Sustainable Development Goals
Abstract/Objectives
This project focuses on the Touqian River Basin in Hsinchu, integrating artificial intelligence (AI) image recognition technology and Internet of Things (IoT) monitoring devices to establish an AI-based Image Recognition and Analysis System for River Waste Monitoring. The main objective is to automatically detect and classify riverbank waste through real-time image surveillance and AI recognition models, thereby improving the efficiency and accuracy of river pollution monitoring and environmental management decisions. In the first phase, high-speed dome cameras were installed along the Matatai Trail to enable real-time image capture, night vision monitoring, and data transmission, forming the foundation of the image database. The second phase expands the dataset through image collection, labeling, and augmentation, employing the YOLOv11 model for training with an achieved recognition accuracy of over 88%. The project also develops a waste reporting and data analytics service. Ultimately, it aims to integrate real-time recognition, user reporting, and visualized data analysis into a sustainable, automated river waste monitoring system that promotes collaboration among local communities, volunteers, and government agencies, advancing smart environmental management and sustainable development goals.
Results/Contributions

This project, "Touqian River Waste AI Image Recognition System," led by Professor Neng-Fu Huang's team from the Department of Computer Science at National Tsing Hua University, aims to address the challenge of monitoring waste pollution in the Touqian River, Hsinchu's vital waterway. Facing the difficulty of capturing real-time, comprehensive data on river waste, the project's core content involves developing and deploying an advanced AI image recognition system.

The project began by installing real-time cameras at key points in the Touqian River basin (such as Matagudao) to automatically collect images. To train a high-accuracy recognition model, the team integrated a diverse image database, including: 1) Actual images captured by on-site cameras (over 500 collected); 2) Public open-source waste datasets like TrashNet (over 4,700 collected); 3) Samples photographed by the team (over 300 collected). Furthermore, the project introduced Generative AI technology to create images of specific or rare types of waste, significantly enhancing the model's training effectiveness and recognition breadth.

Project Results:

This project successfully constructed an automated, intelligent platform for monitoring river waste. Key results include:

  1. AI Recognition Core: An advanced AI recognition model was successfully trained and deployed. It can accurately identify various types of waste in images, such as plastic bottles, paper, and metal cans.
  2. Real-time Monitoring and Alert System: The system automatically analyzes camera-feed images 24/7. Upon detecting waste, it immediately triggers an alert mechanism (e.g., via Line Notify) to inform relevant authorities for cleanup, drastically reducing response time.
  3. Data Dashboard and Trend Analysis: The project developed a visual data dashboard. This platform not only displays real-time waste detection results but also features statistical analysis functions. It reveals waste hotspots, common types, and trend periods, providing a strong scientific basis for allocating cleaning resources and formulating pollution prevention strategies.
Keywords
Touqian River Basin, River Waste Monitoring, AI Image Recognition, IoT Surveillance System, Environmental Sustainability Management
References
Contact Information
分析與環境科學研究所
a0908588968@gmail.com