Insulator Defect Detection Network Integrating the GD Mechanism with Attention Mechanism

Aiming at the problem of poor detection effect of insulator defect images in complex inspection environment due to small target and large scale difference, an insulator defect detection algorithm based on improved YOLOv8 is proposed. Firstly, a new attention mechanism, Channel Prior Multi-scale Convolutional Attention, is proposed, which is added after Backbone to enhance the feature output and further strengthen the identification of different insulator defects; secondly, the introduction of the Gather and distribute mechanism to reconstruct the neck network, which improves the fusion ability of multi-scale target features; finally, multi-scale feature combination is achieved by adding the Scale sequence feature fusion module to the front of the Head, which enhances the detection ability of multi-scale and small objects. The experimental results show that the mAP0.5 of the algorithm in this paper reaches 92.3%, which is 7.5% higher than that of the standard YOLOv8, and the average accuracy of the flashover and broken defect detection is increased by 14.0% and 8.3%, respectively, and the defect detection algorithm proposed in this paper is able to better and accurately identify defects in insulators of different scales.