Insulator defect detection based on improved YOLOX

Detecting insulator defects is a crucial stage in the process of inspecting a power grid. To increase the accuracy of insulator defect identification, this study proposes a detection technique based on an enhanced version of the YOLOX algorithm, which can improve the accuracy of insulator defect detection while maintaining the detection speed. The algorithm adds the Efficient Channel Attention (ECA) mechanism to the Spatial Pyramid Pooling (SPP) based on YOLOX-s, i.e., ECA-SPP, adds the ECA mechanism to the feature pyramid for cross-channel interaction, suppresses interference information and strengthens the model's focus on target features according to the generated corresponding channel weight values, and then fuses the recalibrated feature maps with deeper features to improve the target feature expressiveness. The Complete Intersection over Union (CIoU) is utilised to calculate the loss, while the distance between the centroids of the two frames and the aspect ratio is factored into the penalty term, and the loss function is continuously changed and updated to accelerate the convergence speed of the model. The test findings demonstrate that the revised mean of Average Precision (mAP) algorithm for YOLOX achieves 93.21% accuracy, which is 6.17% higher than the original algorithm.