Research on surface loss detection algorithm of wind turbine blade based on FRE-DETR network

Abstract

Wind turbine generators operate in harsh areas for a long time, resulting in frequent problems such as blade breakage, and traditional blade defect detection methods have low detection accuracy. In this paper, an end-toend target detection algorithm FRE-DETR based on wind turbine blade defects is designed, and the detection speed and detection accuracy of the end-to-end detection model are further improved by redesigning the feature extraction location in the backbone network and proposing a feature selection and fusion module. FRE-DETR is tested on a wind turbine blade defect dataset, and the results show that the model improves the detection accuracy by 2% compared with RTDETR-R18. The inference speed is already higher than RTDETR-R18 when the step size is larger than 2. The Gflops of the model is only 66.8% of that of RTDETR-R18, which also greatly reduces the computational requirements of the hardware when deployed. FRE-DETR meets the requirements of real-time detection.

Keywords

• integrated energy system
• end-to-end algorithm
• wind turbine blade defect
• target detect; computer vision