This paper proposes an improved YOLOv8 model for insulator defect detection, aiming to enhance detection accuracy and efficiency. The improved model, named YOLOv8-GSC, integrates a Multi-scale Large Kernel Attention (MLKA) module, a GSC_C2f module, and the SIoU loss function. The MLKA module enhances feature extraction by capturing both local and non-local contextual information, while the GSC_C2f module reduces computational complexity and improves feature representation. The SIoU loss function accelerates model convergence and improves small target detection accuracy. Experimental results show that the improved model achieves an mAP of 99.22% and an FPS of 55.73, outperforming the original YOLOv8s and YOLOv5s models. The model size is only 30.18MB, making it suitable for real-time applications. The improved model demonstrates superior performance in detection accuracy, speed, and robustness, making it effective for insulator defect detection in complex environments. The study also compares the improved model with other detection algorithms, including Faster R-CNN, SSD, CenterNet, YOLOv3, YOLOv4, YOLOv5s, and YOLOv8s, showing that the improved model achieves the highest mAP and is the fastest among all tested models. The results validate the effectiveness of the proposed improvements in enhancing the performance of insulator defect detection.This paper proposes an improved YOLOv8 model for insulator defect detection, aiming to enhance detection accuracy and efficiency. The improved model, named YOLOv8-GSC, integrates a Multi-scale Large Kernel Attention (MLKA) module, a GSC_C2f module, and the SIoU loss function. The MLKA module enhances feature extraction by capturing both local and non-local contextual information, while the GSC_C2f module reduces computational complexity and improves feature representation. The SIoU loss function accelerates model convergence and improves small target detection accuracy. Experimental results show that the improved model achieves an mAP of 99.22% and an FPS of 55.73, outperforming the original YOLOv8s and YOLOv5s models. The model size is only 30.18MB, making it suitable for real-time applications. The improved model demonstrates superior performance in detection accuracy, speed, and robustness, making it effective for insulator defect detection in complex environments. The study also compares the improved model with other detection algorithms, including Faster R-CNN, SSD, CenterNet, YOLOv3, YOLOv4, YOLOv5s, and YOLOv8s, showing that the improved model achieves the highest mAP and is the fastest among all tested models. The results validate the effectiveness of the proposed improvements in enhancing the performance of insulator defect detection.