为提升输电线路缺陷识别效果,研究了一种利用边缘算法优化Faster R-CNN算法的输电线路缺陷识别方法。通过无人机采集输电线路图像,采用极值中值滤波算法降噪,输入Faster R-CNN模型,提取缺陷特征,利用RPN网络确定目标候选区域;利用边缘算法优化Faster R-CNN算法以确定像素点梯度幅值,并抑制非极大值;训练模型,完成输电线路缺陷识别。测试结果显示,该算法能够提升各主要缺陷类别的识别准确率,准确率达85%以上。

To improve the effectiveness of transmission line defect recognition, this paper studied a transmission line defect recognition method that utilized edge algorithms to optimize the Faster R-CNN algorithm. Transmission line images through drones were collected. Extreme median filtering algorithm to reduce noise was used. Faster R-CNN model was inputted and defect features were extracted. RPN network to determine target candidate regions was used. Faster R-CNN algorithm using edge algorithm was optimized to determine pixel gradient amplitude and suppress non maximum values. The model was trained to complete the identification of transmission line defects. The test results showed that the algorithm studied can improve the recognition accuracy of various major defect categories, with an accuracy rate of over 85%.