EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Multi-Scale Attention-Enhanced YOLO26 for Wind Power Equipment Surface Defect Detection

Longlong Pan, Wei Sun, Panfeng Ba, Ying Wang

Applied Sciences · 02.11.2026 · 16 (21) · 10861

DOI: 10.3390/app162110861 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 02.11.2026. Verfasst von Longlong Pan und 3 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Research on an Intelligent Monitoring and Early-Warning System for Wind Turbine Hoisting Sites.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Surface defects in wind power equipment exhibit substantial variations in scale, morphology, texture, and orientation and are frequently affected by complex backgrounds, illumination changes, and surface contamination, posing challenges to reliable visual inspection. This study proposes a YOLO26-based surface defect detection framework that integrates multi-scale contextual modeling, spatial–channel collaborative attention, and directional feature recalibration. Specifically, the original SPPF module is replaced with SPPSCCAP to combine multi-scale pooling, spatial–channel reconstruction, and global contextual information. CADP-SCSA is subsequently introduced to jointly model directional spatial information and inter-channel dependencies, while ELA-B modules are embedded before the P3, P4, and P5 detection heads to preserve horizontal and vertical positional information. Experiments were conducted on a public wind power equipment dataset containing 9583 images and 22,856 annotated instances across seven defect categories. Under a unified training and testing protocol, comprehensive ablation experiments demonstrate that the effects of the three enhancement mechanisms are not simply additive. Among the evaluated configurations, the three-module Model H achieved the highest values across the four reported detection metrics, with a precision of 0.9120, a recall of 0.8860, an mAP50 of 0.9060, and an mAP50–95 of 0.6660. Compared with the baseline model, these results represent improvements of 5.57, 8.99, 8.63, and 9.44 percentage points, respectively. These findings demonstrate the importance of layer-specific feature enhancement and module coupling in multi-class defect detection and provide a practical basis for vision-based inspection and condition assessment of wind power equipment.

Förderung

  • Research on an Intelligent Monitoring and Early-Warning System for Wind Turbine Hoisting Sites

Zitieren

APA

Pan, L., Sun, W., Ba, P., & Wang, Y. (2026). Multi-Scale Attention-Enhanced YOLO26 for Wind Power Equipment Surface Defect Detection. Applied Sciences, 16(21), 10861. https://doi.org/10.3390/app162110861

Vancouver

Pan L, Sun W, Ba P, Wang Y. Multi-Scale Attention-Enhanced YOLO26 for Wind Power Equipment Surface Defect Detection. Applied Sciences. 2026;16(21):10861. doi:10.3390/app162110861

BibTeX
@article{Pan2026,
  author = {Pan, Longlong and Sun, Wei and Ba, Panfeng and Wang, Ying},
  title = {Multi-Scale Attention-Enhanced YOLO26 for Wind Power Equipment Surface Defect Detection},
  journal = {Applied Sciences},
  year = {2026},
  volume = {16},
  number = {21},
  pages = {10861},
  doi = {10.3390/app162110861}
}
Kostenlos starten →