A Methodological Approach Integrating Deep Learning Techniques to Detect and Classify Anthracnose Disease for Sustaining Mango Leaf Health

Mango, a large, cultivated fruit crop in India is the recent research area where research in detecting diseases found in mango leaves, fruits or stems is widely done. This is done to protect it as around $\mathbf{4 0 \%}$ of crop gets destroyed due to pests and diseases. In this paper, study on one of the most popular mango leaf diseases, Anthracnose is done. The paper focuses on detecting the seriousness level of Anthracnose diseased mango leaf. These seriousness levels range from 1 to 10. A hybrid model consisting of fusion of CNN and Random Forest is proposed to extract the diseased part and classify it among seriousness level respectively. The suggested model is also distinguished with other pretrained models such as VGG16, CNN, ResNet50, and RNN to show its outstanding performance in predicting the seriousness level of Anthracnose. The efficiency at each seriousness level of the proposed method is evaluated through precision, recall and F1-score. The results indicated that precision ranges from $\mathbf{9 1. 1 2 \%}$ to $\mathbf{9 7. 8 3 \%}$, recall measures from $\mathbf{9 1. 3 2 \%}$ to $\mathbf{9 9. 2 7 \%}$ and F1-score measures from $92.63 \%$ to $96.64 \%$ when calculated over all the classes. These outcomes exhibit the competence of the proposed CNN+RF model to categorize the seriousness level of anthracnose disease in mango leaf.