基于注意力机制的多尺度缺陷绝缘子检测算法

Multi-Scale Defect Insulator Detection Algorithm Based on Attention Mechanism

绝缘子是保障输电线路平稳运行的重要电力部件之一。对于故障区域小、周围所处环境复杂的缺陷绝缘子检测是一项极具挑战性的任务,传统的检测方法存在精度不高、效率低和缺少大型公开数据集的问题。针对以上问题,提出一种基于注意力机制的多尺度缺陷绝缘子检测算法。在YOLOv3算法的基础上,使用K-means++匹配新的锚点坐标,将通道注意力机制SENet结构融入特征提取网络Darknet53中,增加多个检测尺度提升检测精度,并使用数据增强技术扩充缺陷绝缘子数据集。实验结果表明,上述方法在满足实时检测的要求下获得了94.42%的平均准确率和95.74%的召回率。

Insulators are one of the important power components to ensure the smooth operation of transmission lines,where the detection of defective insulators with small fault areas and complex environment is a challenging task.Nevertheless,traditional detection methods suffer from low accuracy,low efficiency,and lack of large public data sets.To address the above problems,a multi-scale defect insulator detection algorithm based on attention mechanism is proposed in this work.Based on YOLOv3 algorithm,K-means++was employed to match the new anchor coordinates,the channel attention mechanism SENet structure was integrated into feature extraction network Darknet53,multiple detection scales were added to improve detection accuracy,and the data augmentation technology was used to expand the data set of defect insulator.The experimental results show that the method achieves an average precision of 94.42%and a recall rate of 95.74%while still meets the requirements of real-time detection.