计算机工程与应用2026，Vol.62Issue(17)：363-377,15.DOI:10.3778/j.issn.1002-8331.2507-0396
融合小波优化与分组协同特征的风机表面轻量化检测
Lightweight Wind Turbine Surface Defect Detection with Wavelet-Optimized Collaborative Group Feature Fusion
摘要
Abstract
Addressing issues of high model complexity,insufficient context capture,and low precision in detecting minute defects for wind turbine surface defect inspection,this paper proposes SWCM-DETR,a lightweight improved RT-DETR model.Firstly,a redesigned StarNet backbone network(StarNet-R)is constructed to reduce parameter count and enhance feature discriminability through architectural reconstruction.Subsequently,a wavelet-based feature preservation mecha-nism is established,achieving structural maintenance and anti-aliasing during feature scaling via Haar wavelet transforma-tion.Furthermore,a grouped-cascaded attention architecture is constructed to strengthen the perception capability for minute defects through multi-scale feature interaction.Finally,a Wise-Inner-MPDIoU(WIMPDIoU)hybrid loss function is pro-posed,integrating dynamic focusing and geometric constraints to optimize bounding box regression accuracy.Experimental results on a wind turbine surface defect dataset show that SWCM-DETR achieves an mAP50 of 89.83%,a 3.6 percentage-points improvement over the baseline model(86.23%),and the model parameters decrease by 44.9%and computations re-duce by 48.2%.This model effectively enhances wind turbine surface defect inspection accuracy and efficiency.关键词
缺陷检测/轻量化检测/小波池化/级联分组注意力Key words
defect detection/lightweight detection/wavelet pooling/cascaded group attention分类
信息技术与安全科学引用本文复制引用
刘明龙,蒋廷耀,王敏..融合小波优化与分组协同特征的风机表面轻量化检测[J].计算机工程与应用,2026,62(17):363-377,15.基金项目
国家自然科学基金(42406173). （42406173）
