期刊名称:IAENG International Journal of Computer Science
印刷版ISSN:1819-656X
电子版ISSN:1819-9224
出版年度:2020
卷号:47
期号:3
语种:English
出版社:IAENG - International Association of Engineers
摘要:The copper bus-bar in the power distribution cabinet is a device conveying current and connecting electrical components in the circuit. If the fastening bolts on copper bus-bar are loose or falling off, high current can melt the copper bus-bar, causing significant damage. A defect detection method on power distribution equipment based on an improved cascaded network is presented, which fully embodies the advantages of deep-learning on feature extraction, region proposal, small object recognition. This method provides high accuracy, real-time and high-efficiency identification effect for bolts fastening status (normal, falling off and loose). In the paper, Faster-RCNN is improved by adding support vector machine (SVM) after RPN (Region Proposal Network) to improve classification accuracy on the foreground and background images. Only the foreground features including target are sent to the subsequent network for training to make this method take good balance between accuracy and training efficiency. In addition, the gravitational search operator is introduced into the Normal Mutation cat swarm (NMCSO) algorithm to improve the global optimization ability and significantly enhance the classification precision of SVM. The experimental results show that the improved cascaded network is superior to the existing deep-learning algorithms and traditional machine learning methods in accuracy and training time for power distribution equipment detection. The mean average precision (mAP) for image set of this paper is 89.51%, and the training time is the shortest.