SCRFPN: spatial and channel reconstruction feature pyramid networks for insulator detection in transmission lines

Insulator defects commonly occur in high-voltage transmission lines and can compromise the insulator's ability to isolate voltage and support conductors, thereby impacting the safe and stable operation of the power system. Therefore, quickly and accurately locating insulator defect positions is of significant importance. However, due to the diverse types and irregular shapes of insulator defects, detection accuracy is often low. To address this, in this paper, we design a Spatial and Channel Reconstruction Feature Pyramid Network (SCRFPN) to enhance the model's feature extraction and fusion capabilities. Furthermore, we introduce a dyhead module after the SCRFPN, which integrates scale-aware, spatial-aware, and task-aware attention, enabling the model to focus more effectively on defective areas of the insulators in images. Finally, we construct an insulator defect dataset that encompasses various scenarios, providing a benchmark for detecting insulator defects in high-voltage transmission lines. By integrating SCRFPN into a basic cascade RCNN model, we achieved a 2.3% increase in average precision (AP), significantly improving the model's detection performance.