correctly without bias toward any particular
its crop recommendation system.
feature. Finally, feature engineering extracts key characteristics
III. PROBLEM STATEMENT related to soil, weather, and crops to create new variables that
enhance the model's predictive performance. For example,
Agricultural productivity in India is significantly hindered by combining temperature and humidity data can yield insights
issues such as improper crop selection, incorrect fertilizer use, into optimal growing conditions. These well-processed features
and delayed disease detection. Farmers often lack access to allow the models to make more accurate recommendations
accurate information on soil health, climate conditions, and based on the input data [24], [25].
suitable crops, leading to poor decision-making and suboptimal
3. Algorithm Development: The core of AgroVision lies in Crop Recommendation System: The system collects environm
developing algorithms that utilize both machine learning and ental factors (precipitation, temperature, humidity) and soil prop
deep learning techniques to generate intelligent erties (pH, N, P, K) to assess soil health and growing conditions.
recommendations for farmers. Various models are created using
Data preprocessing involves cleaning, transformation, and featur
popular frameworks like TensorFlow and PyTorch, which
facilitate the implementation of complex algorithms efficiently. e reduction to ensure high-quality inputs. The data is split into tra
For plant disease detection, Convolutional Neural Networks ining and testing sets, and algorithms like SVM, Naive Bayes, K
(CNNs) are employed due to their effectiveness in image ernel Ridge Regression, and Random Forest are applied. Rando
recognition tasks; they analyze visual features in images to m Forest, with its accuracy and multi-feature handling, predicts s
accurately identify plant diseases [20], [21]. uitable crops based on soil and environmental conditions. The sy
stem helps farmers improve productivity, conserve resources, an
4. Training and Validation: Once the algorithms are
d promote sustainable agriculture.
developed, they undergo rigorous training and validation
processes to ensure their effectiveness. The collected datasets Fertilizer Suggestion System: The system gathers soil data (N,
are divided into training, validation, and test sets; this division P, K, pH, moisture), environmental factors, and crop-specific nut
allows for a comprehensive evaluation of model performance. rient needs. Preprocessing includes cleaning, normalization, and
The training data is used to fit the models while validation data
helps fine-tune them and prevent overfitting—a common issue feature engineering to highlight critical factors like soil type and
where models perform well on training data but poorly on new climate. Machine learning models, including SVM and Random
data. Hyperparameter tuning optimizes the models by adjusting Forest, analyze the data, with Random Forest chosen for its accu
variables such as learning rates and the number of layers in racy and ability to handle complex features. The model provides
CNNs for disease detection; this fine-tuning process is crucial real-time, user-friendly fertilizer recommendations via a web int
for enhancing model performance. Cross-validation ensures that erface, improving crop yields and supporting sustainable practice
the models generalize well to unseen data by testing them on
s.
different subsets of the dataset, increasing their accuracy and
reliability when deployed in real-world scenarios[13],[14]. Plant Disease Detection System: This system uses CNNs, parti
cularly ResNet, trained on thousands of healthy and diseased pla
5. Web-Interface Integration: The final step involves nt images to classify diseases based on visual features like color,
integrating the trained models into a web-based platform using
Flask, a Python web framework that simplifies web application texture, and patterns. ResNet, with identity skip connections, add
development. This integration makes the platform accessible to resses the vanishing gradient problem, improving model accurac
users across various devices, allowing farmers to input critical y and performance. It excels at detecting subtle patterns, enablin
data such as soil properties or upload images of plants for g early and accurate diagnosis of plant diseases, reducing crop lo
disease diagnosis easily. The interface processes this input sses, and ensuring sustainable farming practices. The system em
quickly and provides real-time recommendations on optimal powers farmers with data-driven insights to protect and enhance
crops to plant, appropriate fertilizer usage based on soil
crop health.
conditions, and potential plant diseases that may affect their
crops. VI. ALGORITHM
The user-friendly design ensures that even farmers with limited
technical knowledge can navigate the platform effortlessly. By Machine Learning Algorithms for Crop and Fertilizer
providing instant feedback through actionable insights, Recommendations: Machine learning models are utilized to
AgroVision empowers farmers to make informed decisions that provide personalized crop and fertilizer recommendations.
enhance crop yields, optimize resource use efficiently, and These models are designed to predict suitable crops and
effectively manage plant diseases. fertilizers based on environmental conditions, soil health, and
other user inputs, optimizing farming practices for better yield
Overall, AgroVision combines complex ML and DL models and resource management.
with an intuitive web interface to deliver actionable insights
directly to farmers. By following this methodical approach, Algorithm used for recommendation system Random Forest
AgroVision helps farmers increase productivity while reducing The Random Forest algorithm is a powerful ensemble learning
waste and adopting sustainable farming practices. method that enhances accuracy, reduces overfitting, and handles
both classification and regression tasks effectively.
1. Bootstrap Aggregation (Bagging):
V. SYSTEM ARCHITECTURE  From the dataset D containing n (e.g., soil data,
weather data, crop yields) samples, B bootstrap
datasets D1,D2,…,DB are created.
 Each bootstrap dataset Db is generated by sampling
with replacement from D, ensuring variation in the
training data for individual trees, which reduces
overfitting.

2. Decision Tree Construction:

 Each bootstrap sample Db is used to train a decision
tree Tb .
 Feature Selection at Nodes:
 A random subset of m features is selected from
Fig 2. Architecture of AgroVision the total M features (e.g., soil pH, rainfall,
temperature, nutrient levels).
  The best feature among the m selected features For a layer L, the block output yL combines the residual
is chosen to split the node, ensuring randomness mapping FL (x)with the input x. This hierarchical approach
and reducing correlation among the trees. captures features like edges, textures, and complex patterns
crucial for plant disease identification.

4. Global Average Pooling (GAP)

3. Prediction: Instead of fully connected layers, global average pooling
 Each tree Tb predicts a class label hb (x) for the input reduces spatial dimensions:
instance x.
 The final class prediction H(x) is determined by 1
H W
majority voting across all trees: GAP=
H ×W
∑ ∑ yi , j
H(x)=mode{h1(x),h2(x),…,hB(x)} i=1 j=1
Where:
4. Quality of Splits:  H: Height of the feature map.
 Gini impurity measures the likelihood of incorrect  W: Width of the feature map.
classification if a random instance is labeled based on This operation generates a compact representation for
the distribution at a node: each feature map channel.
C
G=∑ P i (1− P i) 5. Output Layer
i=1 The pooled features are passed to a softmax layer for
Where: C: Number of classes. classification:
Pi: Proportion of samples belonging
to class i. zc
e
P ( c|x )=
5. Out-of-Bag (OOB) Error: ∑ ez k

k
 Each bootstrap dataset leaves out some samples,
known as OOB samples. Where:
 These OOB samples are used to test the trees trained  Zc : Logit (score) for class c.
on their corresponding bootstrap datasets.  P(c∣x): Probability of class c (e.g., healthy or diseased
This provides an unbiased estimate of model performance plant).
without the need for a separate validation set.
6. Loss Function
Deep Learning (CNNs): Convolutional Neural Networks The network is trained using the cross-entropy loss:
C
(CNNs) are used for plant disease detection. CNNs are
particularly effective for image recognition tasks, enabling the
ℒ=− ∑ yc log P(c∨x )
c=1
application to analyze images of plants and identify various Where:
diseases. This helps farmers take timely action to protect their  C: Total number of classes.
crops from potential threats.  Yc : True label (1 if class ccc is correct, else 0).
 P(c∣x): Predicted probability of class c.
1. Input and Convolution ResNet's mathematical foundation ensures its suitability for
The input image X is passed through initial convolutional tasks requiring precise and deep feature learning, such as
layers: identifying and classifying plant diseases.
Z = W ∗X + b
 W: Filter weights. VII. RESULTS AND ANALYSIS
 ∗: Convolution operation.
 b: Bias term. The evaluation of AgroVision involved testing its
The output Z is typically followed by batch recommendations and disease detection capabilities using
normalization and activation (e.g., ReLU). benchmark datasets and real-world scenarios. Several
performance metrics, including accuracy, precision, recall, and
2. Residual Block F1-score, were employed to assess the system’s effectiveness.
In a residual block, the input xxx is combined with a learned
residual mapping F(x,{W}):
y=F(x, {W})+x
Where:
 F(x,{W}): Transformation function (includes
convolution, batch normalization, and activation).
 x: Input to the block (identity mapping).
 y: Output of the block.
This addition allows the network to learn modifications
(residuals) rather than the entire transformation, improving
gradient flow during backpropagation. Fig 2. Comparisonof algorithms
This bar chart compares the accuracy of different algorithms.
3. Stacking Residual Blocks Among the algorithms evaluated, Random Forest (RF)
Multiple residual blocks are stacked, progressively learning demonstrates the highest accuracy.In the context of crop and
hierarchical features: fertilizer recommendation, this indicates that Random Forest is
Y L= FL (x) + x better at capturing complex patterns and relationships in the
data compared to the other algorithms (e.g., Decision Tree,
Naive Bayes, SVM, Logistic Regression).
 [5]
A.Sharma, R. Patel, and K. Tiwari, "Machine Learning for
Model Training Accuracy Epochs
ResNet50 99.2 30
Model for plant disease detection Detection and Prediction of Crop Diseases and Pests: A
Comprehensive Survey," International Journal of Engineering
ResNet-based plant disease detection model achieved an Research & Technology, 2021.
[6]
impressive accuracy of 99.2%.The model employs the ResNet P.Das, A. Singh, and R. Tiwari, "A Survey on Intelligent
Agriculture: Fertilizer Recommendation System," Journal of
architecture for efficient classification of plant diseases by
Intelligent Agriculture, 2021.
leveraging its powerful feature extraction abilities. [7]
V. Sharma, K. Tiwari, and S. Yadav, "Data-Driven Crop Disease
Detection Using Machine Learning Algorithms," Machine Learning
Screenshots: in Agriculture, 2022.
AgroVision Homepage: Introduces AgroVision as an AI [8]
A. Patel, V. Yadav, and S. Verma, "AI-Based Fertilizer and
assistant for farming, focusing on crop, fertilizer, and disease Irrigation Recommendation Systems for Agriculture," AI for
management in a simple interface. Sustainable Agriculture, 2022.
[9]
Pranesh Kulkarni, Atharva Karwande, Tejas Kolhe, “Plant Disease
Detection Using Image Processing and Machine Learning,” in 2022
International Journal of Engineering Research & Technology
(IJERT), 2022.
[10]
Khaki, S., & Wang, L., “Crop Yield Prediction Using Deep Neural
Networks,” in 2022 International Journal of Engineering Research
& Technology (IJERT), 2022.
[11]
Gupta, A., Nagda, D., Nikhare, P., Sandbhor, “Smart Crop
Prediction Using IoT and Machine Learning,” in 2022
Our Services International Journal of Engineering Research & Technology
(IJERT), 2022.
[11]
J. Zhang, L. Li, and T. Zhou, "Crop and Fertilizer
Recommendation System for Precision Agriculture," Computers
and Electronics in Agriculture, 2022.
[12]
R. Agarwal, M. Kumar, and S. Mishra, "Agricultural Disease
Detection with Deep Learning for Sustainable Farming," Journal
of Agricultural Informatics, 2022.
[13]
H. Kumar, S. Gupta, and P. Singh, "Machine Learning Models for
Disease Identification in Agricultural Crops," Agricultural
Plant Disease Detection Module: Analyzes plant images to Systems, 2023.
[14]
detect diseases and offers advice. Christine Musanse, Anthony Vodacek, Alfred Uwitonze, “Data-
Driven Analysis and Machine Learning-Based Crop and Fertilizer
Recommendation System for Revolutionizing Farming Practices,”
in 2023 Journal of IEEE Transactions on Artificial Intelligence,
2023.
[15]
S. Patel, A. Sharma, and R. Kumar, "Precision Agriculture: Crop
Prediction and Fertilization Using Machine Learning,"
International Journal of Precision Agriculture, 2023.
[16]
S. Bhardwaj, A. Gupta, and R. Verma, “Deep Learning for Plant
Plant Disease Module output: Gives output describes Disease Detection Using Convolutional Neural Networks,” in
prevention for Apple Scab, a fungal disease in wet conditions. 2023 International Conference on Sustainable Agriculture
Technologies, 2023.
[17]
A. Rao, M. Tiwari, and N. Sharma, “Automated Plant Disease
Identification Using Transfer Learning Techniques,” in 2023
International Journal of Computational Agriculture, 2023.
[18]
G. Zhang, T. Lin, and K. Zhou, “Application of Vision
Transformers in Plant Disease Detection,” in Journal of Artificial
Intelligence in Agriculture, 2023.
[19]
D. Patel, V. Kumar, and M. Singh, “A Comparative Study of Deep
Learning Techniques for Plant Disease Recognition,” in Journal of
Advances in Agricultural Informatics, 2024.
[20]
H. Wang, Y. Li, and J. Zhao, “ResNet-Based Framework for Leaf
VII.REFERENCES Disease Classification in Real-Time Environments,” in Computers
and Electronics in Agriculture, 2024.
[21]
[1]
Deepa Shetty, Rashmi Naveen, Chinmai Shetty, “A Machine S. Kumar, P. Singh, and R. Roy, “Efficient and Accurate Detection
Learning Technique for Identification of Plant Diseases in Leaves,” of Plant Diseases Using Lightweight CNN Models,” in 2023 IEEE
in 2021 Sixth International Conference on Inventive Computation International Conference on Smart Agriculture Systems, 2023.
[22]
Technologies, 2021. V. Ingale, B. Wankar, K. Jadhav, T. Adedoja, V.K. Borate,
[2]
Aditya Kalekar, Vijay Ganesh, Yogesh Kale, “Weather Forecasting “Healthcare is Being Revolutionized by AI-Powered Solutions
& Crop Recommendation,” in 2021 International Journal of and Technological Integration for Easily Accessible and Efficient
Engineering Research & Technology (IJERT), 2021. Medical Care,” in 2024 15th International Conference on
[3]
Usman Ahmed, Jerry Chun-Wei Lin, Gautam Srivastava, “A Computing, 2024.
[23]
Nutrient Recommendation System for Soil Fertilization Based on S. Sonavane, A. Khade, V.B. Gaikwad, “Novel Approach for
Evolutionary Computation,” in 2021 Computers and Electronics Localization of Indian Car Number Plate Recognition System
in Agriculture, 2021. Using Support Vector Machine,” in International Journal of
[4]
Bendre, M.R., Thool, R.C., Thool, V.R., “Big Data in Precision Advanced Research in Computer Science and Software
Agriculture: Weather Forecasting for Future Farming,” in 2021 The Engineering, 2013.
[24]
International Conference on Next Generation Computing S.M. Sonavane, G.R. Prashantha, P.D. Nikam, A.V.R. Mayuri,
Technologies, 2021. “Optimizing QoS and Security in Agriculture IoT Deployments: A
Bioinspired Q-Learning Model with Customized Shards,” in
*Heliyon*, 2024.