YOLO in Precision Aquaculture: A Decadal Bibliometric and Systematic Review of Applications, Architectural Adaptations, and Deployment Challenges

Computer vision and artificial intelligence technologies have been increasingly adopted in recent years to support operational and management decisions in aquaculture. Among the various convolutional neural network (CNN) frameworks investigated for these applications, You Only Look Once (YOLO) models have gained prominence for the balance between detection performance and inference speed, which is critical for real-time decision support. This study critically evaluates and synthesizes the applications and domain-specific advances of the YOLO architecture in aquaculture through a decadal (2015–2025) bibliometric and systematic literature review. An extensive Scopus search yielded 201 documents, of which 181 were retained for bibliometric analysis and 48 for systematic review. Across six identified application domains, specifically animal detection and tracking, health assessment, behavior monitoring, precision feeding, biomass monitoring, and system inspection, the aquaculture-specific optimization of YOLO models and their real-world deployment challenges were critically examined. Our synthesis reveals that while architectural refinements, such as integrating attention mechanisms, lightweight backbones, redesigned loss functions, and transformer modules, have improved detection performance under controlled conditions, a critical gap persists in their real-world validation. Most reviewed models were developed and tested in small, well-controlled laboratory tanks that do not reflect the turbidity, stocking density, and optical complexity of commercial-scale farms. The lack of diverse open-source benchmark datasets and inconsistent reporting of deployment metrics further compound this gap, representing the most critical barriers to wider adoption of YOLO-based aquaculture monitoring. Future efforts should therefore prioritize the development of diverse, commercially representative datasets and rigorous real-world validation frameworks to advance YOLO-based aquaculture monitoring. • Six key aquaculture application domains were identified and systematically reviewed • YOLOv5 and YOLOv8 were the most widely adopted versions • Attention mechanisms and lightweight backbones were the dominant modification strategies • Only 16.7% of studies validated models under commercial/semi-commercial-scale aquaculture conditions • Real-world validation and multi-task architectures emerged as the most critical priorities