Integrating vision-based AI and large language models for real-time water pollution surveillance.

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Water pollution is a critical global issue affecting billions, stemming from natural and anthropogenic sources.
• Effective monitoring of surface water quality is essential for timely intervention and mitigation strategies.
• Existing methods often lack real-time capabilities and comprehensive contextual analysis of pollution events.

Purpose of the Study:

• To develop an automated, real-time vision-based surveillance system for monitoring surface-level water pollution.
• To integrate large language models (LLMs) for generating contextual information on detected pollutants.
• To provide local authorities with an effective tool for water quality management and environmental protection.

Main Methods:

• A multi-model system combining a camera, Raspberry Pi for frame processing, and LLMs (ChatGPT API) for analysis.
• Utilizing the YOLOv5 object detection model trained on seven pollutant types: algal blooms, synthetic foams, dead fish, oil spills, wooden logs, industrial waste, and trash.
• Developing a system capable of autonomous surveillance and alerting authorities without human intervention.

Main Results:

• Accurate detection of various water pollutants using the vision-based system.
• Generation of contextual information regarding pollutant type, causes, and environmental/health impacts via LLMs.
• Successful demonstration of real-time monitoring and automated alerting capabilities.

Conclusions:

• The integrated system offers a novel approach to real-time water pollution surveillance and management.
• LLM integration enhances the system's utility by providing actionable insights beyond simple detection.
• The system has the potential for broad application, including integration with drones and mobile platforms for comprehensive environmental monitoring.