Abstract: To systematically review the application of computer vision in the field of aquaculture,this paper provides an in- depth analysis of its current implementations and challenges across various stages of the farming process,while also offering insights into future development trends. The aim is to provide theoretical support and technical references for the intelligent transformation and upgrading of aquaculture. This study focuses on the specific application pathways and performance of visual recognition algorithms-such as convolutional neural networks and the YOLO series in aquaculture. It also elaborates on the advantages and development potential of multi-modal fusion algorithms in integrating visual images,acoustic signals,and water quality monitoring data. Existing research demonstrates that computer vision technologies can significantly enhance the precision management and production efficiency of aquaculture operations. Multi-modal fusion algorithms,in particular,have shown outstanding performance in key tasks such as fish behavior recognition and quantitative analysis of feeding intensity. However,computer vision algorithms still face challenges in practical applications,including poor image quality caused by complex underwater imaging environments and increased recognition difficulty due to diverse fish behavior patterns. Looking ahead,with the optimization of deep learning algorithms,further application of multi -modal fusion technology,and cross - disciplinary integration with technologies such as the Internet of Things and aquaculture robotics,computer vision is expected to provide critical technical support for the efficient, precise, and sustainable development of aquaculture. This will play a significant role in ensuring global aquatic product supply and food security.

Key words: computer vision, aquaculture, algorithm application, multimodal fusion, intelligent aquaculture

摘要： 为系统梳理计算机视觉在水产养殖领域的应用现状 ,深入 分析了计算机视觉在养殖全流程中的应用现状与现存挑战 ,并对未来发展趋势进行展望 ,以期为水产养殖的智能化转型升级提供理论支持与技术参考 。本研究重点围绕卷积神经网络、YOLO 系列算法等视觉识别算法在水产养殖中的具体应用路径与性能表现展开探讨 ,同时详细阐述了多模态融合算法在整合视觉图像、声学信号及水质监测数据等方面的优势与发展潜力 。现有研究表明 ,计算机视觉技术可显著提升水产养殖的精准化管理水平与生产效率 ; 多模态融合算法在鱼类行为识别精度、摄食强度量化分析等关键任务中表现尤为突出 。然而 ,计算机视觉算法在实际应用中仍面临水下成像环境复杂导致图像质量不佳、鱼类行为模式多样增加识别难度等问题 。未来 ,随着深度学习算法的优化、多模态融合技术的深入应用 ,以及与物联网、养殖机器人等技术的跨领域协同融合 ,计算机视觉技术将为水产养殖业的高效化、精准化、绿色可持续发展提供关键技术支持 ,对保障全球水产品供应与粮食安全具有重要意义。

关键词: 计算机视觉, 水产养殖, 算法应用, 多模态融合, 智能化养殖