Salmon in the Loop

The Gradient · 2 years ago

A consultant describes using computer vision and machine learning to automate fish counting at hydroelectric dams, which is required by federal regulators to monitor impacts on endangered salmon populations. Hydropower utilities typically seek systems that achieve 95% accuracy compared to manual visual counts, and the process involves defining problem scope, establishing performance goals, collecting labeled training data, and selecting appropriate models while integrating human expert review of low-confidence cases. The implementation requires navigating both technical constraints and regulatory requirements, treating fish counting as a human-in-the-loop machine learning challenge that combines automated detection with biologist expertise.