Insulator Detection Based on Improved Lightweight Rotation Detector
Yu Lin, Lianfang Tian, Qiliang Du, Fengrui Qu · 2023
Insulator overheating defect detection is an important task in power inspection. The traditional horizontal detector is not precise enough for positioning inclined insulators. Although the general rotation detector can produce a more accurate rotation detection box, it is not suitable for edge device with limited computing power due to its large amount of model parameters. In this paper, we propose a lightweight Rotated RetinaNet (LR-RetinaNet) to address these issues. Firstly, we propose Brief PAFPN to achieve more efficient feature fusion and effectively reduce model parameters. Secondly, we design a lighter Brief Rotated Head to address the increased model computational complexity caused by the heavy head of RetinaNet. Finally, we conducted a series of experiments, which showed that our proposed LR-RetinaNet can effectively reduce model parameters while ensuring detection accuracy, making it more suitable for edge device.