Deep Learning-Based Defect Identification for Transmission Tower Bolts:Optimization of YOLOv3 and ResNet50 Algorithms

Abstract

In the power industry, it is often necessary to carry out inspection on the transmission tower, including the detection of bolts on the tower. In the past, the method of detecting whether the bolts are tightened is manually completed, which has a large workload and low efficiency. Therefore, this paper proposes to use deep learning method to detect and identify the defects of bolts on the tower to improve the detection efficiency. In bolt detection, the data enhancement method is used to increase the number of samples. In the link of bolt detection, YOLOv3 algorithm is used. In order to improve the detection accuracy of the algorithm, the algorithm is also optimized. In the identification of bolt defects, ResNet50 network, data enhancement and transfer learning are adopted to solve the problem of bolt defect identification, and the ResNet50 network is optimized to improve the recognition quality of the algorithm. The recognition accuracy of nut and screw defects is 0.95 and 0.90 respectively. The feasibility of the identification method is confirmed, which can be used to identify bolt defects of transmission tower and improve the detection efficiency and quality of transmission tower