基于U-Net网络和椭圆度量学习的防震锤锈蚀识别

Identification of Anti-vibration Hammer Corrosion of High-voltage Transmission Lines Based on U-Net Network and Elliptic Metric Learning

高压输电线路中金属锈蚀会严重危害输电线路的安全运行。针对高压输电线背景复杂、缺乏有效锈蚀检测手段以及锈蚀检测准确率低等问题,提出了一种基于U-Net网络和度量学习的高压输电线防震锤锈蚀检测方法。相比其他深度网络,U-Net网络的参数量较少且直观,在小样本下具有较优的性能,利用U-Net网络可以将复杂背景条件下的高压输电线路中的防震锤完整分割出来。对分割后的防震锤图像提取HSV颜色特征和LBP纹理特征,并引入能够反映样本空间结构信息或语义信息的椭圆度量,通过椭圆度量学习实现高压输电线防震锤锈蚀的识别。实验结果表明,相比于支持向量机、BP神经网络、决策树等检测方法,该方法能够高效、准确地识别复杂背景环境下的高压输电线防震锤锈蚀。

Metal corrosion in high-voltage transmission lines can seriously endanger the safe operation of transmission lines.Aiming at the problems of complex background of high-voltage transmission lines,lack of effective corrosion detection methods,and low accuracy of corrosion detection,we propose a method for detecting corrosion of anti-vibration hammers of high-voltage transmission lines based on U-Net network and metric learning.Compared with other deep networks,the U-Net network has fewer parameters and is intuitive with better performance in a small sample.U-Net networks can be used to completely isolate the seismic hammers in high-voltage transmission lines under complex background conditions.The HSV color features and LBP texture features are extracted from the segmented seismic image,and an ellipse metric that reflects the spatial structure information or semantic information of the sample is introduced.The ellipse metric learning is used to identify the anticorrosive hammer corrosion of high-voltage power lines.Experiment shows that compared with support vector machine,BP neural network,decision tree and other detection methods,the proposed method can efficiently and accurately identify the anti-vibration hammer corrosion of high-voltage transmission lines in complex background environments.