Designing Ocean Vision AI: An Investigation of Community Needs for Imaging-based Ocean Conservation

Title of the Paper

Designing Ocean Vision AI: An Investigation of Community Needs for Imaging-based Ocean Conservation

Paper Information

• Field of Study: Ocean Science and Artificial Intelligence (AI)
• Keywords: Ocean Science, Sustainability, Data Portal, Machine Learning Interface, Citizen Science, Human-Centered Design

Research Background and Issues

• Identified Problems or Challenges:

  • Ocean researchers face significant challenges in data analysis. Despite the increasing use of imaging devices, automated tools for analyzing such data remain limited.
  • Over 300,000 hours of underwater video have been collected, yet only about 15% of the data has been annotated by human experts. Current data analysis methods are inefficient and lack scalability.
  • The absence of unified image training datasets and standardized data management frameworks hinders the development of machine learning models.
  • There is insufficient culture of ocean data sharing, with issues of trust and unequal resource distribution.

• Significance:

  • Research on marine biodiversity and ecological processes is a critical component of addressing climate change and promoting sustainable development.
  • Developing automated data processing tools can accelerate scientific research and support industries related to the "blue economy."

• Research Motivation and Related Work:

  • Relevant communities have begun exploring the potential of machine learning and open data structures in ocean science, but these efforts are often limited to narrow applications.
  • This study aims to investigate how machine learning-driven imaging analysis tools can better meet diverse community needs through human-centered design approaches.

Solution

• Proposed Solution:

  • Develop a toolset named “Ocean Vision AI (OVAI)” that integrates the open database FathomNet to process ocean images and video data.
  • OVAI provides end-to-end data analysis functionalities, including data uploading, annotation, search, machine learning model training, and community collaboration features.

• Innovative Features:

  • Employ human-centered design (HCD) methods to collaboratively define needs with stakeholders.
  • Innovatively build the FathomNet image database through community collaboration, offering unified data labels for 202,063 marine species.
  • Include gamification and educational features to engage citizen scientists and ocean enthusiasts.
  • Offer modular design tools tailored to different user types (scientists, businesses, the public, etc.).

• Implementation Steps and Technologies:

  • Conduct 36 in-depth interviews and a global workshop with 246 participants to gather community needs.
  • Design functionalities such as data filtering (based on geographic location or conceptual keywords), machine learning model training, collaborative annotation, and validation.
  • Develop open interfaces and APIs to facilitate data and model sharing, supporting data hosting for both households and enterprises.

Research Outcomes

• Specific Results:

  • Identified four core issues: challenges in data sharing, lack of machine learning knowledge, absence of convenient tools for visual data processing, and diverse community needs for ML tools.
  • Recognized 12 user archetypes, including academic researchers, ocean enthusiasts, nonprofit organizations, policymakers, etc., and designed targeted functionalities for each user category.

• Advantages and Comparisons:

  • The tools provided by OVAI significantly enhance data processing efficiency and offer greater applicability and multifunctionality compared to existing niche tools.
  • Promotes data standardization and community collaboration while providing user-friendly interfaces that lower the barrier to AI-powered participation for various users.

• Experimental or Project Evaluation Results:

  • The feasibility and community acceptance of OVAI tools were validated through thematic analysis and user testing.
  • User feedback highlighted OVAI's potential to address critical gaps in current data workflows, storage, and analysis systems.

• Limitations and Future Directions:

  • Participants were predominantly from the Northern Hemisphere, which may introduce geographic bias.
  • Lack of long-term funding support and global community coverage.
  • Future work includes improving the culture of data sharing, enhancing support for low-resource regions, and optimizing OVAI’s gamification interface to further attract public engagement.