Defect Identification Technology for Rotate Object in Transmission Lines Based on Multi-scale Residual Network

In this paper,a multi-scale residual network based defect identification technology for rotate targets in transmission lines is proposed. By combining the Res2Net module and the Squeeze-and-Excitation Networks to improve the attention of the backbone network ResNet-50 for multi-scale features, using five parameters (x,y,w,h,theta) to represent the directed bounding box, and using angle regression to improve the detection sensitivity of rotate anchor boxes, which greatly improves the Defect diagnosis effect for large metal fittings in transmission lines.