PROJECT
EXHIBITION
DSN2098 - 1 Group
130
“KRISHI CARE ”
AI-Based Plant Disease
Detection System
Using Image
Classification and
Fertilizer
TEAM MEMBERS
Om Kumar Aadish Roshmik Harshit Tiwari
Abhyanand
Singh
23BAI100 23BAI113
Chaturvedi 23BAI100
Agrawal 23BAI104 Sharma23BAI110
76 67 14 35 47
Supervised by : Dr. M K Jayanthi
Kannan
INTRODUCTION
Agriculture is a cornerstone of global food security and economic development, yet it faces
persistent challenges from plant diseases that significantly reduce crop yields and increase
production costs. These diseases pose a severe threat to farmers, especially smallholders in
rural and underserved areas, where traditional methods of disease identification are often
inaccessible, time-consuming, and reliant on expert intervention. Delayed diagnosis and
incorrect treatment further exacerbate the problem, resulting in substantial economic losses
and diminished agricultural productivity. Advancements in artificial intelligence (AI) and
machine learning (ML) have revolutionized the potential for addressing these challenges. AI-
powered systems that integrate image classification with actionable insights provide scalable
and efficient solutions for early disease detection and management. With the growing adoption
of smartphones and digital tools, such technologies are becoming increasingly accessible, even
in remote regions. This paper introduces Krishi Care, an integrated system that combines deep
learning for plant disease detection with a recommendation engine offering tailored advice on
fertilizers, pesticides, and preventive measures. The inclusion of real-time weather guidance
ensures context-aware recommendations, making the system practical and effective. By
blending cutting-edge technology with a user-friendly interface, Krishi Care seeks to empower
Existing Work
and Literature
Review
Different Research Papers Explored by Respective Team
Members
OM KUMAR ( Objective ) ( Efficiency )
SINGH ( Methodology
Used ) • VGG16 achieved high accuracy for
23BAI10076 1) Automate Plant Disease 3)Leverage Advanced AI for 1) Image Acquisition:
Detection Accuracy • High-quality images are crops like tomatoes and cucumbers.
Develop an AI-based Use state-of-the-art captured using cameras • EfficientNet reached 96% accuracy
Title system using CNNs to deep learning models like and IoT devices.
for cucumber disease detection.
:
Revolutionizing
automatically detect VGG16 and EfficientNet 2) Dataset:
• YOLOv5 performed with 75.64%
and classify diseases in for precise classification • The PlantVillage dataset
Agriculture with crops like tomato, and disease is used, containing accuracy for chili disease detection in
4)Real-Time Monitoring with
Artificial potato, chili, and identification. 54,306 images of both
2) Improve Agricultural IoT Integration real-time field conditions.
Intelligence: Plant cucumber by analyzing healthy and diseased
Productivity
leaf images. Incorporate IoT devices leaves from various plant
Disease Detection (sensors, drones) for real-
Enable early disease species (e.g., tomato,
Methods, detection to help farmers time monitoring of crops in potato, chili).
Applications, and take timely action, the field, making the 3) Preprocessing: (ISSUES)
system scalable and
journal
Their :
Limitations reducing crop losses and
practical for large
• Techniques like color
1) Environmental Variability:
improving yield. space transformations
Model performance can be affected by changes
Frontiers in agricultural applications. and noise removal are
in lighting, moisture, and other environmental
Plant Science ( Technology Used ) applied to standardize
factors.
Year • Artificial Intelligence (AI):
and clean the images.
4) Segmentation:
2) Data Quality:
:
13 March • Machine Learning (ML): Models such as SVM, • Leaves are segmented to
The presence of noise and extraneous elements
like soil and grass in the images can complicate
D
2024 Random Forest, and Artificial Neural Networks
separate diseased areas
using methods like
disease detection.
OI
https://doi.org/ (ANN). thresholding and
3) Real-Time Deployment:
:
10.3389/fpls.20 • Deep Learning (DL): clustering.
Challenges arise in deploying the models in
real-time due to hardware limitations and the
24.1356260 ⚬ Convolutional Neural Networks (CNN), 5) Feature Extraction:
• Features like color, need for IoT integration for field use.
UR including architectures like VGG16, texture, and morphology 4) Multiple Diseases:
L:
https://www.frontier EfficientNet, and YOLOv5 for real-time are extracted for Identifying overlapping or co-occurring diseases
sin.org/journals/pla in a single leaf remains a challenge.
object detection. classification.
nt-science/articles/1 5) Dataset Limitations:
6) Classification
0.3389/fpls.2024.13 • IoT (Internet of Things): Integrated with AI to Dataset imbalance and the use of images from
56260/full • CNN models such as
monitor real-time plant health using sensors VGG16, YOLOv5, and controlled environments may limit
AADISH ( Objective ) ( Efficiency )
CHATURVEDI ( Methodology
Used )
23BAI11367 1) Automated Disease 3) Global Food Security:
• Image Preprocessing: Images
• The best model achieved an accuracy of 99.35%
Detection: Contribute to mitigating on the PlantVillage dataset.
Develop a system that global food security were resized to 256x256 pixels • GoogLeNet consistently outperformed AlexNet, and
Title automatically detects
threats by providing a
tool that can diagnose
and converted into different transfer learning yielded better results than
formats (color, grayscale,
: Deep
Using
plant diseases using diseases early,
segmented).
training from scratch.
image-based particularly aiding • Performance remained high even when using only
Learning for recognition techniques, smallholder farmers who • Deep Learning Approach: Two 20% of the data for training (98.21% accuracy),
reducing the need for rely heavily on healthy CNN architectures, AlexNet and demonstrating the robustness of the approach.
Image-Based 4)Smartphone Integration:
crops.
manual
2) diagnostics.
Leveraging Deep GoogLeNet, were tested with • However, performance dropped when the system
Plant Disease Explore the potential for
Learning: various configurations, including was tested on images from external sources
integrating the system
Detection Utilize advanced deep training from scratch and (around 31% accuracy).
with smartphones,
learning models, transfer learning.
making real-time disease
journal : specifically
detection accessible
• Training and Evaluation: Models
convolutional neural were trained on different train-
networks (CNNs), to globally through mobile
Frontiers in test splits (80-20, 60-40, etc.),
Plant Science
identify plant diseases
with high accuracy.
devices.
and performance was measured (ISSUES)
( Technology Used ) using accuracy, precision, recall,
and F1 score. • Generalization: The models performed well on
• Hyperparameters: Stochastic controlled datasets but had reduced accuracy on
Year • Deep Learning Models: Convolutional Neural Networks
gradient descent with learning real-world images from external sources.
: SEP 2016
22
(CNNs) were used, specifically AlexNet and GoogLeNet
architectures.
rate adjustments was used, and • Overfitting: While overfitting was minimized, more
experiments ran for 30 epochs diverse and larger datasets are needed for the
D • Transfer Learning: Some experiments applied transfer
each. model to generalize better to unseen
OI learning, where pre-trained models on large image
environments.
:
https://doi.org/10 datasets (ImageNet) were adapted to the task of plant
• Practicality: The approach is limited to images of
.3389/fpls.2016.0 disease detection.
individual leaves on a homogeneous background,
1419
UR • Dataset : The PlantVillage dataset consisting of 54,306
which is not always feasible in real-world
images of plant leaves, representing 14 crops and 26
L:
https://www.frontiersin.org
/journals/plant-science/arti different diseases.
applications.
cles/10.3389/fpls.2016.014
19/full
ROSHMIK ( Objective ) ( Efficiency )
AGRAWAL ( Methodology
Used ) 1) Performance:
23BAI10014 1) Review of Deep Learning 3) Analyze Current Deep learning models outperform traditional image
Applications in Plant Techniques: processing methods, providing higher accuracy,
1) Classification Network:
Disease and Pest Detection Explore classification, robustness, and better feature extraction capabilities.
• Used to classify images of
Title Summarize deep detection, and
plant diseases and pests,
2) Detection Accuracy:
:
Plant diseases
learning advancements segmentation networks
leveraging CNNs for image
Deep learning-based methods, like Faster R-CNN and
in plant disease and used in detection.. YOLO, have shown high mean Average Precision (mAP)
feature extraction.
and pests pest detection. for plant disease and pest detection, with some
2) Comparison with 4) Identify Challenges and models achieving up to 99.64% accuracy in specific
detection Traditional Detection Future Trends: 2) Detection Network:
scenarios (e.g., tomato disease detection).
based Methods: Discuss current • Involves detecting specific
3) Speed :
Highlight advantages challenges and propose areas of the image where the
on deep One-stage detection networks (like YOLO) offer faster
of deep learning over future research diseases or pests are located. inference speeds compared to two-stage networks
learning
journal : traditional techniques. directions for practical Methods include Faster R-CNN (like Faster R-CNN), making them more suitable for
applications. and YOLO for object detection.
Liu and Wang real-time applications.
Plant Methods 3) Segmentation Network (ISSUES)
( Technology Used ) • Utilizes networks like FCN
(Fully Convolutional Networks)
1) Small Dataset Size:
Deep learning methods rely on large datasets, but
and Mask R-CNN to separate
1)Deep Learning: many plant disease datasets are limited in size, which
Year The review focuses on deep learning technologies,
diseased areas from healthy
can hinder model performance.
: FEB 2021
24 particularly Convolutional Neural Networks (CNN).
plant regions.
2) Early Detection Challenges:
D 2) Image Recognition: 4) Datasets:
Identifying diseases in the early stages is
difficult due to small lesion size, and
OI
https://doi.org/10.118
6/s13007-021-00722-9
Digital image processing is the primary means of • Analysis includes using
background noise complicates image analysis.
disease detection.
: datasets such as PlantVillage
3) Lighting and Occlusion Problems:
3) Open Source Tools: and other self-collected or
Natural variations in lighting and plant
UR The paper mentions deep learning frameworks like publicly available datasets to
train and evaluate models.
occlusions can negatively affect detection
L:
https://plantmethods.bio
medcentral.com/articles/1
TensorFlow, PyTorch, and Caffe. accuracy.
0.1186/s13007-021-00722
4) Common CNN Models: 4) Complexity of Models:
-9 Popular networks such as AlexNet, GoogleLeNet, Some models require significant computational
VGGNet, ResNet, and Inception V4 were discussed for resources, making them difficult to deploy on
their roles . mobile or low-power devices.
HARSHIT ( Efficiency )
TIWARI ( Objective ) ( Methodology 1) Improved Accuracy:
1) Enhance Crop Disease Used ) The CNN-based model shows high accuracy in
23BAI10435 Detection Accuracy: 3) Automate Disease 1) Need for Large, High-Quality
detecting multiple crop diseases, with
Detection Process: Datasets:
• Improve the improvements of up to 92% in classification
Develop a system that • Leaf images of various
accuracy of performance.
Title identifying plant automates the detection crops are collected,
2)Reduced Training Time:
:
Crop Disease diseases from leaf and classification of crop cleaned, and preprocessed
using resizing,
By using transfer learning, the system
images using deep diseases to aid farmers significantly reduces the training time compared
Detection with early intervention.. normalization, and
learning algorithms. to traditional methods.
4)Ensure Scalability and augmentation techniques
Using Deep Versatility:
3)Enhanced Robustness through Data Augmentation:
2) Reduce Processing to increase dataset
dependency edges. Data augmentation techniques help increase the
Learning Time:: Implement a model diversity.
model's generalizability across different
Techniques Employ techniques that capable of handling
environmental conditions and lighting.
allow rapid analysis of different crops and 2) Overfitting with Small Datasets:
.journal : diseases with high • A deep learning model
4) Fast and Scalable::
large datasets to speed The model can be scaled to detect diseases
up the disease adaptability. (CNN) is trained on labeled
International across various crops with minimal changes,
detection process.. leaf images to classify
Journal of making it adaptable and easy to deploy.
diseases based on visual
Engineering
( Technology Used ) symptoms. 1) Need for Large, High-Quality Datasets:
Research & The system requires a large number of labeled
2) Need for Large, High-
Technology 2) Overfitting with Small 3) Dependence on Preprocessing images for each disease to maintain high
Quality Datasets:
(IJERT) Datasets: Quality: accuracy, which may not always be available.
CNNs are used to
Year automatically extract
Models like ResNet and • Pre-trained models such as
ResNet and VGG16 are fine-
2) Overfitting with Small Datasets:
:25 Sep 2021 features from images,
VGG16 are employed
using pre-trained tuned on the dataset to
Overfitting can occur, particularly when the
available data is limited or lacks variability,
providing an accurate
weights to boost leverage existing reducing the model's generalization capacity.
diagnosis of the crop
D disease.
accuracy while
minimizing
2) training time
Limited Performance on
knowledge and improve
detection performance.
3) Dependence on Preprocessing Quality:
2) Dependence on
OI
10.17577/IJERT Preprocessing Quality: Unseen Crops/Diseases:
The accuracy of the model is heavily reliant on
the preprocessing techniques applied to the
:
V9IS060591 Techniques like resizing, The model is 4) Evaluation Using Accuracy and input images (like normalization and
normalization, and data implemented using F1 Scores: augmentation).
augmentation (rotation, Python, with TensorFlow • The model's performance is 4) Limited Performance on Unseen Crops/Diseases:
flipping) are used to and Keras as the deep evaluated using standard The system's performance may drop when
improve the robustness learning frameworks. metrics like accuracy, exposed to new crop varieties or diseases that
PROPOSED WORK
The proposed model for plant disease detection is built around a custom Convolutional Neural
Network (CNN) architecture, tailored for efficient image classification into various disease
categories. Instead of relying on pre-trained architectures like MobileNetV2 or ResNet50, the
model is lightweight and specifically designed for this task. It processes input images of size
256×256×3 through convolutional layers with ReLU activation for feature extraction,
MaxPooling layers for dimensionality reduction, a Flatten layer to convert 2D feature maps into
a 1D vector, and fully connected Dense layers for classification into categories such as
"Healthy," "Powdery," and "Rust." The model concludes with a softmax activation function for
multi-class classification.The model is compiled using the Adam optimizer and categorical
cross-entropy loss, ensuring compatibility with multi-class classification tasks. Training involves
preprocessed datasets and validation to monitor performance, optimizing for high accuracy and
low loss. A user-friendly interface, developed using Streamlit, enhances accessibility, allowing
users to upload plant images and receive disease predictions seamlessly.This system is
straightforward and effective, balancing simplicity with functionality. Its modularity and
scalability make it adaptable for similar image classification tasks. The integration of Streamlit
ensures usability for non-technical users, particularly in agricultural settings where real-time
METHODOLOGY
Krishi Care is developed through a structured methodology, starting with defining
objectives and understanding user needs, such as real-time feedback and ease of
use. A diverse dataset of plant images, categorized by disease types, is collected and
preprocessed through resizing, normalization, and augmentation for consistency and
robustness. A custom Convolutional Neural Network (CNN) is designed with layers for
feature extraction, dimensionality reduction, and classification, optimized using the
Adam optimizer and categorical cross-entropy loss.The model is trained and validated
on a split dataset, with performance metrics monitored to refine its accuracy. It is
deployed using a Streamlit application, providing a user-friendly interface for
uploading plant images and receiving disease predictions with treatment
recommendations. The system is rigorously tested with unseen data to ensure
reliability and hosted on a cloud platform for scalability and accessibility. Regular
updates to the model and dataset enhance accuracy and allow adaptation to new
disease categories, delivering an efficient and scalable solution.
NOVELTY OF THE PROJECT
⚬ The plant disease detection system introduces a unique blend of advanced deep learning
techniques and user-friendly design, making it an innovative solution in agricultural disease
management. At its core, the system employs a custom Convolutional Neural Network
(CNN) optimized for classifying plant diseases with over 96% accuracy and providing real-
time predictions within one second. Beyond detection, the system features a
recommendation engine that offers tailored solutions, such as treatment methods and
fertilizer suggestions, enhancing its practical utility for farmers and agricultural
professionals.
⚬ What sets this project apart is its deployment through Streamlit, which delivers an
interactive and accessible interface, enabling even non-technical users to seamlessly upload
images and receive insights. The inclusion of visual tools, such as disease distribution
graphs and textual explanations, further enhances its usability and educational value.
⚬ The modular architecture supports scalability, allowing the integration of new disease
categories or IoT devices like soil sensors. Combined with robust security measures, this
system offers a scalable, secure, and user-friendly approach to improving crop health and
productivity.
HARDWARE AND SOFTWARE REQUIREMENT
Hardware Requirements:
• Processor: Multi-core processor with a clock speed of at least 2.5 GHz (e.g., Intel i5/i7 or AMD Ryzen
5/7).
• RAM: Minimum 8 GB (16 GB recommended for optimal performance).
• Storage: At least 256 GB SSD for fast data access and storage of models/datasets.
• GPU: NVIDIA CUDA-enabled GPU (e.g., NVIDIA GTX 1050 or higher) for training and inference
acceleration.
Software Requirements:
• Operating System: Windows 10/11, macOS, or Linux.
• Programming Language: Python 3.x.
• Libraries and Frameworks:
⚬ TensorFlow and Keras for CNN implementation.
⚬ NumPy for numerical operations.
⚬ Pillow for image preprocessing.
⚬ Streamlit for the user interface.
⚬ Plotly for visualizing model performance.
• Integrated Development Environment (IDE): Jupyter Notebook or PyCharm for development and
SYSTEM ARCHITECTURE
DIAGRAM
Module Description
1. User Interaction Module
Purpose: Enables users to upload images of plant leaves for disease analysis.
Key Features:
·Image upload functionality supporting common formats (.jpg, .png).
·Error handling for invalid or corrupted file formats.
Responsive and user-friendly design for easy access via a web interface built with Streamlit.
2. Disease Detection Module
Purpose: Processes the uploaded image and classifies plant diseases using the trained
CNN model.
Key Features:
·Model Inference: Utilizes MobileNetV2, a lightweight and efficient deep learning model,
for real-time predictions.
·Data Preprocessing: Normalizes and augments images before feeding them into the
model to enhance prediction accuracy.
·Accuracy: Delivers high confidence predictions for diseases based on the PlantVillage
3. Results Display Module
Purpose: Provides users with the analysis results in a clear and interpretable format.
Key Features:
·Displays the predicted disease name along with a confidence score.
·Includes additional actionable information, such as links to recommendations and preventive
measures for the diagnosed disease.
·Supports real-time updates to ensure smooth and quick feedback to the user.
4. Recommendation Module
Purpose: Provides disease-specific remedies to assist farmers in managing plant health.
Key Features:
·Offers recommendations on fertilizers, pesticides, and preventive actions tailored to the
identified disease.
·Integrates real-time weather data for context-aware advice (e.g., optimal application timing
based on local conditions).
5. Feedback Collection Module
Purpose: Captures user feedback to improve system performance and usability.
Key Features:
·Allows users to report incorrect predictions or provide general suggestions.
·Collects data for model retraining and feature enhancements.
6. Backend Integration Module
Purpose: Ensures efficient processing and model deployment for real-time predictions.
Key Features:
·Powered by TensorFlow for model inference.
·Supports GPU acceleration for faster computations.
·Optimized to handle concurrent requests with minimal latency.
7. Software Stack
Programming Language: Python
Frameworks: TensorFlow, Streamlit
Libraries: NumPy, Pillow
8. Steps for Implementation
a) Data Preprocessing: Normalize and augment input images to improve model generalization.
b) Model Training: Train the MobileNetV2 model on GPU using a 70-30 train-test split.
c) Web Integration: Deploy the trained model using Streamlit for real-time access to predictions and
recommendations.
By implementing these functional modules, the system delivers a comprehensive and accessible
solution for plant disease detection and management. Each module is designed to enhance user
experience and ensure accurate, actionable insights
Module Workflow
“IMPLEMENTATION
AND
CODING ”
“CROP CARE ”
AI-Based Plant Disease Detection System Using Image
Classification
PROBLEM STATEMENT
Plant diseases significantly impact agricultural productivity, leading to reduced yields
and economic losses. Early and accurate detection of diseases is crucial for effective
intervention and mitigation of damage. However, traditional manual inspection
methods are time-consuming, subjective, and error-prone. Additionally, many farmers
lack access to timely and reliable disease detection systems.
To address this, the proposed solution is an AI-based plant disease detection system
that utilizes image classification techniques to identify plant health issues from images
of leaves. By developing a web-based application that integrates deep learning models
(such as Convolutional Neural Networks), the system can analyze plant images and
provide disease predictions in real-time. This solution aims to assist farmers and
agricultural professionals in making informed decisions, leading to healthier crops,
improved yield, and minimized losses.
Objective
Automated Plant
s User-Friendly Minimization of
Disease Detection Interface Economic Losses
Provide a simple and Support the agricultural
Develop an AI-based system intuitive web application sector by helping to reduce
capable of automatically using Streamlit, allowing the economic impact of
detecting plant diseases users (farmers, plant diseases, improving
from leaf images using deep agriculturists) to upload both yield quality and
learning models, eliminating plant images and receive quantity.
the need for manual quick, actionable disease
inspection diagnoses without needing
technical expertise.
Early Detection and Real-Time Disease
Scalable Solution
Prevention Classification
Create a system that can be Facilitate early detection of
scaled for use with different diseases, allowing for timely Implement a Convolutional
plant species and diseases intervention to reduce crop Neural Network (CNN) to
by leveraging transfer loss and improve accurately classify various
learning techniques and agricultural productivity. plant diseases in real-time,
adding more plant data to enabling immediate
the model. diagnosis and action.
Real-Life Scope Of the
Agricultural
Project
Reduction in Accessibility for Small-
Monitoring Pesticide Use Scale Farmers
Helps apply treatments Provides easy access to AI-driven
Integrates AI into precision more accurately, minimizing disease detection via mobile apps,
farming for real-time crop unnecessary pesticide use empowering farmers with limited
health monitoring, reducing and lowering costs. technical skills
manual effort and
enhancing protection.
Increased Crop Integration with IoT Contribution to
Yield and Quality Devices Research and Data
The AI system can be
Early disease detection deployed in conjunction
Collection
he application can also be used as a
allows quick action, with IoT-based smart research tool, collecting data on
preventing large-scale crop farming solutions, such as disease outbreaks, patterns, and
loss and boosting cameras or sensors in the trends across regions, helping
productivity. field that continuously researchers and governments track
monitor crop health. the spread of diseases and
implement better control measures.
Farmers and Government and Agricultural
Growers Agricultural Consultants and
Agencies Agronomists
Agri-tech Target Research
Companies Consumers Institutions and
Universities
Fertilizer and Greenhouses and Large-scale Farms
Indoor Farming and Agribusiness
Pesticide Operators Corporations
Companies
COMPONENTS of project
All
Set!! Data Collection
Deployment
Backend API for
Development Model
Inference Machine
Learning
Frontend
Model
Usability / Application
Field Use by Precision IoT Integration
Farmers Agriculture
Farmers can easily upload The AI-driven system can be The system can be linked to
images of plants through a integrated with drones or Internet of Things (IoT) devices like
mobile or web app to sensors to monitor large smart cameras and field sensors for
diagnose diseases. This fields, identifying disease automated, real-time monitoring,
helps them make informed outbreaks early, thus reducing the need for human
decisions on treatment, enabling more effective intervention.
ensuring timely action. resource allocation.
Agricultural Advisory Research and Data
Services Collection
Consultants and agronomists can use the he application can also be used as a
system to provide remote diagnosis and research tool, collecting data on disease
actionable advice to farmers, especially in outbreaks, patterns, and trends across
regions where manual inspection is regions, helping researchers and
impractical. governments track the spread of diseases
and implement better control measures.
“RESULT INPUT”
“RESULT OUTPUT ”
CONCLUSION
In conclusion, the proposed system offers a comprehensive solution to the challenges of
plant disease detection and management by integrating cutting-edge AI technologies with
a user-centric design. By utilizing image-based diagnostics, expert recommendations, and
weather insights, the system empowers farmers to make data-driven decisions that
enhance crop health and productivity. This multi-faceted approach allows for early disease
detection, timely intervention, and better resource management, which ultimately
improves agricultural sustainability. Looking ahead, the system’s future development will
focus on expanding the disease dataset to cover a broader range of plant diseases and
incorporate soil analysis to provide more accurate and actionable insights. Additionally,
IoT integration will enable real-time monitoring of environmental factors, further
enhancing decision-making capabilities. To ensure accessibility for a wider audience,
multilingual support will be introduced, helping farmers from different linguistic
backgrounds benefit from the system. These advancements will strengthen the system’s
impact, offering farmers worldwide an effective and scalable solution to combat plant
Thank
you