LEAF DISEASE DETECTION USING MACHINE LEARNING

This paper presents a hybrid approach integrating Machine Learning (ML) and Convolutional Neural Networks (CNN) for automated tobacco leaf disease detection and classification. The proposed system combines handcrafted feature extraction methods such as HSV color histograms, Local Binary Patterns (LBP), Histogram of Oriented Gradients (HOG), and shape descriptors with deep learning architectures like ResNet50, VGG16, EfficientNetB0, and MobileNetV2. A hybrid ResNet50 model integrating CNN deep features with ensemble ML classifiers achieved an accuracy of 97%. The proposed method demonstrates superior efficiency, interpretability, and real-time deployability via a Streamlit web interface. Tobacco is a critical cash crop in India, yet its yield is severely compromised by various leaf diseases. Traditional diagnosis is often manual, slow, and lacks scalability. This paper introduces a robust machine learning framework for the automated detection and classification of four major tobacco leaf conditions: Frog Eye Leaf Spot, Powdery Mildew, Tobacco Mosaic Virus (TMV), and Healthy Leaf. The system employs a sophisticated multi-feature engineering pipeline combining color (HSV), texture (LBP), shape, and gradient (HOG) features, resulting in a high- dimensional feature vector (8,644 features) per image. This vector is fed into an Ensemble Voting Classifier (Random Forest, SVC, and Logistic Regression). Through rigorous data augmentation (59 original images expanded to 320 samples) and optimization, the system achieves a professional-grade classification accuracy of 90.6%, significantly outperforming a baseline model. The final model is deployed via a Streamlit web application for real-time diagnosis and comprehensive management recommendations.