Detection of Pin Defects in Transmission Lines Based on Dynamic Receptive Field

Zhenyu Zhang, Jianming Du, Shaowei Qian, Chengjun Xie

Abstract: Pin plays a role in fixing components and stabilizing structures in transmission lines. At present, UAVs are introduced into the inspection of transmission lines. Due to the small size of the pins, which account for fewer than 0.04% of the pixels in the aerial image, their defects are difficult to be detected. At the same time, the complex background and different shooting angles are prone to cause the defect samples difficult to be identified, so the detection accuracy is not high and robust. Aiming at these problems, this paper proposes a detection method for pin defects in transmission lines based on the dynamic receptive field. In this method, a Dynamic Receptive Field (DRF) module is used to extract the contextual features of pin defects, which effectively fuses receptive fields of different sizes and in-channel information to improve detection accuracy. Second, a Spatial Activation Region Proposal Network (SARPN) is proposed to enhance the information acquisition of regions of interest in the proposal network. The experimental results show that the proposed method performs a noticeable effect on the two-stage object detection method, in which mAP is increased by 4.4% and recall is increased by 8.6% based on Cascade RCNN.