基于Faster R-CNN的输电线路缺陷识别模型研究

随着多旋翼无人机引入输电线路巡检作业后,对巡线人员通过图像判断线路上设备是否有缺陷提出了新的挑战。为了帮助巡线人员做出准确决策,提高发现缺陷的能力,基于深度卷积神经网络,搭建了适用于无人机图像识别的输电线路缺陷识别网络模型。首先详细描述了输电线路缺陷识别图像数据库的建立过程,然后通过分析对比三个预训练前端网络的性能及多个参数对网络模型识别准确率的影响,得到基于Faster R-CNN的输电线路缺陷最优识别网络模型。经过测试集验证,提出的缺陷识别网络模型的识别准确率达到了90%以上,单张图片耗时达到了毫秒级,在识别准确率和耗时上均明显优于其他识别网络模型,为实际巡线工作中的输电线路缺陷判别提供智能有效的决策依据,是机器学习在智能电网中应用的有益探索。

A defect recognition model of transmission line based on Faster R-CNN

With the introduction of the multi-rotor UAV into the transmission line inspection operation,it is a new challenge for the patrol worker to judge whether the equipment on the line is defective through the image.In order to help the patrol worker make accurate decisions and improve the ability to find defects,a transmission line defect identification network model is built for the UAV image recognition based on the deep convolutional neural network.The establishment of the transmission line defect identification image database is firstly described,and then the performance of three pre-trained front-end networks and the different influences of multiple parameters on the recognition accuracy of network models is analyzed and compared.Finally,a transmission line defect network model based on Faster R-CNN is established.After the test set verification,t he recognition accuracy of the defect recognition network model proposed in this paper reaches more than 90%,meanwhile,t he time consuming of a single picture reaches milliseconds,which is obviously superior to other methods in recognition accuracy and time consumption.And the defect recognition network model provides an intelligent and effective decision-making basis for the identification of transmission line defects in actual line work,and is a useful exploration for the application of machine learning in smart grids.