基于深度残差UNet网络的电气设备红外图像分割方法

红外图像处理是实现电气故障诊断的有效手段，而电气设备分割是故障检测的关键环节。针对复杂背景下红外图像电气设备分割难问题，本文采用深度残差网络与UNet网络相结合，深度残差网络替代VGG16对UNet网络进行特征提取和编码，构建深度残差系列Res-Unet网络实现对电气设备的分割。以电流互感器和断路器两种电气设备红外图像分割为例测试Res-Unet网络分割效果，并与传统的UNet网络和Deeplabv3+网络进行对比。通过对数量为876的样本进行测试，实验结果表明，Res18-UNet能够准确地分割电气设备，对电流互感器和断路器的分割准确率超93%，平均交并比大于89%，且分割准确性优于UNet及Deeplabv3+网络模型，为实现电气故障智能诊断奠定基础。

Deep Residual UNet Network-based Infrared Image Segmentation Method for Electrical Equipment

Infrared thermal image processing is an effective method for detecting defects in electrical equipment. Aiming at the problem of electrical equipment segmentation in infrared thermal images with a complex background, in this study we propose a deep residual UNet network for infrared thermal image segmentation. Using a deep residual network to replace VGG16 to perform feature extraction and coding for the UNet network, a deep residual series UNET network was constructed to segment electrical equipment. To validate the effectiveness of the Res-UNet network, infrared images, including current transformers and circuit breakers, were used to test the segmentation results and were compared with the traditional UNet and Deeplabv3+ networks. The networks were tested using 876 images. The experimental results show that RES18-UNET can accurately segment electrical equipment; the segmentation precision of current transformers and circuit breakers is greater than 93%, and the mean intersection over union (MIoU) is greater than 89%. Our method obtains more accurate segmentation results than UNet and Deeplabv3+, setting the basis for intelligent diagnosis of electrical faults.