7.YOLO-DCS:An improved method for detecting surface defectson wind turbine blades based on YOLOv11n

Abstract:

To address the issues of weak feature representation for wind turbine blade targets,complex backgrounds,and missed detections of small objects,this paper proposes an improved YOLOv11nbased detection algorithm named YOLODCS. In the backbone,adiverse branch block (DBB)and crosslevel channel attention (CLCA)are introduced to enhance key feature representation,whereDBB increases mAP@ 0. 5 by 1. 6% and CLCA improves mAP@ 0. 5 - 0. 95 by 3. 0% . Furthermore,a C2PSA_CLCA module is constructed by integrating the local aggregation mechanism of C2PSA with the global attention of CLCA,and thereby the multiscale featurelearning is strengthened,leading to an improvement of 8. 0% in smallobject mAP. In the neck,a SlimNeck architecture with GSConvand VoVGSCSP is adopted to improve feature fusion,resulting in an improvement of 5. 7% in overall mAP@ 0. 5 - 0. 95 with only additional 0. 4 M parameters. Experimental results show that YOLODCS attains 92. 9% recall,93. 4% precision,95. 7% mAP@ 0. 5,and 76. 8% mAP@ 0. 5 - 0. 95,outperforming the baseline YOLOv11n by 3. 7% ,2. 7% ,3. 5% ,and 5. 7% ,respectively,demonstrating the effectiveness of the proposed method.

Keywords: wind turbine blades;YOLOv11n;feature integration;missed detection of small targets