Paddy Leaf Disease Classification Using Kullback Leiberg Divergence Loss Based Convolutional Neural Network

Paddy disease significantly reduce both the yield and quality of the crops, making early and accurate identification of the disease crucial for efficient prevention and control. However, achieving a balance between lightweight models and the need for precise classification of paddy diseases is challenging. This is primarily due to the dispersed nature of disease symptoms and the interference of noisy backgrounds, which complicates the detection process. To overcome this problem, this research proposes the Kullback Leiberg Divergence Loss-based Convolutional Neural Network (KLDCNN) approach for the classification of paddy diseases. The KLD improves the CNN's capability to differentiate between predicted probability and true distribution of the classes, resulting in better accuracy. In the segmentation process, a High-Resolution Network (HRNet) is used to localize the features within the image through maintaining the image representation throughout the network. The KLD-CNN approach is assessed by utilizing various performance metrics like accuracy, precision, recall and F1-score based on the Paddy Doctor and Rice Leaf Disease dataset. The experimental outcomes show that the proposed ELD-CNN approach attains a superior accuracy of $\mathbf{9 9. 8 1 \%}$ and $\mathbf{9 9. 8 2 \%}$ on the Paddy Doctor and Rice Leaf Disease dataset as compared to the existing works like Lesion-aware visual transformer and Custom VGG-16.