Abstract: The operating environment of wind turbine blades is complex, with strong winds, sand, and other factors affecting the quality of surface defect images. Additionally, due to issues with remote transmission or storage of equipment, image data is often subject to noise interference, which in turn affects the training of convolutional neural network models, leading to misjudgments and missed defect detections. This paper proposes a zero-shot surface defect denoising model for blades based on ZS-N2N. The model uses a simple two-layer network structure and, without the need for training data or noise distribution information, achieves denoising of defect images at a low computational cost. Experimental results show that the proposed model can effectively remove noise from defect images under various noise conditions.