|
Sexton trawl camera used to collect video footage inside pollock trawls to test the effectiveness of bycatch reduction devices in nets. Credit: NOAA F
Artificial Intelligence Accelerates Data Analysis to Reduce Incidental Salmon Catch in the Pollock Fishery
UNITED STATES
Thursday, October 09, 2025, 00:10 (GMT + 9)
Scientists at the Alaska Fisheries Science Center are using a deep learning model (YOLOv11) to detect pollock and salmon in fishing net videos, enhancing the efficiency and sustainability of the fishery.
JUNEAU, ALASKA – In a significant step forward for sustainable fisheries management, scientists from NOAA Fisheries' Alaska Fisheries Science Center have developed a method to analyze video data at unprecedented speed, thanks to Artificial Intelligence (AI). This achievement is crucial for evaluating the effectiveness of exclusion devices designed to reduce the incidental catch, or bycatch, of salmon in the lucrative pollock fishery.
The research team employed an advanced form of AI known as deep learning, utilizing the You Only Look Once, version 11 (YOLOv11) object detection model. This model, designed for object detection and classification in images, was customized to specifically identify pollock and salmon in videos taken inside fishing nets.
Example image of the salmon excluder that is placed in the last taper section of the net with the camera placement (white box) and the approximate field of view (dashed triangle) shown. The diameter of the net is approximately 2 m at the beginning of the excluder. Credit: NOAA Fisheries.
YOLOv11: A Leap in Efficiency
YOLOv11’s capability allows scientists to semi-automate the video review process, which previously relied entirely on human effort. This not only accelerates the evaluation of the effectiveness of bycatch reduction devices (known as salmon excluders) but also facilitates the observation of fish behavior to improve these devices' performance.
 Katherine C.Wilson, a physical scientist and lead of the study at NOAA Fisheries' Alaska Fisheries Science Center, highlighted the new system's efficiency:
"We were able to train a publicly available deep learning object detection model to identify salmon and pollock. The model compared well to human performance, with some variability. And the models save us a ton of time. They can process fishing tows in a matter of hours; humans would need days or weeks to review the same data."
The Need for AI Against the Bycatch Challenge
The pollock fishery has faced significant challenges in the past, with record-high levels of incidental catch of Chinook salmon in 2005 and Chum salmon in 2006. Since then, the commercial fishing industry, in collaboration with scientists and engineers, has invested in:
Examples of fish detections from three frames for a model trained to detect both pollock and salmon (top, multi-class model) and a model trained to only detect salmon (bottom, salmon only model). Salmon (white) and pollock (grey) detections are shown with the confidence score of the detection as a percentage at the top of the boxes. False positive salmon detections are present for herring in (B, bottom-left), (C), and (F) and salmon is missed in (D) and (E). Credit: NOAA Fisheries.
These tools aim to prevent salmon capture from the outset. Salmon exclusion devices are a key part of this strategy, allowing salmon to escape the net while pollock remain inside. To monitor their effectiveness, cameras with lighting are positioned at the entry point of these devices, recording videos that scientists then review.
The increase in video collection, thanks to the availability of low-cost, high-quality camera systems, has generated a massive amount of data. This is where machine learning becomes indispensable, freeing scientists from the tedious task of reviewing countless hours of footage.
Collaboration and Future Perspectives
The study was a collaborative effort with the Pacific States Marine Fisheries Commission, and data were collected by a B&N Fisheries contracted vessel..
Wilson added: “There are improvements to be made, but this work is promising. It was challenging for people familiar with salmon identification to review so many hours of video and identify every salmon. Automated long-term video monitoring will be more cost-effective and reduce human error.”
The trained YOLOv11 model detected 97% of the fish with an 82% prediction accuracy. Although its performance varied under different conditions (such as the presence of krill, fish density, or low light), the ability to identify multiple species, such as herring, could minimize classification errors.
Fishing net towing behind a vessel. Credit: NOAA Fisheries
Scientists plan to continue working to optimize these models, especially by testing them on videos with a higher presence of Chinook salmon, including smaller individuals, and exploring the capacity to distinguish between Chum and Chinook salmon.
Wilson concluded that the automation of video analysis has the potential to “save time and money,” providing the fishing industry with additional tools to reduce bycatch and move toward greater sustainability. "Our salmon and pollock detection work demonstrated that deep learning methods are accessible and robust," she affirmed.
editorial@seafood.media
www.seafood.media
|
Print
