Escrever uma avaliação PREreview

Rice leaf disease poses a significant threat to global food security and ecological stability. While existing studies predominantly concentrate on detecting symptoms after visible lesions emerge, early-stage disease features—which are often subtle—are critical for timely intervention. This paper introduces a novel detection approach based on an enhanced YOLOv11n architecture, tailored for the precise and efficient recognition of early-stage rice leaf disease indicators. To address the limitations of traditional detection techniques in identifying fine-grained features, we propose three key modules: the Multi-branch Large-kernel Fusion Depthwise (MLFD) module, the Multi-scale Dilated Transformer-based Attention (MDTA) module, and the Lightweight Detection Head (Lo-Head). The MLFD module enhances multi-scale feature extraction via parallel pathways and depthwise convolutions with large kernels. The MDTA module integrates both spatial and channel attention through a multi-head mechanism, improving the representation of diverse lesion features. Meanwhile, the Lo-Head detection head significantly reduces model complexity and parameter count, facilitating deployment on edge devices without compromising accuracy. Experimental results show that the proposed network achieves substantial performance gains. At an input resolution of 640×640, the model reaches a mean Average Precision (mAP@50:95) of 0.7927—an increase of 1.84 percentage points over the baseline YOLOv11n. It also outperforms Faster R-CNN, YOLOv5n, YOLOv8n, and YOLOv10n by 17%, 7.2%, 3%, and 2.5% respectively, while maintaining a low computational load of 6.2 GFLOPs and 2.66M parameters. These findings underscore the model’s potential for real-world agricultural applications, particularly in enabling early detection and precise disease control. The proposed method represents a step toward proactive plant health monitoring and precision agriculture.