Advancing fishery dependent and independent habitat assessments using automated image analysis: A fisheries management agency case study.

Journal: PloS one
Published Date:

Abstract

Advances in artificial intelligence and machine learning have revolutionised data analysis, including in the field of marine and fisheries sciences. However, many fisheries agencies manage sensitive or proprietary data that cannot be shared externally, which can limit the adoption of externally hosted artificial intelligence platforms. In this study, we develop and evaluate two residual network-based automatic image annotation models to process fishery specific habitat data to support ecosystem-based fisheries management in the Exmouth Gulf Prawn Managed Fishery in Western Australia. Using an extensive dataset of 13,128 manually annotated benthic habitat images, we train a grid-based annotation model and an image-level object detection model. Both models demonstrated high overall accuracy, with the grid-based model achieving 90.8% and the image-level model 92.9%. Patch-wise accuracy of the image-level model was 74.2%, highlighting its ability to classify broader spatial context without requiring point-based labelling. Precision and recall values for both models often exceeded 70% for dominant habitat classes such as unconsolidated substrate, macroalgae, and seagrass. The development of these models supports the potential for cost-effective, robust, and scalable in-house habitat classification for fishery or ecoregion specific habitat data to support timely decision-making. Further, the grid-based model uniquely integrates spatial precision with compatibility to existing manual data workflows, enabling seamless adoption within many existing fisheries monitoring programs. Despite limitations, such as a class imbalanced dataset, both models present a scalable, data secure solution for fisheries management agencies. This study establishes a foundation for integrating artificial intelligence driven image analysis of proprietary fisheries data, to further support responsive, standardised and data-informed decision making.

Authors

  • Scott N Evans
    Western Australian Fisheries and Marine Research Laboratories, Department of Primary Industries and Regional Development, Government of Western Australia, Hillarys, Western Australia, Australia.
  • Bronson Philippa
    Marine Data Technology Hub, College of Science and Engineering, James Cook University, Townsville, Queensland, Australia.
  • Carlo Mattone
    Marine Data Technology Hub, College of Science and Engineering, James Cook University, Townsville, Queensland, Australia.
  • Nick Konzewitsch
    Western Australian Fisheries and Marine Research Laboratories, Department of Primary Industries and Regional Development, Government of Western Australia, Hillarys, Western Australia, Australia.
  • Renae K Hovey
    School of Biological Sciences and UWA Oceans Institute, University of Western Australia, Crawley, Western Australia, Australia.
  • Marcus Sheaves
    James Cook University, Townsville, Australia.
  • Gary A Kendrick
    School of Biological Sciences and Oceans Institute, The University of Western Australia, Crawley, WA 6009, Australia.
  • Lynda M Bellchambers
    Western Australian Fisheries and Marine Research Laboratories, Department of Primary Industries and Regional Development, Government of Western Australia, Hillarys, Western Australia, Australia.