Elevating Mango Leaf Disease Classification Utilizing Dense ViT

Abstract

Mango is a significant economic crop, generating billions of dollars annually in global trade, particularly from major producers like India, China, and Thailand. However, mango production faces substantial challenges due to mango leaf diseases such as anthracnose and powdery mildew, which can reduce yields by up to $30 \%$. According to recent statistics, these diseases affect over $25 \%$ of mango orchards worldwide, necessitating increased investment in disease management and resistant cultivars. Image processing has come a long way in assisting in detecting and preventing mango leaf diseases. The objective of this study is the propose a model combining the reliability of DenseNet121 and the embedding methods of Vision Transformer, DenseViT for the classification of Mango Leaf Diseases. The model was trained and tested on a dataset containing 1800 images across 8 classes achieving an accuracy of $98.14 \%$ alongside 0.99,0.995 and 1 for the recall, precision and f1score indicating the robustness of the model. It can easily be deployed in as a real-life application which can help both novice and veteran farmers.