Skip to main content
AboutProblemsGuidelinesTerms & conditions

Plant Leaf Disease Detection using Vision Model

Problem Statement​

Plant leaf diseases severely impact plant health and crop yield. Early detection is crucial, but manual inspection is time-consuming, requires expert knowledge, and is prone to errors.

Key challenges addressed:

  • Visual similarity between different leaf diseases.

  • Highly imbalanced class distributions.

  • Large number of disease categories.

  • Need for an automated and scalable image-based diagnosis system.

The goal is to accurately classify plant leaf diseases using only leaf images, without relying on environmental or crop-level information.

Objective​

To develop a deep learning–based vision model that can accurately identify plant leaf diseases from images, enabling early detection and actionable insights for farmers and plant specialists.

Work Done So Far​

  • Developed an EfficientNet-based CNN model (variants B4–B7) for leaf disease classification.

  • Preprocessed images (resize, crop, normalize) for model input.

  • Trained and evaluated models on:

  • Rice Plant Leaf Disease Dataset (19 classes).

  • Potato–Tomato Plant Leaf Disease Dataset (13 classes).

  • Handled class imbalance during training.

  • Selected model based on accuracy and precision metrics.

  • EfficientNet-B4 identified as best balance between performance and computational efficiency.

System Flow​

Plant Leaf Image → Image Preprocessing (Resize, Crop, Normalize) → EfficientNet CNN Model (B4 / B5 / B6 / B7) → Plant Leaf Disease Prediction (Accuracy, Precision)

Key Features​

  • Multi-class classification for plant leaf diseases.

  • Handles imbalanced datasets.

  • Works across multiple crops (rice, potato, tomato).

  • Provides actionable disease identification to users.

  • Supports scalable automated detection.

Technologies Used​

  • Frameworks: PyTorch, TIMM, TorchVision

  • Models: EfficientNet (B4–B7 variants)

  • Datasets: Rice Plant Leaf Disease (19 classes), Potato–Tomato Plant Leaf Disease (13 classes)

  • Training Techniques: Transfer learning, class imbalance handling, precision-accuracy-based model selection

Future Scopes​

  • Expand Dataset: Include more plant species, field images, and additional diseases.

  • Improve Model Robustness: Handle variations in lighting, image quality, and leaf orientation.

  • Multi-Modal Inputs: Combine leaf images with environmental data (temperature, humidity) to improve predictions.

  • Continuous Model Updates: Retrain periodically with new data to track evolving disease patterns.

  • Edge Deployment: Optimize model for mobile and IoT devices.

  • Integration with Advisory Systems: Connect predictions with automated notifications and guidance for farmers.

  • Explore Other Models: Experiment with different architectures such as ResNet, DenseNet, Vision Transformers, or Vision-Language Models (VLMs) to further improve performance and accuracy.

  • Ensemble Approaches: Combine predictions from multiple models to boost overall robustness and reduce misclassification.

  • Explainable AI (XAI): Implement techniques like Grad-CAM or attention maps to provide visual explanations of model decisions, helping farmers understand predictions.

  • Active Learning & Semi-Supervised Learning: Use unlabelled or partially labelled data to continuously improve model performance without extensive manual annotation.

Impact​

This vision model enables fast, accurate, and scalable plant disease detection directly from leaf images. It helps farmers and plant specialists take timely action, improves crop management, and reduces losses due to undetected diseases.