Abstract: In transmission corridor inspection, insulator targets are generally small and exhibit significant scale variations, which can easily lead to false detections and missed detections of defects. To address these problems, an improved YOLOv8-dynamic enhanced window (DEW) based insulator defect detection algorithm was proposed. Firstly, a YOLOv8n detection network was constructed. To enhance the adaptive spatial aggregation capability of model for insulator targets with different poses and scales, the third-generation deformable convolution (DCNV3) was introduced. Partial original C2f modules and standard convolution modules in the backbone network were respectively replaced with C2f_DCNV3 and DCNV3_YOLO modules. Secondly, efficient channel attention (ECA) was introduced to enhance the recognition capability of model for target defect features. Finally, wise intersection over union loss (WIOU) was adopted to replace complete intersection over union loss (CIOU), so as to reduce the bounding box regression loss and improve the localization quality of prediction boxes. Experimental results show that the model file size, parameter quantity and computational complexity of the improved model are all slightly reduced, the number of floating-point operations is decreased by 1.3×10⁹, and the detection speed reaches 26 frames/s. For insulator defect detection, the η _mAP50 of reaches 98.3%, and the η _mAP50:95 reaches 86.6%. Its overall performance is superior to other mainstream object detection algorithms.