Fast diagnosis of wind turbine blade images based on regional clipping and YOLO-Dehazenet

Abstract

Wind turbine blade fault detection under hazy conditions is a critical technology for ensuring the reliability of wind power systems. Current algorithms face challenges in achieving a balance between dehazing effects and real-time processing requirements, especially in resource-constrained environments like those on drone-mounted platforms, presenting a bottleneck that needs to be addressed. Therefore, this article proposes an efficient real-time regional clipping image dehazing method (YOLO-Dehazenet). This method comprises two components: a large target detection model (Cut-YOLO) and an image dehazing model (Fast-Dehazenet). First, the Cut-YOLO model performs global precise localization of wind turbine blades in infrared images and maps these locations onto visible light images to achieve target area cropping. Subsequently, Fast-Dehazenet performs efficient dehazing on the cropped local images. Extensive experiments demonstrate that the proposed method outperforms existing technologies in dehazing performance, achieving improvements of 0.5–2 and 0.2–1.8% on the peak signal-to-noise ratio and structural similarity index metrics, respectively. Meanwhile, the inference time for a single image is just 14.1 ms. Additionally, this method demonstrates effectiveness in the field of wind turbine blade fault detection. Notably, it remains stable in resource-constrained environments, offering robust support for industrial deployment. Furthermore, the proposed image cropping technique is applicable to various dehazing models, demonstrating its broad utility.