A real-world underwater video dataset with environmental parameters in Recirculating Aquaponic Systems (RAS)

This dataset is designed as a training resource for fish tracking and localization models, specifically targeting tilapia behavior analysis in real-world Recirculating Aquaponic Systems (RAS). It consists of 31 curated video clips, each 30 seconds long, extracted from full-length recordings. The selection process prioritized high fish presence, ensuring that the dataset is highly relevant for computer vision-based behavior classification.
Capturing fish in real aquaculture conditions presents significant challenges for vision systems due to factors such as variable lighting, water turbidity, and fish density. This dataset bridges the gap between controlled laboratory settings and actual production environments, enabling the development of robust tracking models capable of handling the complexities of underwater vision in aquaculture.
The dataset also includes physicochemical parameters such as pH, temperature, dissolved oxygen, and turbidity. The recordings were conducted at Granja La Familia Tilapia in San Cristóbal, Querétaro, Mexico, ensuring real-world applicability for research and industry applications.