WTD-YOLO: an improved wind turbine defect detection algorithm

To address the challenges in wind turbine defect detection tasks, specifically the significant feature disparities between structural and non-structural defects and the insufficient detection accuracy of existing models, this paper proposes a targeted wind turbine defect detection algorithm, WTD-YOLO. This defect detection algorithm is built upon YOLO v11 and enhances detection accuracy through improvements in both the Backbone and Neck architectures. In the Backbone component, a C2DAT module based on deformable attention mechanism is designed and integrated to improve feature perception of non-structural defect sets through dynamic attention weight allocation. In the Neck component, an HSCFPN structure is proposed, incorporating high-resolution feature information and recalibration mechanisms to achieve multi-scale feature aggregation and precisely capture fine-grained features of structural defects. Experimentals on wind turbine defect detection datasets demonstrate that WTD-YOLO outperforms other conventional object detection algorithms in terms of detection performance.