Crop Leaf Segmentation and Disease Detection Based on Shepherd Wide Residual Network

Abstract

The rapid rise in population necessitates a significant increase in agricultural productivity to satisfy the increasing demand for food. Timely identification of crop diseases is essential to ensure food security. Timely identification of crop disorders is vital for effective disease management and mitigating declines in crop yields. However, manually monitoring leaf diseases is labour-intensive and requires extensive knowledge of plant pathogens and considerable time and effort. The primary objective of this study is to develop an efficient and accurate deep learning-based approach named Shepherd Wide Residual Network (ShWRes-Net) for the automated detection and classification of crop leaf diseases, thereby reducing reliance on manual diagnosis and improving disease management in agriculture. The process begins with collecting crop leaf images from various datasets, then subjecting them to pre-processing leveraging a Wiener filter to mitigate noise. Leaf segmentation is then performed utilising the Dual-Branch U-Net model. Additionally, feature extraction is performed using a Complete Local Binary Pattern and Pyramid Histogram of Oriented Gradients. Finally, the identification of crop diseases is accomplished through the introduction of the ShWRes-Net model, which combines the Shepard Convolutional Neural Network with the Wide Residual Network. The ShWRes-Net method achieved a True Negative Rate of 90.877%, a True Positive Rate of 94.876% and an accuracy of 92.986%.