An Oriented Object Detection Network for Cable Terminal Equipment in Infrared Images

Cable terminal equipment is an important part of power transmission lines. Detecting cable terminal equipment in infrared images is a prerequisite for realizing temperature measurement of it, which plays an important role in maintaining the transmission line stable. A detection network for cable terminal equipment in infrared images is proposed in this paper, including the feature extraction network, the feature fusion network and the oriented object detection head. The AngleConv module, the 2DDynaConv module and the LCSAM module are proposed and embedded into the feature extraction network. In addition, the LCIoU is proposed to calculate the prediction loss of the position, the long side length and the short side length, which is more suitable for the large ratio of the long side to the short side of cable terminal equipment than CIoU. Some ablation experiments are conducted, which prove the effectiveness of these modules and methods proposed. In the self-made dataset of infrared images of cable terminal equipment, the proposed network achieves mAP.50 of 95.4% and mAP.75 of 75.3%, and outperforms some mainstream oriented object detection networks.