In industrial manufacturing, it is critical to perform accurate defect detection in order to maintain product quality and ensure work safety. Machine vision technology enables defect detection with high accuracy due to intelligent image processing algorithms. However, the clarity of the captured image affects the detection results. Due to capture equipment or cost constraints, images may be blurry or of low resolution, making accurate defect detection difficult. To overcome this challenge, super-resolution technology is adopted to improve image quality and thus optimize defect detection results. Reference-based super resolution model was used to improve the quality of low-resolution metal surface images. The super-resolution images generated were then processed using the modified YOLO model to realize the deep learning-based surface defect detection. The NEU-DET dataset was used for experimental validation. The experimental results show that the integration of super-resolution and YOLO greatly improves detection accuracy. This approach demonstrates the higher PSNR and SSIM of combining super-resolution and advanced detection models in industrial quality control applications.