Keypoint-Based Behavior Detection and Individual Identification in Nile Tilapia

Aquaculture needs scalable, noninvasive tools that can see what farmers miss. This project will build an AI system that recognizes individual Nile tilapia and flags behavior shifts signaling stress or disease. Using YOLO for fast detection and a key-point model for anatomical landmarks, we will extract precise coordinates for the snout, operculum, fin bases, and tips. From these points, we compute distances, angles, and motion cues to create stable biometric signatures and behavior indicators robust to growth and lighting changes. Data will be captured in Brazil with high-resolution imaging under controlled conditions and rigorously annotated; modeling and validation will be led at Dalhousie in Canada, in close collaboration with UNESP. The outcome is a noninvasive, scalable pipeline that enables longitudinal identification, early-warning dashboards, and reproducible welfare analytics. Expected benefits include reduced manual handling, faster health interventions, improved feed conversion, and stronger selective-breeding programs through objective phenotypes. The project will deliver a curated dataset, open methods, and a proof-of-concept tool ready for farm pilots, positioning Dalhousie and UNESP as leaders in digital aquaculture and strengthening CanadaBrazil research ties. By pairing computer vision with aquaculture science, we aim to set a new benchmark for precision husbandry and sustainable fish production at scale.

Faculty Supervisor:

Suresh Raja Neethirajan

Student:

Partner:

Universidade Estadual Paulista "Julio de Mesquita Filho"

Discipline:

Life Sciences

Sector:

Aquaculture and Fishing

University:

Dalhousie University

Program: