| 薛阳*,蔡畅*,卢秋红**,徐笑*.轻量化改进型YOLOv8的多类别绝缘子缺陷检测[J].高技术通讯(中文),2025,35(9):933~942 |
| 轻量化改进型YOLOv8的多类别绝缘子缺陷检测 |
| Multi-category insulator defect detection for lightweight improved YOLOv8 |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 09. 002 |
| 中文关键词: 智能巡检； 缺陷检测； YOLOv8网络； 特征融合； 非对称检测头 |
| 英文关键词: intelligent inspection, defect detection, YOLOv8 (you only look once version 8) network, feature fusion, asymmetric detection head |
| 基金项目: |
| 作者 | 单位 | | 薛阳* | (*上海电力大学自动化工程学院上海 200090)
(**上海合时智能科技有限公司上海 200040) | | 蔡畅* | | | 卢秋红** | | | 徐笑* | |
|
| 摘要点击次数: 15 |
| 全文下载次数: 13 |
| 中文摘要: |
| 为推动输电线路智能化巡检模式，本文针对人机协同巡检模式下的图像差异大及干扰因素多等问题，提出一种轻量化改进型YOLOv8(you only look once version 8)的多类别绝缘子缺陷检测算法。首先在特征提取网络中融合可变形大核注意力的同时进行轻量化，提升网络对不同目标轮廓与尺寸的适用性；其次引入渐进的特征融合策略以改善不同层次特征间的语义差距，提高网络的检测精度；并设计轻量化非对称检测头，进一步减少参数冗余；最后改进边框损失函数有效降低由密集遮挡造成的漏检和误检数量。实验结果表明，本文算法相较于原算法检测精度提升了7.7%，参数量和计算量分别减少了26.4%和30.2%，并在密集、遮挡、多类别目标缺陷检测中的评价指标均领先于当前主流的几类目标检测算法，显著提高了复杂环境下的多类别绝缘子缺陷检测，实现了检测精度和速度的双重提升。 |
| 英文摘要: |
| To promote the intelligent inspection mode of transmission lines, this paper proposes a lightweight and improved YOLOv8 (you only look once version 8) multi class insulator defect detection algorithm to address the problems of large image differences and multiple interference factors in the human-machine collaborative inspection mode. Firstly, in the feature extraction network, deformable large kernel attention is integrated while lightweighting to enhance the network’s applicability to different target contours and sizes; Secondly, a progressive feature fusion strategy is introduced to improve the semantic gap between features at different levels and enhance the detection accuracy of the network; And a lightweight asymmetric detection head is designed to further reduce parameter redundancy; Finally, improving the border loss function effectively reduces the number of missed and false detections caused by dense occlusion. The experimental results show that the detection accuracy of the proposed algorithm has improved by 7.7% compared to the original algorithm, with a reduction of 26.4% in parameter count and 30.2% in computation. Furthermore, evaluation metrics of the proposed method for dense, occluded, and multi class object defect detection are ahead of several mainstream object detection algorithms, significantly improving the detection of multi class insulator defects in complex environments and achieving a dual improvement in detection accuracy and speed. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
