基于深度学习的电力巡检图像实时处理与识别算法研究  

Research on deep learning based real⁃time processing and recognition algorithm of power patrol image

作　　者：石志彬 罗望春[1] 莫兵兵[1] 张福 SHI Zhibin;LUO Wangchun;MO Bingbing;ZHANG Fu(Maintenance&Test Center,China Southern Power Grid EHV Transmission Company,Guangzhou 510663,China)

机构地区：[1]中国南方电网超高压输电公司检修试验中心,广东广州510663

出　　处：《电子设计工程》2022年第23期189-193,共5页Electronic Design Engineering

基　　金：中国南方电网有限责任公司科技项目(CGYKJXM00000028)。

摘　　要：应用人工智能技术对航拍采集的巡检图像进行目标检测和缺陷识别已成为现代电力巡检的发展趋势。文中基于深度学习技术提出了一种改进的识别算法,通过改进Faster-RCNN模型,结合HSI颜色特征提取,实现图像实时处理与输电线路的各类常规故障识别。将航拍图片归一化处理,RGB颜色模型转化为HSI颜色模型,遍历HSI空间的每个像素点,根据图片颜色特征判断像素点是否发生故障;建立Dense Net网络,将RoI Align层与预测层连接,应用改进Faster-RCNN目标检测模型对巡检线路训练数据集进行目标缺陷识别。实验分析结果表明,文中所提方法的故障缺陷识别精确率可达92.54%,具有实时性强、识别精度高等特点。The application of artificial intelligence technology to detect the target and identify the defects of aerial photography inspection image has become the development trend of modern power inspection.Based on the deep learning technology,an improved recognition algorithm is proposed.By improving Faster⁃RCNN model and combining with HSI color feature extraction,the real⁃time image processing and the common fault identification of transmission lines are realized.The normalized processing of aerial photos is carried out,and RGB color model is transformed into HSI color model.Every pixel point in HSI space is traversed to determine whether the pixel points fail according to the color characteristics of the images;The Dense Net network is established,which connects RoI Align layer with prediction layer.The improved Faster⁃RCNN target detection model is used to identify the target defects in the patrol line training data set.The experimental results show that the fault defect identification accuracy of the method can reach 92.54%,which has the characteristics of real⁃time and high recognition accuracy.

关 键 词：深度学习 图像处理 目标检测 缺陷识别 

分 类 号：TP807[自动化与计算机技术—检测技术与自动化装置]