SSW-YOLO: An Enhanced YOLO11-Based Framework for Multi-Class Rice Disease and Pest-Induced Damage Detection

Under complex field conditions, multi-class rice disease and pest-induced damage detection remains challenging. These challenges include substantial variations in target scale, densely distributed small targets, and blurred boundaries. To overcome these difficulties, this work presents a multi-class rice disease and pest-induced damage detection framework, SSW-YOLO. It is derived from YOLO11n to improve detection performance. First, SPDConv is incorporated into the backbone to preserve small-scale and densely distributed lesion features during conventional downsampling. C3k2-SAConv is used in the neck to adaptively adjust receptive fields to facilitate information exchange and fusion among multi-scale features. Finally, WIoU v3 is adopted instead of the original CIoU loss to achieve more stable bounding box regression for lesions with irregular morphology and blurred boundaries. WIoU v3 employs a dynamic non-monotonic focusing strategy that reduces the influence of poor-quality samples during optimization and improves localization performance. In a single-run evaluation, SSW-YOLO obtained 80.7% Precision, 70.1% Recall, and 77.2% mAP@0.5. Relative to YOLO11n, the corresponding gains were 3.7, 3.8 and 3.8 percentage points, respectively. In addition, SSW-YOLO has 4.18 M parameters and requires 10.4 GFLOPs. These results show that SSW-YOLO improves the performance of multi-class rice disease and pest-induced damage detection on the evaluated dataset and shows potential for accurate rice disease identification and intelligent monitoring.