A Hybrid CNN-SVM Model for Multi- Classification of Mango Leaf Disease

Abstract

Mango leaf spot diseases significantly affect the health and yield of mango trees, posing a serious threat to agricultural productivity in tropical regions. Traditional disease detection methods are often labor-intensive, time-consuming, and prone to human error. To address this issue, this research proposes a hybrid deep learning and machine learning approach for the automatic multi-class classification of mango leaf spot diseases. The system leverages a Convolutional Neural Network (CNN) using the pre-trained VGG16 model as a feature extractor, combined with three classical classifiers: Support Vector Machine (SVM), Random Forest (RF), Deep feature extraction (VGG16 CNN), Classical ML classifiers (SVM, RF, KNN, LR, etc.), Multi-class classification and K-Nearest Neighbors (KNN). A comprehensive dataset containing images of eight categories including seven common diseases (Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Powdery Mildew, Sooty Mould) and one healthy class was used for training and validation. The CNN-based feature extraction enabled the effective representation of leaf textures and disease patterns, which were then classified by the individual models. The SVM classifier achieved the highest accuracy among the tested algorithms, outperforming RF and KNN. Evaluation metrics such as accuracy, classification report, and confusion matrix were used to assess performance, while Principal Component Analysis (PCA) and visualizations provided insights into data distribution and class separation. This hybrid model demonstrates strong potential for aiding farmers and agricultural professionals in early and accurate identification of mango leaf diseases, contributing to timely intervention and improved crop management.