| 摘 要: 针对绝缘子检测背景条件复杂,对小目标和被遮挡绝缘子缺陷检测精度低的问题,提出了改进的YOLOv10n绝缘子缺陷检测系统。在模型优化方面采取EfficientViT网络替换C2f模块和卷积模块,减少模型复杂程度,提高检测效率,并选取加入 ACmix注意力机制,更高效提取特征信息,用于有效提升检测精度。引入SIoU损失函数,提高模型收敛速度和鲁棒性。通过实验验证改进后模型的检测效果,最终实验结果表明,改进后的 AEYOLOv10n网络能在大量图像中精准有效地识别绝缘子的缺陷,精确率、召回率分别达到了96.3%和93.8%,平均精度由90.4%提升到了98.1%,为电力系统运行提供有力支持。 |
| 关键词: 深度学习 绝缘子目标检测 EfficientViT ACmix注意力机制 损失函数优化 |
|
中图分类号: TP391.4
文献标识码: A
|
| 基金项目: 国家重点研发计划项目(2017YFC0821001-2) |
|
| Improved YOLO-Based Insulator Defect Detection |
|
ZHU Lizhong, YANG Biqi
|
(School of Automation and Electrical Engineering, Shenyang Ligong University, Shenyang 110159, China)
zlz2686312@sina.com; 2252448739@qq.com
|
| Abstract: To address the challenges of complex backgrounds in insulator detection and low accuracy in identifying small or occluded insulator defects, this study proposes an enhanced YOLOv10n insulator defect detection
system. For model optimization, the C2f module and convolutional blocks are replaced with the EfficientViT network to reduce model complexity and improve detection efficiency. The ACmix attention mechanism is incorporated to extract
feature information more effectively, thereby enhancing detection accuracy. The SIoU loss function is introduced to accelerate model convergence and improve robustness. Experimental validation demonstrates that the improved AEYOLOv10n network achieves precise and efficient identification of insulator defects across large-scale image datasets.The final results show precision and recall rates of 96.3% and 93.8% , respectively, while the mean average precision
(mAP) increased from 90.4% to 98.1% . This advancement contributes significantly to power system operations. |
| Keywords: deep learning insulator object detection EfficientViT ACmix attention mechanism loss function optimization |
