RiceLDD-YOLO: Optimizing the YOLOv13 Model to Improve the Efficiency of Rice Leaf Disease Detection

Rice is a staple crop that underpins global food security, providing sustenance for more than half of the world’s population. However, its productivity is increasingly threatened by foliar diseases, which cause substantial economic losses and reduce yield quality across diverse cultivation environments. Early and accurate detection of leaf diseases is therefore crucial for timely intervention and the development of intelligent agricultural monitoring systems. In this paper, RiceLDD-YOLO, an enhanced version of YOLOv13, is proposed specifically for robust rice leaf disease detection under real-field conditions. The proposed model incorporates three key improvements: an ImDS-C3k2 convolutional module that strengthens deep feature extraction and preserves fine-grained lesion patterns, an improved multi-scale aggregation head integrating the SPPF module to enhance contextual representation, and a specialized Rice-IoU loss function that stabilizes bounding-box regression for small, elongated, and irregular disease regions. Experimental results demonstrate that RiceLDD-YOLO significantly improves detection accuracy, achieving an mAP50 of 56.1% and mAP _50:95 of 33.5%, while maintaining a real-time inference speed of 1.7 ms, ensuring the model remains lightweight and suitable for deployment on edge devices. These findings highlight the potential of RiceLDD-YOLO as a practical and effective solution for intelligent crop-health monitoring and precision agriculture applications.