Even in the age of digitalisation and capitalism, agriculture still plays a significant role in many economies, such as in certain Asian countries where mangoes have become an important export commodity. However, plant diseases put serious constraints on both productivity and quality. Existing methods for identifying disease typically rely on the experience of farmers and are time-consuming and error-prone. In this study, we propose a new hybrid framework consisting of a custom Convolutional Neural Network (CNN) and a Support Vector Machine (SVM) classifier to classify eight mango leaf conditions: Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Powdery Mildew, Healthy, and Sooty Mould. The model uses a dataset of 4000 images collected from mango orchards throughout Bangladesh and incorporates rigorous pre-processing and data augmentation to help improve model robustness and generalisability. The results indicate that the hybrid CNN-SVM model performs best, outperforming state-of-the-art models with an accuracy of 99.75%. The research thus emphasises the role of deep learning and machine learning in enabling more accurate disease detection in agriculture, benefitting farms and the environment via sustainable practices and higher crop yields. KEYWORDS Classification, custom CNN model, deep learning, disease detection, feature extraction, machine learning
Salma et al. (Wed,) studied this question.