基于U-Net的输电线覆冰识别

Identification of ice coating on transmission line based on U-Net

为了解决传统的依靠边缘检测+霍夫直线检测的这类覆冰识别方法易受覆冰监控环境背景的干扰,且只能识别边缘为直线这种单一类型覆冰的问题,本文采用深度学习中的语义分割的方法,不依靠任何先验特征,能对覆冰图像中的每一个像素分类,最终提出了输电线路覆冰识别方法,通过下采样网络提取覆冰特征、上采样网络恢复图像大小,在上采样时融合了低层次特征以提高最终分割精度;在训练过程中使用均方误差作为损失函数、适应性矩估计优化算法对网络参数进行更新。最终的实验结果表明,基于全卷积神经网络的覆冰识别方法相对于传统的数字图像处理法具有明显的优势,即使对于非规则的覆冰也能准确地识别,平均交并比值达到80.1%,该结果表明,这是一种理想的输电线覆冰识别方法。

In order to solve the traditional icing recognition method that relies on edge detection and Hough line detection,it is susceptible to interference from the background of the icing monitoring environment,and can only recognize the single type of icing that the edge is a straight line.This paper uses the semantic segmentation method in deep learning,does not rely on any prior features,can classify each pixel in the ice-coated image,and finally proposes a transmission line ice-coating recognition method,which extracts the ice-coated features and upper-coating features through a down-sampling network.The sampling network restores the image size,and integrates low-level features during upsampling to improve the final segmentation accuracy.during the training process,the mean square error is used as the loss function and the adaptive moment estimation optimization algorithm is used to update the network parameters.The final experimental results show that the convolutional neural network method has obvious advantages over the traditional digital image processing method,and it can recognize the irregular icing accurately,the average intersection-to-intersection ratio is 80.1%.The results show that this method is an ideal method for identifying icing on transmission lines.