基于深度学习的输电线路锈蚀检测方法的研究  

Study on detection method of transmission line rusty based on deep learning

作　　者：李辉[1] 钟平 戴玉静 吕东辉[2] Li Hui;Zhong Ping;Dai Yujing;Lv Donghui(NARI Technology Co.,Ltd.Shenzhen Branch,Shenzhen 518000,China;School of Communication and Information Engineering,Shanghai University,Shanghai 200444,China)

机构地区：[1]国电南瑞科技股份有限公司深圳分公司,深圳518000 [2]上海大学通信与信息工程学院,上海200444

出　　处：《电子测量技术》2018年第22期54-59,共6页Electronic Measurement Technology

基　　金：国家自然科学基金(61471230)项目资助

摘　　要：高压输电线路环境复杂、生锈部件形状差异较大,其中的锈蚀难以检测。针对该问题,提出基于深度学习的Faster-RCNN目标检测模型与锈蚀HSI颜色特征结合的锈蚀检测方法。先利用锈蚀的颜色特征去除图像中的干扰背景,再使用Faster-RCNN对去除干扰背景后的图像进行锈蚀检测。与单独采用基于HSI颜色特征的锈蚀检测方法和单独采用深度学习方法相比,深度学习算法与HSI颜色特征结合的方法在判别锈蚀有无的正确率和召回率上有较大提高,在定位的准确性上相对于基于HSI颜色特征的锈蚀检测方法也有较大提高,与单独采用深度学习方法检测的准确率较为接近。实验结果表明,该方法符合实际需求,且优于基于HSI颜色特征的锈蚀检测方法和单独使用深度学习方法。The environment of high-voltage transmission lines is complex, and the shape of rusted parts is quite different. The rust is difficult to detect. In order to solve this problem, a rust detection method combining Faster-RCNN with rust HSI color characteristics was proposed. Firstly, the rusted color feature was used to remove the interference background in the image, then Faster-RCNN was used to detect the image after removing the interference background. Compared with using rust detection method based on HSI color characteristics and the independent use of the deep learning method alone, the combination of the deep learning algorithm and rust HSI color characteristics greatly improves the accuracy and recall rate of the presence or absence of rust. Compared with rust detection method based on HSI color features, the accuracy of positioning is also greatly improved, which is very close to the accuracy of using the Faster-RCNN algorithm to detect the rust positioning. The experimental results show that the method accords with the actual demand and is superior to rust detection method based on HSI color features and the independent use of deep learning methods.

关 键 词：目标检测 深度学习 锈蚀识别 输电线路 图像处理 

分 类 号：TN911.73[电子电信—通信与信息系统]