YOLOv8 improved network structure diagram.

In recent years, the demand for efficient and accurate defect detection algorithms in industrial production has been increasing. However, industrial cameras may be affected by water fog during image acquisition, resulting in image blurring and quality degradation, which increases the difficulty of defect detection. This paper proposes an industrial defect detection algorithm incorporating dehazing technology to enhance detection performance in complex environments. Experimental results show that using an optimized dehazing processing method on industrial images affected by water fog achieves an average PSNR of 34.9 dB and an SSIM of 0.951. The overall performance surpasses CNN and MADNet models, and verification using the improved YOLOv8 model significantly enhances defect detection confidence while greatly reducing missed detections. Further research indicates that this method is not only applicable to industrial defect detection but can also be transferred to personnel localization and rescue tasks in fire and smoke environments. This study provides a novel technical approach for industrial defect detection in complex environments and offers valuable references for image processing and object detection tasks in other fields.