Violation-YOLO:面向电力作业违章识别的实时目标检测模型OA

To address the issues of insufficient feature extraction for irregularly shaped and multi-scale targets and the incom-plete coverage of sample categories in power operation violation recognition,this paper proposes a Violation-YOLO detection method based on an improved YOLOv8.Firstly,this method proposes a Deformable-C2f(DC2f)module,which incorporates de-formable convolution and depthwise separable convolution to enhance the feature representation of irregularly shaped targets while reducing the number of parameters.Secondly,the proposed method proposes an Adaptive Multiscale Channel Attention(AMCA)module that integrates multi-scale depthwise separable convolution,adaptive convolution,and channel attention mechanisms,thereby enhancing the feature extraction capability for targets of various scales with relatively low parameter count.Finally,the proposed method constructs a target detection dataset for power operation violation recognition,named the EOVR-v1.0,which includes several target categories not present in publicly available datasets.On the EOVR-v1.0 dataset,ablation ex-periments verify the effectiveness of the DC2f module and the AMCA module,as well as their combination.Compared with the baseline YOLOv8 model,mAP@0.5 and mAP@0.5-0.95 are improved by 2.6 percentage points and 3.1 percentage points respec-tively,with a slight decrease in the number of parameters and GFLOPS.Comparative experiments with seven mainstream YOLO models show that the method achieves the best results of 87.6％and 62.6％in mAP@0.5 and mAP@0.5-0.95,respectively,and fully meets the real-time requirements of the power operation violation recognition scenario.