Abstract: To tackle excessive memory use and low precision in insulator detection, we enhanced YOLOv8s by integrating feature interaction and lightweight fusion network traits, creating lightweight insulator defect detection model. First, the subsampled convolution ADown was proposed to boost multi-level feature learning ability and cut computing costs. Second, three-branch C2f-ATripletAt module was designed to promote the backbone network information interaction and reduce the background interference. Then, a new HS-FPN-ECA feature fusion network was designed to reduce small target feature loss and enhance model lightness. Next, Focal-EIoU(enhanced intersection over union) replaced CIoU(complete intersection over union) to speed convergence and address precision issues from imbalanced insulator samples. Finally, CPLID dataset was expanded for experiments. Results show that the model can accurately identify insulators and defects in various scenarios with good robustness, 98.8% accuracy, and 57.7% & 52.1% reduction in parameters & calculations, respectively, the memory size is only 9.8 MB, suitable for unmanned aerial vehicle equipment.