Improved Semantic Segmentation for Rice Leaf Disease Identification With U-Net Architecture and Local-Based Sauvola's Thresholding

A number of rice leaf diseases possess a considerable effect on agriculture farming, making precise and effective disease detection essential. An enhanced semantic segmentation pipeline for accurate disease segmentation is presented in this work. The method is divided into many stages, such as U-Net deep learning architecture for illness segmentation, Sauvola thresholding & Canny edge detection. Prior to segmentation, preprocessing procedures including thresholding & Canny edge detection are carried out to improve input quality and lower false positives. The suggested model gets a higher validation accuracy of 96.12% and an F1 score of 99%, according to experimental validation. In addition to t-SNE plots and ROC curves, the model can clearly identify illnesses like bacterial blight, brown spot, and leaf smut by utilizing confusion matrices. Through the use of technical solutions, the suggested framework advances automated plant disease diagnostic technology, assisting sustainable farming applications and increasing food security.

Перевод пока недоступен