Publication

Abstract : ABSTRACTAgriculture plays a critical role in feeding populations worldwide, yet farmers often lack the specialised knowledge required to detect and treat diseases in crops, which can lead to delays in disease diagnosis. This challenge is particularly evident in the case of rice crops, where early detection of leaf diseases is essential for minimising losses. Although numerous methods for classifying rice leaf diseases have been proposed, many of them have shown limited effectiveness due to the complexity and diversity of the diseases. To address this gap, an advanced method for rice leaf disease classification named Dense Kronecker Net (DK‐Net) is devised. Firstly, an input image is given into image preprocessing, which is done utilising a Wiener filter. Thereafter, image segmentation is conducted utilising M‐segNet. Then, image augmentation takes place using flipping, cropping, and rotation techniques. After that, the segmented image is delivered to the feature extraction process and extracted features include Grey Level Co‐occurrence Matrix (GLCM), entropy‐based Complete Local Binary Pattern (CLBP), and Local Gabor Directional Pattern (LGDP). Finally, leaf disease classification is exhibited utilising DK‐Net, which is a combination of DenseNet and Deep Kronecker Net. The DK‐Net achieved outstanding performance with the highest accuracy of 91.3%, True positive rate (TPR) of 91.4%, and True negative rate (TNR) of 91.6%. These results demonstrate that DK‐Net outperforms previous methods, offering a more accurate and robust solution for the early detection of rice leaf diseases.