机电工程技术2025，Vol.54Issue(6)：16-22,7.DOI:10.3969/j.issn.1009-9492.2024.00191
融合多重注意力及高阶特征的绝缘子缺陷检测方法
Insulator Defect Detection Method Based on Multi-attention and High-order Context Information
摘要
Abstract
To address the issue of dispersed target distribution and inconspicuous features in insulator defect image detection,as well as low detection accuracy,an improved YOLOv8-based insulator and defect detection algorithm model,HIEDet(Highly Efficient Insulator and Defect Detector),is proposed.Firstly,a backbone feature extraction network is designed,employing double-layer routing attention mechanisms in areas with larger feature maps to maintain global modeling capabilities,and efficient gradient flow multi-scale attention in areas with smaller feature maps.The approach captures global and local spatial dependencies through grouping and parallel sub-networks operating at multiple scales,thereby enhancing the detection performance of the network model.Then,a high-order joint feature distribution mechanism network is designed to alleviate the problem of feature information loss caused by the PANet structure,while simultaneously strengthening the model's ability to fuse multi-scale features.Finally,SIoU is adopted as the regression loss function for the HIEDet network,aiding in the convergence and effectiveness of the training process,and improving the model's detection performance on insulator defect image detection tasks.Experimental results validate that the improved HIEDet model increases mAP@50 by 2.2%and mAP by 1.6%on the insulator defect dataset compared to YOLOv8,indicating the effectiveness of the proposed HIEDet model in insulator defect detection tasks.关键词
绝缘子/多重注意力机制/高阶特征融合/目标检测/YOLOv8Key words
insulator/multi-attention mechanism/high-order feature fusion/object detection/YOLOv8分类
计算机与自动化引用本文复制引用
严锴,曾子豪,邓文娟,汪志成,黄奖华,周书民..融合多重注意力及高阶特征的绝缘子缺陷检测方法[J].机电工程技术,2025,54(6):16-22,7.基金项目
江西省自然科学基金重点项目(20232ACB202004) （20232ACB202004）
江西省技术创新引导类项目(20212BDH80008) （20212BDH80008）
