基于深度学习的无人机巡检图像销钉故障检测

Pin fault detection in UAV inspection image based on deep learning

摘要:

随着经济和社会的发展,发电量和用电量逐年上升。安全的电力保障关系到国计民生,在常年的使用过程中,由于电力传输的输电线路受到外界环境的影响,使得输电线路部件容易出现不同程度的破损,其中销钉是固定螺母的关键零件,销钉的脱落会导致各部件之间连接的不稳定,这给输电网络的安全运行带来了极大的隐患。随着深度学习技术在计算机视觉领域中的应用,使得机器自动识别销钉这一输电线路系统中的微小部件成为现实。采用Faster R-CNN算法对无人机巡检图像中的销钉脱落故障进行识别,并讨论了不同分类器对识别结果的影响,然后对ACF+Adaboost、Hough+LSD和Faster R-CNN检测方法进行比较。实验结果表明,基于Faster R-CNN的目标检测方法对于输电线路中销钉脱落故障的识别率可达到96%,同时对正常销钉的识别率最高可达98%。

Abstract:

With the development of economy and society, power generation and electricity consumption increase year by year. Safe power supply is related to national economy and people"s livelihood. In the process of many years of use, due to the transmission of power transmission line is often influenced by the external environment, making it easier for the transmission line components appear different degree of damage. The pin is the key to the fixed nut parts. The shedding of pin will lead to an unstable connection between the components that brings great challenge to the safe operation of power transmission network. With the application of deep learning technology in the field of computer vision, the machine automatic identification pin which is a tiny part in the transmission line system has become a reality. In this paper, Faster R-CNN algorithm was used to identify pin shedding fault in Unmanned Aerial Vehicle (UAV) patrol image, and the impact of different classifiers on recognition results was discussed. Then Aggregate Channel Features (ACF)+Adaboost, Hough+ Line Segment Detector (LSD) and Faster R-CNN recognition methods were compared. The experimental results show that the recognition rate of Faster R-CNN based target detection method for pin falling fault in transmission lines can reach 96%, and the recognition rate of normal pin can reach 98% at the same time.