Abstract: Acquiring high-quality underwater crack images and achieving efficient identification and quantifi-cation are crucial for enhancing dam inspection efficiency. To address underwater image degradation and crack quantification challenges, this study proposes a deep learning and transfer learning-based method for underwater image enhancement and crack analysis. An underwater imaging platform was constructed for data acquisition, and a conditional diffusion model was developed using public marine image datasets as prior knowledge for cross-domain multi-source enhancement. Crack detection was performed using YOLOv12, followed by morphological operations for feature quantification. Ex-perimental results demonstrate that the proposed enhancement method significantly outperforms conventional approaches in visual quality, no-reference metrics, and pixel allocation. The integrated detection model improves accuracy while reducing missed detections, and the quantification method effectively extracts crack parameters. The developed "enhancement- identification-quantification" closed-loop framework provides an effective technical solution for intelligent underwater inspections.