AI Spots River Trash from Riverbanks

Millions of tons of plastic and debris flow through rivers into oceans each year, harming wildlife and water quality. Current monitoring methods, like visual checks or boat sampling, are slow and miss key data. Researchers have now developed an AI system using fixed riverbank cameras that continuously detects floating waste, offering a low-cost, automated solution to track pollution in real-time.

The team discovered that specific AI models, particularly YOLOv11-m, can identify floating debris like plastic bottles and bags with high accuracy. In tests, this model achieved a mean average precision of 96.2% for detecting objects, meaning it correctly identifies and locates most debris in images. It also processes images quickly, handling about 1.86 frames per second on standard computing hardware, making it suitable for ongoing monitoring.

To build the system, researchers installed pan-tilt-zoom cameras along the Steingiessen River in France, capturing over 6,000 images of intentionally added debris, including glass, polystyrene, and plastic items. They manually labeled these images to train the AI, using techniques to prevent data leakage—where similar images in training and test sets could falsely boost performance. By grouping images based on visual and temporal similarities, they ensured the AI's results reflect real-world conditions, not just memorized data.

Results showed the AI not only detects debris but also estimates its size using geometric corrections. For example, the system reduced errors in height predictions from 2.08 cm to 1.50 cm after applying corrections, getting close to the theoretical limit of precision given the camera's resolution. This allows for reliable measurements of debris dimensions, which is crucial for calculating pollution volumes. The approach was validated on a different river, the Bruche, demonstrating its adaptability to new environments without retraining.

This innovation matters because it provides a scalable way to monitor river health, helping communities and authorities address plastic pollution before it reaches oceans. By automating detection, it reduces the need for labor-intensive surveys and supports global efforts like the One Health initiative, which links environmental and human well-being. For instance, better tracking could inform cleanup strategies and policy decisions, potentially reducing risks to aquatic life and public health from microplastics.

Limitations include the AI's struggle with highly imbalanced data, where some debris types are rare, and challenges like water reflections that can distort detections. The study notes that future work should expand datasets and incorporate rotation-aware models to handle tilted objects better. Despite this, the system lays a foundation for widespread, affordable river monitoring systems that could transform how we protect freshwater ecosystems.