Rice diseases such as bacterial leaf blight, brown spot, and blast pose significant threats to global food security by reducing crop yields, making early and accurate detection crucial. This study aims to improve automated segmentation of rice leaf diseases using advanced deep learning techniques, specifically U-Net with MobileNetV2, DeepLabV3+ with EfficientNetB4, and U-Net with EfficientNetB7 integrated with attention gates. The models were evaluated on a dataset of diseased rice leaves for segmentation accuracy, computational efficiency, and robustness in real-world conditions. Results show that the U-Net with EfficientNetB7 and attention gates outperforms the other models, particularly for complex leaf images, achieving superior accuracy and generalization. This research provides a practical, real-time solution for early disease detection in rice, contributing to precision agriculture by helping reduce crop losses and optimize the use of agrochemicals, ultimately promoting sustainable crop management.