| 摘 要: 针对绝缘子缺陷目标小、分布零散等问题,提出一种基于YOLOv8的轻量化检测算法。通过自适应降采样(ADown)技术融合平均池化与最大池化优势,降低局部变化敏感性并提升参数利用效率;结合高效远程注意力网络(ELAN)和跨阶段局部网络(CSPNet),设计轻量化骨干网络RepGELAN,借鉴RepVGG卷积结构优化卷积堆叠,实现轻量化与高效推理;应用双向特征金字塔网络(BiFPN)增强特征整合,提升小目标检测能力。实验表明,算法参数量减少了28.88%,mAP@50提升了1.6%,mAP@95提升了0.7%,平衡了轻量化与性能需求。 |
| 关键词: YOLOv8 绝缘子缺陷检测 轻量化 ELAN CSPNet BiFPN |
|
中图分类号: TP391.4
文献标识码: A
|
| 基金项目: 国家自然科学基金项目(52167007) |
|
| Lightweight Insulator Defect Detection Algorithm Based on Improved YOLOv8 |
|
JI Yuxun, LI Handong
|
(College of Electrical Engineering, Guizhou University, Guiyang 550025, China)
Jiyx062930@126.com; xdli1@gzu.edu.cn
|
| Abstract: To address issues such as small target sizes and scattered distribution of insulator defects, this paper proposes a lightweight detection algorithm based on YOLOv8. By integrating the advantages of average pooling and max pooling through Adaptive Downsampling ( ADown) technology, the algorithm reduces sensitivity to local variations while improving parameter utilization efficiency. Leveraging the Efficient Long-range Attention Network (ELAN) and Cross Stage Partial Network (CSPNet), a lightweight backbone network RepGELAN is designed. This network optimizes convolutional stacking by drawing inspiration from RepVGG’s convolutional structure, achieving both lightweight design and efficient inference. The Bidirectional Feature Pyramid Network (BiFPN) is employed to enhance feature integration and boost small-target detection capability. Experimental results demonstrate a 28.88% reduction in parameter count, with mAP @50 increasing by 1.6% and mAP @95 by 0.7% , effectively balancing lightweight requirements and performance demands. |
| Keywords: YOLOv8 insulator defect detection lightweight ELAN CSPNet BiFPN |
