Obtaining clear underwater images is crucial for smart aquaculture, so it is necessary to repair degraded underwater images. Although underwater image restoration techniques have achieved remarkable results in recent years, the scarcity of labeled data poses a significant challenge to continued advancement. It is well known that semi-supervised learning can make use of unlabeled data. In this study, we proposed a semi-supervised underwater image enhancement method, MCR-UIE, which utilized multimodal contrastive learning and a dynamic quality reliability repository to leverage the unlabeled data during training. This approach used multimodal feature contrast regularization to prevent the overfitting of incorrect labels, and secondly, introduced a dynamic quality reliability repository to update the output as pseudo ground truth. The robustness and generalization of the model in pseudo-label generation and unlabeled data learning were improved. Extensive experiments conducted on the UIEB and LSUI datasets demonstrated that the proposed method consistently outperformed existing traditional and deep learning-based approaches in both quantitative and qualitative evaluations. Furthermore, its successful application to images captured from deep-sea cage aquaculture environments validated its practical value. These results indicated that MCR-UIE held strong potential for real-world deployment in intelligent monitoring and visual perception tasks in complex underwater scenarios.

Accurate underwater fish perception is a prerequisite for intelligent aquaculture, providing essential support for automated growth monitoring and precision feeding. Nevertheless, underwater imaging is severely affected by light scattering and cluttered backgrounds, making reliable fish perception and image-based non-contact fish size estimation particularly challenging. To overcome these limitations, this paper presents a geometry-aware underwater fish perception framework based on YOLOv5 for simultaneous object detection, pose estimation, and fish body size measurement. The proposed framework incorporates a Fish Geometry Coordinate Attention (FGCA) module into the backbone to enhance geometric feature representation by jointly modeling horizontal and vertical spatial dependencies, thereby improving the discrimination of elongated fish structures. In addition, a Dynamic Edge Relation Modeling (DERM) module is integrated into the neck to explicitly capture edge cues and structural relationships among key body regions, leading to more robust feature aggregation and more accurate key point localization under complex underwater conditions. Based on the predicted key points, the length and width of the fish can be estimated in a non-contact manner. Experimental results on a public underwater fish dataset demonstrate that the proposed framework achieves a Box mAP_0.5 of 93.07% and a Pose mAP_0.5 of 93.10%, consistently outperforming several representative YOLO-based detectors. Moreover, the proposed method maintains low body size estimation errors across different fish size categories. These results indicate that the proposed framework provides an accurate and practical solution for integrated underwater fish perception and image-based non-contact fish size estimation, offering significant potential for intelligent aquaculture and automated fish growth monitoring.