2022-11-01-Capgemini-海洋数据和人工智能用于物种保护（英）_14页_1mb

Ocean Data and AI for Species Conservation

Overview:
The project, OceAIn, aims to detect anomalies in ocean data using AI to support marine ecosystem research and species conservation. The Lofoten-Vestralen Ocean Observatory in Norway provides cross-sectional data from depths of up to 200m, measured by sensors recording temperature, sound, currents, and biological data.

Key Features:

1. Data Collection

   • Sensors measure directional sounds, biomass, currents, and water chemistry.
   • Data from 7 sensor nodes and land-based stations have accumulated over 100 terabytes since 2013.

2. AI Model Development

   • Initial approach used unsupervised AI due to its lack of need for labeled data but proved ineffective due to data noise and high dimensionality.
   • Shifted to supervised AI, resulting in better F1 scores (e.g., supervised models scored up to 0.85, surpassing baseline models).
   • Models process multiple data types in Docker containers managed via Kubernetes.

3. Visualization

   • A web app powered by Python, Docker, and Three.js visualizes oceanographic data.
   • Users can explore time-based trends or sensor-specific features (e.g., EK60 shows biomass as color gradients).

4. Infrastructure & Architecture

   • Cloud-native architecture (e.g., using Airflow for workflow scheduling) supports scalable and automated processing.
   • The system runs on CSV → JSON → image formats to reduce data transmission.

Results & Outcomes

• The platform automates anomaly detection, saving researchers time from manual analysis.
• Supervised AI models clearly outperform unsupervised ones, a key insight from the project.

Outlook

• Further automate the workflow and enable continuous AI retraining.
• Enhance visualization tools and ensure the system’s continuous availability.

Conclusion
Collaboration between marine biology and IT drives innovative tools for species conservation, improving understanding of ocean ecosystems through advanced data analysis.