Improved YOLOv8s for Wheat Scab Disease Detection

Wang Hao; Yu Xiao

doi:10.14083/j.issn.1001-4942.2025.08.019

Abstract：To address the issue of low recognition accuracy of wheat scab lesions detected by existing models,which hindered subsequent prevention and control efforts,this study proposed a wheat scab lesion de-tection algorithm named YOLOv8s-SCR,which was an improved version based on the YOLOv8s model.Using YOLOv8s as the basic network,this algorithm incorporated the advantages of ShuffleNet v2 and introduced the Squeeze and Excitation(SE)channel attention mechanism for enhancement,which could not only lightened the network model but also enhanced the model's focus on key features.Leveraging the adaptive training capa-bility of the CARAFE(Content Aware ReAssembly of FEatures)module,the nearest-neighbor interpolation upsampling in the original YOLOv8s was replaced,which enabled the provision of richer semantic information during the upsampling process.A multi-scale and trainable RFB(Receptive Field Block)module was em-ployed to further improve the model's detection performance by fusing features at different scales.Experimental tests demonstrated that the YOLOv8s-SCR model reduced the number of network parameters by 21.02％and decreased FLOPS(FLoating-point Operations Per Second)by 24.48％compared to the original model.On the test set,the mean average precision(mAP)of the model increased from 84.6％in the original model to 90.5％,representing 5.9 percentage points of improvement,thereby validated the effectiveness of the improved model in wheat scab detection.In summary,the YOLOv8s-SCR model proposed in this study could swiftly and effectively detect wheat scab lesions on wheat ears,providing robust support for subsequent prevention and control efforts.