To address the problems of small scale, blurred boundaries, weak features and complex background interference in insulator flashover trace detection from UAV inspection images, an improved YOLOv8-based method is proposed. CBAM and a receptive field enhancement structure are introduced into the Backbone to strengthen feature representation of key trace regions. A weighted bidirectional feature fusion structure with a preserved P2 branch is constructed in the Neck to enhance the fusion of shallow details and deep semantic features. A P2 small-object detection head and detail enhancement blocks before the P2/P3 branches are added in the Head to improve localization and recognition of tiny flashover traces. A hard-sample closed-loop optimization strategy is further used to improve model stability. Experimental results show that the proposed method can effectively detect insulator flashover traces and has potential for engineering application.