U-Net-Based Segmentation of Disease-Affected Areas in Rice Fields Using Aerial Imagery

Rice production often faces serious challenges due to plant disease attacks that can significantly reduce crop yields. Early detection of infected areas is a crucial step in mitigating production losses. With the advancement of agricultural technology, the use of aerial imagery from Unmanned Aerial Vehicles (UAVs) has become a potential solution to support rapid and efficient disease detection. This study applies U-Net-based image segmentation to detect disease-infected rice fields using aerial imagery. The U-Net model used is a standard U-Net with four different backbone types, to explore the effect of backbones on segmentation accuracy. The research process includes image preprocessing, model training, and model performance evaluation using the Intersection Over Union (IoU) and Dice Similarity Coefficient (DSC) metrics. In this study, a secondary dataset was used, consisting of aerial images of diseased rice field areas. The results show that the U-Net model with the EfficientNetB3 backbone provides the best performance with an mIoU value of 0.7102 and an mDSC of 0.8283 in the test data, outperforming the standard U-Net model.