Salmon in the Loop: Computer Vision and Human-in-the-Loop Fish Counting at Hydroelectric Dams
A consultant shares experiences from fish counting at hydroelectric dams, a sociotechnical problem undergoing digital transformation. The article explains the regulatory context (FERC), the challenges of manual fish counts, and how computer vision with human-in-the-loop systems can automate and improve accuracy. It outlines steps for building such systems, including problem definition, performance goals, data collection, model selection, and monitoring.
The article, written by a consultant, describes the problem of fish counting at large hydroelectric dams, which is a regulated activity under the Federal Energy Regulatory Commission (FERC). Dam operators must demonstrate compliance with environmental regulations by conducting fish passage studies, primarily through visual counts by trained biologists. These manual counts are error-prone and labor-intensive due to difficult conditions. To improve efficiency and accuracy, organizations are exploring computer vision and machine learning, using human-in-the-loop systems that combine human expertise with algorithmic consistency. The author outlines the steps for building such a system: defining the problem space in negotiation with stakeholders, establishing performance goals (often a 95% accuracy threshold compared to human counts), collecting and labeling training data, selecting and fine-tuning a model (e.g., pretrained deep learning models), and monitoring system performance. The article highlights the complexities of the sociotechnical aspects, including regulatory compliance, data quality, and trust.
- Сокращения
- FERC = Federal Energy Regulatory Commission — Федеральная комиссия по регулированию энергетики
Source: The Gradient —
original