AquaPulse: Robust Computer Vision and Uncertainty Estimation for Aquatic Ecosystems
Parsa Besharat
Department of Math and Computer Science, Technische Universität Bergakademie Freiberg, Freiberg, Germany
parsabe99@gmail.comAbstract
AquaPulse bridges multi-model YOLO neural vision, BotSORT multi-object tracking, Ensemble Kalman Filtering stochastic data assimilation, and local generative AI into an end-to-end ecosystem telemetry platform. By eliminating manual fish counting and enabling non-invasive automated observation in turbid aquatic environments, AquaPulse provides researchers, marine biologists, and environmental agencies with real-time ecological safety monitoring, extinction risk forecasting, and publication-ready LaTeX scientific reports.
Key Contributions & Architecture
- Multi-Model YOLO Vision Engine: High-throughput object detection tuned for underwater turbidity and specimen detection.
- BotSORT Multi-Object Tracking: Continuous tracking with robust re-identification across occlusions and dynamic aquatic currents.
- Ensemble Kalman Filtering (EnKF): Stochastic data assimilation and state space modeling for real-time uncertainty estimation.
- Automated Scientific Reporting: Integrated local generative AI dialogue and telemetry engines producing publication-ready LaTeX documents and telemetry plots.