【摘要】 电力设备的缺陷检测对于保障电力系统的安全性和稳定性至关重要。然而，目前的缺陷检测系统在处理多样化缺陷和复杂背景时，检测精度和效率难以兼顾，无法满足实际应用需求。针对以上问题，提出了一种基于YOLOv8的电力场景缺陷检测识别方法。首先，利用ShuffleNetV2基础模块替换YOLOv8主干网络中的C2f模块，通过其深度可分离卷积和通道混合技术，大幅降低模型的参数量和计算复杂度。其次，引入三重注意力机制(Triplet Attention),增强模型对于缺陷相关特征的关注，抑制无关信息的干扰。最后，采用了PIoU(Powerful-IoU)损失函数对原有的损失函数进行改进，通过对中等质量锚盒的角点差异进行惩罚，有效提高边界框回归的精度和检测的整体性能。结果显示，所提方法的平均精度达到98.2%,且大小只有14 MB,模型部署在边缘端，对照片的处理效率达到0.12 s/张。所提出的方法为电力设备自动化缺陷检测提供了更为高效的解决方案，具有广泛的应用前景。更多还原

【Abstract】 The defect detection of power equipment is very important to ensure the safety and stability of power system. However, when the current defect detection system deals with diversified defects and complex background, the detection accuracy and efficiency are difficult to meet the needs of practical applications. A defect detection and recognition method of power scene based on YOLOv8 is proposed. Firstly, the C2f module in the backbone network of YOLOv8 is replaced by ShuffleNetV2 base module, and the parameter number and computational complexity of the model are greatly reduced through deep separable convolution and channel mixing technology. Secondly, a triplet attention mechanism is introduced to let model focus on defect-related features and suppress the interference of irrelevant information. Finally, PIoU(Powerful-IoU)loss function is adopted to improve the original loss function, and the accuracy of boundary box regression and the overall detection performance are effectively improved through the penalization of the corner difference of the medium quality anchor box. The results show that the average precision of the model is 98.2%,the size of the model is only 14 MB. The model is deployed at the edge, and the efficiency for photo processing is up to 0.12 s/photo. The method proposed provides a more efficient solution for automatic defect detection of power equipment and has a wide application prospect.更多还原