KAL-YOLO: An Algorithm for Efficient Detection of Wind Turbine Blade Defects

With the development of artificial intelligence-related technologies, the use of deep learning techniques for wind turbine blade defect detection can not only significantly save labor costs, but also improve efficiency. However, limited by the detection scenarios, the detection performance is poor in the face of problems such as small damage areas, complex morphology, low contrast and changes in ambient light.To address this problem, this paper proposes the KAL-YOLO detection algorithm based on YOLOv8s, which introduces the KAN_C2f module into the backbone network, then uses the ADown downsampling module, and finally uses the LADH detection head, and the improved model achieves a precision rate of 87.6%, a recall rate of 86.7%, and a mAP@50 of 87.2% on the wind turbine blade defect test set, mAP@50:95 reaches 59.8%, which is 1.0%, 3.4%, 2.0% and 2.0% higher than the baseline model respectively, and the computation amount is reduced by 33.10%, which improves the detection precision and guarantees the model computation performance at the same time.