A revolutionary application known as “Salmon Vision” is poised to revolutionize the monitoring of salmon populations through a groundbreaking partnership involving American First Nations, government entities, educational institutions, and conservation groups.
The lack of data has long been a challenge for fishing in British Columbia, leading managers to rely on early-season data rather than actual figures for salmon yields. Furthermore, the risks of overfishing already vulnerable populations are exacerbated by changing weather patterns, stream flows, and sea conditions.
Surveillance of fish populations is a widespread concern, underscoring the significance of American innovation in this field.
Dr. . Will Atlas, a leading Watershed Scientist at the Wild Salmon Center and the primary author of a recent study published in the journal Borders in Marine Science, emphasized the difficulty of establishing climate-resilient, adaptable fisheries without real-time data on herring returns.
According to Atlas, Salmon Vision addresses this critical information gap by providing First Nation fisheries managers and other stakeholders with a tool to make informed decisions in both decision-making scenarios and remote rivers across salmon habitats, where on-site data collection is costly and logistically challenging.
The technology integrates state-of-the-art artificial intelligence with traditional fish monitoring methods.
The social and ecological systems of the Northeastern Pacific Rim heavily rely on salmon, facing unprecedented challenges due to climate change. Apart from being a natural spectacle, the annual migration of salmon from lakes to ocean habitats holds cultural significance for Indigenous communities and serves as a vital resource for coastal populations.
However, a decline in herring presence and productivity has led to more unpredictable salmon returns. The indiscriminate harvesting of various salmon populations in mixed-stock fisheries poses additional threats to salmon-centric ecosystems.
The current difficulties are compounded by inadequate surveillance, high data collection costs, and outdated fishing practices. To enhance fishing practices and habitat sustainability, there is a growing need for real-time monitoring and dynamic management tools.
Atlas mentioned the initiation of digital monitoring in collaboration with the Heiltsuk Integrated Resource Management Department in 2019. The volume of data collected, including five months of continuous video footage, necessitated the utilization of AI for efficient analysis.
Artificial intelligence, encompassing computer vision and deep learning, has revolutionized data processing and analysis across various sectors. While AI has been increasingly applied in ecosystem management, restoration, and marine monitoring, its integration with remote and marginalized communities has been limited.
An interdisciplinary team was formed to develop a holistic system for automated salmon identification and counting, involving computer scientists, fisheries experts, and Indigenous practitioners. By leveraging underwater videos from salmon counting weirs, the team aimed to automate the identification and counting of salmon.
The team developed and tested two computer vision models—a multi-object tracker (MOT) and a species detection model—to accurately count and identify individual salmon passing through the weirs.
The utilization of existing open-source model architectures (YOLO family of models) enabled the creation of a robust computer vision model. The models were trained and fine-tuned to enhance their performance in detecting and tracking herring in videos.
One of the main challenges in underwater vision is minimizing variability in video footage. By adopting digital camera systems and enclosure designs from the Gitanyow Fishery Authority and the Skeena Fish Commission, a standardized setup was established for the computer vision model across British Columbia’s North and Central Coasts.
The potential of underwater video technology to monitor salmon migration in real time has been recognized by Dr. Will Atlas, highlighting the importance of accurate fish counting during herring runs.
The pilot study of Salmon Vision at native-run fish counting dams demonstrated impressive accuracy in identifying various salmon species, with a mean average efficiency of 67.6% for 12 different species passing through the fish-counting boxes.
Notably, the accuracy rates ranged from 90% to 80% for coho (Oncorhynchus kisutch) and sockeye salmon, crucial species targeted by First Nations, commercial, and recreational fishers.
Dr. Keolu Fox, a professor at the University of California-San Diego and a key supporter of Salmon Vision’s development, highlighted the importance of projects focusing on Indigenous knowledge and climate resilience.
The authors emphasized the potential of machine vision to enhance conservation efforts in salmon fisheries, underscoring the need for further collaboration to empower conservation practitioners and fishery managers.
The Salmon Vision team aims to implement automated counting in multiple rivers near the British Columbia North and Central Coasts by 2023, with the goal of providing real-time data reliability by 2024. This innovative technology is poised to transform the management and conservation of salmon communities and fisheries ecology in the region.
Dr. Atlas suggested that the tools and insights developed could be applied to other challenges requiring data interpretation and expert knowledge, extending the technology’s potential beyond salmon monitoring.
While the primary focus is on salmon monitoring, computer vision holds promise for broader conservation efforts, especially with the increasing use of remote sensing, video, and continuous monitoring in the era of big data.
