Advancing Non‐Contact Fish Phenotyping via Optical Sensing: Toward Precision Aquaculture

ABSTRACT Non‐contact fish phenotyping has become an essential strategy for precision aquaculture, enabling accurate, high‐throughput, and non‐invasive trait measurement. In this review, a digital phenotyping framework is summarized, including application‐driven trait definition, multimodal data acquisition (2D, 2.5D, and 3D imaging), and automated phenotypic analysis based on computer vision and deep learning. According to current research paradigms, optical phenotyping applications are systematically categorized into three major domains: morphological, behavioral, and appearance‐based phenotyping. Multimodal optical sensing systems now support high‐throughput morphological, behavioral, and appearance‐based phenotyping, enabling applications ranging from biomass estimation and intelligent feeding control to selective breeding, product quality evaluation, species identification, and disease diagnosis. Despite rapid methodological progress, most existing studies remain confined to laboratory or semi‐controlled environments, and large‐scale implementation in commercial production systems is still limited. Key challenges persist in underwater image degradation, posture‐induced measurement uncertainty, data heterogeneity, model generalizability, computational constraints, and the integration of phenotypic and genomic data for precision breeding. Finally, future development pathways are discussed, including efficient and flexible data acquisition platforms, lightweight and real‐time processing, transfer and few‐shot learning, multimodal data fusion, novel phenotype quantification, and standardized open datasets. Based on a comprehensive analysis of over 100 studies, this review maps the current technological landscape and proposes pathways toward scalable, intelligent phenotyping systems that can improve fish welfare, enhance breeding efficiency, and promote sustainable aquaculture management.