Towards Non-invasive Fish Monitoring in Hard-to-Access Habitats Using Autonomous Underwater Vehicles and Machine Learning∗

This paper presents a concept for non-invasive, high spatio-temporal resolution fish monitoring. Autonomous underwater vehicles combined with artificial intelligence enable automatic habitat mapping and fish detection from sonar and camera data. The monitoring approach will help to fill important knowledge gaps on target fish spatio-temporal distribution in hard-to-access areas, give valuable insight into target fish behavior, and help to identify the species' essential habitats, which is relevant for the design of marine protected areas in fish management and conservation. Many of the required hardware, such as underwater vehicles, sensors, and acoustic modems, have become very cost-effective over recent years, making this approach feasible. Unclear, however, remains the question of how to detect fish using images obtained from low-cost camera and sonar devices. Therefore, we have reviewed and discussed suitable machine learning techniques for this task.

Options

Towards Non-invasive Fish Monitoring in Hard-to-Access Habitats Using Autonomous Underwater Vehicles and Machine Learning∗