Leaf Disease Detection Using Deep
Learning
Carried out by:
Guided by:
K.Jaideep
Mr. Saidulu Sir
Priya kumari
Tejaswi
Roshan
ABSTRACT
● Plant diseases significantly affect global crop yields.
● Traditional detection methods are manual and costly.
● CNNs enable fast, accurate leaf disease detection using
images.
● Our model is lightweight and works offline on mobile
devices.
● This approach empowers farmers and promotes
sustainable agriculture.
INTRODUCTION –
Importance of Agriculture
● Agriculture is vital to the economy and food supply.
● Leaf diseases can drastically reduce crop yields.
● Early disease detection is crucial.
INTRODUCTION – Traditional Detection
Challenges
● Manual methods are slow and expert-dependent.
● Lab testing is expensive and not scalable.
● Rural farmers lack access to proper tools.
INTRODUCTION – Why Deep Learning?
● CNNs automate and speed up detection.
● Highly accurate with minimal human input.
● Can be used on basic smartphones.
WHAT IS OUR PROJECT?
● A deep learning-based system for leaf
disease detection.
● Uses image inputs to identify diseases using
CNNs.
● Runs on mobile phones without internet.
● Supports farmers in early detection.
MOTIVATION BEHIND THE PROJECT
● Manual inspection is unreliable.
● Expert consultations and lab tests are costly.
● Widespread use of mobile devices creates
opportunities.
● Bridges the gap between technology and
farming.
ADVANTAGES
● Cost-effective and runs on low-end devices.
● Fast, real-time, and offline capabilities.
● User-friendly interface for non-tech users.
● Scalable and environment-friendly.
.Easy to use by the farmers
PROBLEM DEFINITION
● Most existing systems need high-end
hardware and internet.
● Poor generalization to real-world farm
conditions.
● Complex interfaces are hard for rural farmers.
● Our solution is lightweight, offline, and user-
centric.
LITERATURE SURVEY – Thesis 1
● Title: Leaf Disease Detection Using Deep Learning
● Date of Publication: 15-01-2025
● Published By: Nottingham Trent University, US
● Objective: Detect and classify plant leaf diseases using CNNs.
● Techniques Used: CNN, data augmentation, Softmax classifier
● Results: Achieved over 95% accuracy on PlantVillage dataset.
● Limitations: No field validation; tested only on lab-clean
images.
LITERATURE SURVEY – Thesis 2
● Title: Efficient Plant Leaf Recognition via CNN
● Date of Publication: 10-09-2024
● Published By: MIT AI Research Group
● Objective: Develop a fast CNN model for disease detection.
● Techniques Used: Transfer learning, ReLU, Adam optimizer
● Results: High precision in detecting multiple leaf diseases.
● Limitations: Hardware-intensive; limited device
deployment.
LITERATURE SURVEY – Thesis 3
● Title: Agricultural Disease Prediction Using AI
● Date of Publication: 02-06-2023
● Published By: IIT Delhi, India
● Objective: Use AI to forecast and classify diseases.
● Techniques Used: Hybrid CNN-LSTM model
● Results: Effective prediction model with 93% accuracy.
● Limitations: Model complexity affects mobile
deployment.
LITERATURE SURVEY – Thesis 4
● Title: Smart Farming Using CNN and IoT
● Date of Publication: 20-12-2023
● Published By: Stanford University
● Objective: Integrate CNN detection with IoT sensors.
● Techniques Used: CNN, cloud computing, mobile
interfacing
● Results: Improved monitoring and response time.
● Limitations: Relies on stable network and cloud setup.
LITERATURE SURVEY – Thesis 5
● Title: Real-Time Disease Detection on Android
● Date of Publication: 01-08-2022
● Published By: Google AI India
● Objective: Deploy plant disease detection on Android.
● Techniques Used: TFLite, CNN pruning
● Results: Fast and lightweight mobile app.
● Limitations: Limited to Android OS, lacks iOS support.
LITERATURE SURVEY – Thesis 6
● Title: Lightweight CNN for Field Deployment
● Date of Publication: 11-03-2024
● Published By: University of Toronto
● Objective: Develop a CNN for real-world farm use.
● Techniques Used: Quantization, edge inference
● Results: Runs effectively on low-power devices.
● Limitations: Slight accuracy tradeoff for speed.
EXISTING SYSTEMS vs PROPOSED SYSTEM
● Existing: Requires high-end hardware
Proposed : Runs on low-end smartphones
● Existing: Needs internet for inference
Proposed : Works offline, suitable for rural areas
● Existing: Trained on clean lab datasets
Proposed : Trained on real-world farm images
● Existing: Complex UI
Proposed : Farmer-friendly, simple UI
● Existing: No model update capability
Proposed : Supports feedback and updates
EXISTING SYSTEMS vs PROPOSED SYSTEM
● Existing: Heavy models, online-only, lab datasets.
● Ours: Lightweight, offline, user-friendly, field-tested.
BLOCK DIAGRAM OVERVIEW
1.Input symbols
2.Preprocessing
3.CNN Model
4.Classification
5.Output result
TECHNIQUES USED
● Convolutional Neural Networks (CNNs)
● Transfer Learning with pre-trained models
● Adam Optimizer and Softmax Activation
● Categorical Crossentropy loss function
SOFTWARE & TOOLS USED
● Python, TensorFlow, Keras, OpenCV
● Google Colab, Jupyter Notebook
● TensorFlow Lite for mobile deployment
● Android Studio for app development
SOURCE CODE OVERVIEW
● Streamlit UI for image upload and prediction.
● Model loads .h5 file and classifies input images.
● Mobile deployment uses TensorFlow Lite.
RESULT & ANALYSIS
● Achieved over 95% classification accuracy using CNN.
● Transfer learning improved training efficiency and reduced
overfitting.
● Model performed well on lab and real-world leaf images.
● Confusion matrix shows high precision across disease classes.
● Fast inference: <1 second on mobile devices.
● Data augmentation enhanced model robustness.
● Reliable even under varying lighting conditions.
● Effective for real-time farm usage.
● WORKING FLOWCHART
REAL-WORLD BENEFITS
● Instant decision-making support for farmers.
● Minimized pesticide use through early
detection.
● Works in low-resource and rural areas.
CONCLUSION
● CNN-based app is effective and practical.
● Mobile deployment is cost-efficient and scalable.
● Enables precision agriculture and smart farming.
FUTURE SCOPE
● Support for more plant types and disease
classes.
● Real-time drone integration.
● User-feedback-based model improvement.
● Multilingual UI for global deployment.
REFERENCES
● Mohanty et al., 2016 - Deep Learning for
Plant Disease
● Ferentinos, 2018 - Deep Learning Models for
Diagnosis
● TensorFlow, OpenCV, Google Colab
documentation
Thank you!!
