电力巡检图像中防振锤的区域全卷积网络识别方法

Identification method of region based fully convolution networks for damper in power inspection image

针对目前电力巡检图像中传统的防振锤检测方法仍存在效率低、精度差、计算成本高等问题,提出一种基于区域全卷积网络(region-based fully convolutional networks,R-FCN)的电力巡检图像防振锤智能识别方法.该方法通过特征提取网络自动提取防振锤特征,省却了传统检测方法特征提取的过程,提高了效率.此外,在R-FCN网络中采用位置敏感池化来引入平移变换,抵消全卷积网络造成的平移不变性问题,在检测精度和效率上均有较大提高.实验结果表明,该方法能准确检测出复杂背景下不同形态的防振锤,平均准确率高达88%,具有较强的鲁棒性.

In view of the problems of low efficiency,poor accuracy and high computational cost in the traditional detection methods of dampers in aerial inspection images,the region based fully convolutional networks(R-FCN)method for damper detection of aerial inspection images is proposed.The model based on R-FCN can automatically extract the features of dampers through the feature extraction network,which saves the process of feature extraction of traditional detection methods and improves the efficiency.In addition,the position-sensitive pool is adopted in R-FCN network to introduce translation changes,which can offset translation invariance caused by full convolution network,and greatly improve detection accuracy and efficiency.The experimental results show that the method can accurately detect different forms of dampers under complex background,and the accuracy is up to 88%,it also has strong robustness.