基于Faster RCNN的配网设备红外图像缺陷识别方法

Infrared image defect recognition method of distribution network equipment based on Faster RCNN

配网设备异常通常伴有发热现象,红外图像能够有效检测出发热设备,预防配网事故的发生。随着红外图像采集技术在配网巡检中的广泛应用,积攒了大量配网设备红外图像,传统机器学习方法对缺陷设备检测的准确率低、泛化性差。为此,文中将深度学习技术应用于配网设备红外图像检测,提出了基于Faster RCNN的缺陷检测方法。该方法采用深度残差网络提取图像特征,针对配网设备形状特点优化区域提议网络,借助共享卷积层训练网络。通过对8类典型配网设备缺陷测试表明,该方法对缺陷设备红外图像具有较高的检测准确率,且具有良好的鲁棒性和泛化能力。

Abnormal distribution equipments are often accompanied by fever.Infrared images can effectively detect the starting thermal equipment and prevent distribution network accidents.With the extensive application of infrared image acquisition technology in distribution network inspection,a large number of infrared images of distribution network equipment have been accumulated.Traditional machine learning methods have low accuracy and poor generalization for the detection of defective equipment.Therefore,deep learning technology is applied to the infrared image detection of distribution network equipment,and a defect detection method based on Faster RCNN is proposed.This method uses deep residual network to extract image features.According to the shape characteristics of the distribution network equipment,the regional proposal network is optimized,and the network sharing parameters are constructed through alternating training.Tests on 8 types of typical distribution network equipment defects show that the method has high detection accuracy for infrared images of defective equipment,good robustness and generalization ability.