(Infographics and detailed information coming soon!)
Using CV to Monitor Fish
This project has been designed to assist farmers in monitoring their tanks for behaviours commonly associated with stress, like balling and surface-seeking. These behaviours can be seen by CCTV cameras mounted above individual tanks. Using footage from these cameras, I created and trained a computer vision (CV) model that is to be deployed on the farm soon* as part of a bespoke full-stack solution.
*( once I finished writing my thesis and have time to fly to the farm...early 2026)
My work in this project encompassed data annotation, model development and training, experimentation/performance evaluation and local deployment on native systems. Development is completed, the model awaits local deployment and will be continously supported after deployment.
Accessibility & HCAI
This model has been designed with accessibility and human-centered AI principles:
- It utilizes existing farm infrastructure, resulting in no hardware cost.
- Similarly, it has been designed to work with the existing camera software. This causes minimal disruption to existing workflows and avoids having to retrain workers or lose previous functionality.
- The users/client have been involved in the design process.
- This model is not replacing human work, but supplementing it.
- The architecture of this model has been kept intentionally simple and small, so that it can run locally (not on the cloud) and return predictions fast.
- Running the model locally allows for sensitive farm footage to remain secure and is not uploaded at any point.
- The model is supported and routinely updated with new data, allowing the model to keep learning and growing.
- The model returns information on why it made a decision (following XAI principles).