计算机视觉在水产养殖中的研究进展

Research Progress of Computer Vision in Aquaculture

摘要:

随着全球对水产蛋白需求的激增，从劳动密集型养殖向精准水产养殖转型已成为保障可持续粮食安全的战略必然。计算机视觉（CV）作为该领域的颠覆性技术，实现了对水产生物的非侵入式、高通量和智能化监测。本综述系统性地批判分析了 CV 在水产养殖中的进展，涵盖四大核心维度:生物参数测量（计数与生物量估算）、行为解读（摄食强度与应激分析）、健康诊断（病害与病灶检测）以及环境监测。我们剖析了从传统图像处理到前沿深度学习范式（如 CNN、Transformer）的技术演进，并评估了端到端智能流水线的设计。至关重要的是，本文不仅限于罗列算法，更深入探讨了未解决的学术争议，包括二维单目与三维立体视觉之间的权衡、模型复杂性与实时边缘部署之间的冲突，以及数据驱动方法与生物学先验知识的融合。此外，我们识别了限制产业化应用的关键瓶颈，如域泛化差距和水下成像退化问题。最后，本文提出了未来的技术路线图，倡导采用基础模型（Foundation Models）、数字孪生（Digital Twins）以及云-边-端协同架构。本综述旨在引导研究人员和从业者实现全自主、面向动物福利且鲁棒的智能水产养殖系统。

Abstract:

With the global surge in demand for aquatic protein, the transition from labor-intensive farming to precision aquaculture has become a strategic imperative for ensuring sustainable food security. Computer vision (CV), as a disruptive technology in this field, enables non-invasive, high-throughput, and intelligent monitoring of aquatic organisms. This review systematically critiques CV advancements in aquaculture across four core dimensions: biological parameter measurement (counting and biomass estimation), behavioral interpretation (feeding intensity and stress analysis), health diagnostics (disease and lesion detection), and environmental monitoring. We dissect the technological evolution from traditional image processing to cutting-edge deep learning paradigms (e.g., CNNs, Transformers) and evaluate end-to-end intelligent pipeline designs. Crucially, this paper extends beyond algorithmic enumeration to delve into unresolved academic debates, including trade-offs between 2D monocular and 3D stereo vision, conflicts between model complexity and real-time edge deployment, and the integration of data-driven approaches with biological prior knowledge. Furthermore, we identify key bottlenecks constraining industrial applications, such as domain generalization gaps and underwater imaging degradation. Finally, this paper proposes a future technology roadmap advocating the adoption of Foundation Models, Digital Twins, and cloud-edge-device collaborative architectures. This review aims to guide researchers and practitioners toward achieving fully autonomous, animal-welfare-oriented, and robust intelligent aquaculture systems.