69412 2019-2020 Salmon Detection Data Set Published / External 59353 Conservation Engineering Research Project 2023-03-08 2025-09-25 This dataset contains the annotated data, object detection models, and video data used to fine-tune and evaluate object detection models for salmon and pollock detection in videos collected in a commercial pollock trawl in Alaska. The annotated data contains track annotations for all Pacific salmon (Oncorhynchus spp.) and walleye pollock (Gadus chalcogrammus) present and frame level annotations that describe the trawl and background conditions. The annotated data is composed of 11,572 salmon and 73,394 pollock annotations from 16,989 video frames that came from 184 video clips. Five-fold cross validation was used to evaluate EfficientDet D2 and YOLO11n object detection models, and this dataset includes all of these models and a single-class, salmon-only YOLO11n model. In addition to the annotated video frames that are provided as images, videos from three fishing tows that were not annotated are included with their respective YOLO11n detections. These three fishing tows were used to further evaluate the object detection performance of the best performing pollock and salmon model and the salmon-only model. Bycatch reduction devices (BRDs) are used in the Alaska walleye pollock (Gadus chalcogrammus) fishery to reduce Pacific salmon (Oncorhynchus spp.) bycatch. Evaluation of BRD effectiveness often requires people to process collected or live-feed video, which can be a tedious, expensive, and time-consuming task. Deep learning can be used to automate the detection and classification of fish in videos to support BRD and other fisheries bycatch work. This dataset was developed to evaluate the performance of widely-available, open-source pre-trained object detection models for the task of pollock and salmon detection. The Wilson et al. 2025 study found that the YOLO models performed better than EfficientDet and on average detected 90% of salmon and pollock with 72% accuracy using a 50% detection overlap threshold. Theme detection Theme pollock Theme salmon Theme tracking Theme trawls Theme video Alaska Fisheries Science Center Seattle WA USA RACE division, MACE program Data Set Imagery As Needed Video (digital) The dataset includes the subset of video clips that were collected in the commercial pollock fishery in Alaska and annotated for salmon and pollock, the object detection models that were trained and evaluated, all the video recorded for 3 fishing tows that were used to further evaluate model performance, and the model detections for these 3 fishing tows. F/V Pacific Explorer, David Bryan, Deborah Sharpe , Katherine Hellen-Schneider, Matthew Callahan, Connor Fox Author 2023 Person Wilson, Katherine katherine.wilson@noaa.gov 7600 Sand Point Way NE, Bldg. 4 Seattle FL 98115 United States 2065264474 Co-Author 2020 Organization Pacific States Marine Fisheries Commission PSMFC Lurbur, Moses <mlurbur@gmail.com> Co-Author 2019 Person Yochum, Noelle noelle.yochum@noaa.gov 7600 Sand Point Way NE, Bldg. 4 Seattle WA 98115 United States (206) 526-4432 (916) 719-5920 Point of Contact 2025 Person Wilson, Katherine katherine.wilson@noaa.gov 7600 Sand Point Way NE, Bldg. 4 Seattle FL 98115 United States 2065264474 All 26 tows were within this extent. -166 -164.5 55.33333333333 54.33333333333 Range 2019-06-01 2019-06-30 All 7 tows from Trip 1 were within this extent. -166.03722222222 -165.59805555556 54.69833333333 54.32472222222 Range 2020-08-07 2020-08-10 All 4 tows from trip 2 were within these bounds. -165.72888888889 -165.43055555556 54.78472222222 54.51527777778 Range 2020-08-13 2020-08-16 Standard CC0-1.0 https://creativecommons.org/publicdomain/zero/1.0/ These data were produced by NOAA and are not subject to copyright protection in the United States. NOAA waives any potential copyright and related rights in these data worldwide through the Creative Commons Zero 1.0 Universal Public Domain Dedication (CC0-1.0). Unclassified Video clip frames, annotations, object detection models, full tow videos and their detections can be accessed at https://console.cloud.google.com/storage/browser/nmfs_odp_afsc/RACE/MACE/salmon_pollock_object_detection. If using these data, please cite https://doi.org/10.1093/icesjms/fsaf168. https://console.cloud.google.com/storage/browser/nmfs_odp_afsc/RACE/MACE/salmon_pollock_object_detection/ 2019-2020 Salmon and Pollock Object Detection dataset This folder within the Alaska Fisheries Science Center's Google cloud storage contains the data and object detection models used for the Wilson et al. 2025 study on automated salmon and pollock detection in trawl videos from the Alaska commercial pollock fishery. The 'annotated_data' folder contains a folder for each of the 184 video clips used in the study that provides video frames as png files and the annotations as a custom COCO json file. The 'tow_data' folder contains videos as .mov files and csv files of fish detections for three fishing tows that were also used in the study to further evaluate object detection model performance. The 'models' folder contains the object detection models that were used in the study. The 'documentation' folder contains text files that provide details about the data and the cloud storage file structure. PNG - Portable Network Graphics Format <2MB Uncompressed https://console.cloud.google.com/storage/browser/nmfs_odp_afsc/RACE/MACE/salmon_pollock_object_detection Dataset Dataset that includes annotated data, models, and tow videos and detections used for the Wilson et al. 2025 study. https://doi.org/10.1093/icesjms/fsaf168 Automated fish detection in videos to support commercial fishing sustainability and innovation in the Alaska walleye pollock (Gadus chalcogrammus) trawl fishery Scientific publication https://github.com/noaa-afsc-mace/salmon_presence Salmon presence repo Source code for salmon presence prediction and evaluation. Video from within the trawl of the commercial pollock fishery in Alaska and the associated annotations of salmon and pollock. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. See Wilson et. 2025 publications for details. Annotations were done by trained personnel and reviewed by a lead scientist. Yes 0 No No Other The video portion of this project was collected onboard the FV Pacific Explorer by Noelle Yochum and Katherine Hellen-Schneider in 2019 and by Connor Fox in 2020 with Sexton housings equipped with Mobius cameras. Processing of video (identifying salmon and entering data) was conducted by David Bryan, Michelle Dyroy, Noelle Yochum, and Katherine Wilson. Video annotation (bounding box tracks of salmon and pollock) was done by Moses Lurbur, Deborah Sharpe, Katherine Hellen-Schneider, and Matthew Callahan. Annotation QA/QC was conducted by Katherine Wilson. Sexton Video Camera System instrument Video Recorder The Sexton video camera system consists of an underwater housing rated to 1000m, 2 LED lights, a Mobius2 Action Camera, and a power supply (either 2 NiMH 19.2v or 2 Li-Ion 14.4v batteries). The Mobius2 camera can record 1800 HD video at 30 frames per second and is suitable for low light scenarios. 79020 Data Set gov.noaa.nmfs.inport:69412 Katherine Wilson 2023-03-09T02:35:34 Katherine Wilson 2026-01-30T19:42:25 2025-09-26 Alaska Fisheries Science Center AFSC 7600 Sand Point Way N.E., Building 4 Seattle WA 98115 USA https://www.afsc.noaa.gov 0700-1700 Pacific Time 1001 Public No No 2025-09-09 1 Year 2026-09-09