Multi-target defect intelligent recognition of transmission line based on improved Faster-RCNN

In order to achieve rapid and accurate recognition of a large number of images of transmission line, and improve the ability to detect defects, according to the characteristics of defects in transmission line, an improved multi-target defect recognition algorithm based on the Faster-RCNN was proposed in this study. The algorithm adopted bilinear interpolation method and selected alternating optimization training method of Faster-RCNN. ROI Pooling in convolutional neural network was improved to ROI Align regional feature aggregation. Using the target image datasets of 7 types of transmission line, the structure and parameters of VGG16, VGG_CNN_M1024 and ZF network models under the framework of Faster-RCNN were optimized, and by comparing the precision rate-recall rate of the three network models, the refined VGG16 was selected as the feature extraction network of Faster-RCNN. The mean Average Precision (mAP) of the network model for target recognition of transmission line was about 88.01%. Also, the loss curves of region proposal network (RPN) regression and classification, Fast-RCNN regression and classification showed that the effectiveness of the improved VGG16 network model is ideal. Further, the test results indicated that the improved Faster-RCNN for target recognition on transmission line had a recall rate of 0.8252 to 0.9869, with an average of 0.9254; an accuracy of 0.5400 to 0.8616, with an average of 0.7273; F1 value (F-measure) of 0.6618 to 0.9230, with an average of 0.8114; the average accuracy of this model was 86.57% and the average detection time was 0.307 s. Capitalizing on these study findings, the improved Faster-RCNN has high effectiveness and real-time performance for defect recognition of transmission line, and can provide intelligent and effective decision-making basis for the detection of transmission lines defects in actual line inspections.