Study on defect detection in lightweight insulators based on improved YOLOv8

Aiming at the problems of insulator detection such as low recognition accuracy for small targets and the algorithm model is too large and difficult to be deployed to the edge devices. In this paper, a lightweight YOLOv8-ASF-P2 insulator defect detection model is designed. The model introduces the ASF and P2 detection layer, and at the same time according to the idea of ASF to add the P2 detection layer, the new network structure is trained and then pruned. After pruning, the mAP of this algorithm is 89.5%, the model size is 2.1 MB, and the detection speed is 144.9FPS. 1.7% improvement in mAP, 64.7% reduction in model size, and 40% improvement in detection speed compared with the YOLOv8 algorithm, which verifies the effectiveness of the improved method.