We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Ocean and River Plastic Waste Detection Using Computer Vision
Summary
Millions of tons of plastic pour into our oceans every year, much of it carried by rivers—but tracking where it comes from has always relied on slow, manual surveys. This study built an AI camera system that automatically spots and counts floating plastic waste from drone and shoreline footage with over 91% accuracy, offering a faster way to find pollution hotspots. Better tracking matters because plastic waste that lingers in waterways breaks down into microplastics, which end up in the fish and water we consume, making tools like this useful for guiding cleanup efforts before pollution reaches our food and water supply.
Plastic pollution in aquatic ecosystems has emerged as one of the most pervasive environmental crises of the modern era, with an estimated 8 to 11 million metric tonnes of plastic waste entering the world's oceans annually, a substantial proportion of which is transported through river systems originating from inland sources. The Central Pollution Control Board of India identifies the Ganga, Cauvery, and several peninsular river systems as significant contributors to marine plastic loading, with riverine plastic transport patterns showing strong seasonal correlation with monsoon-driven surface runoff. Conventional plastic pollution monitoring relies on manual shoreline surveys and periodic sampling, both of which are labour-intensive, spatially sparse, and unable to provide the continuous, wide-area coverage necessary for effective pollution source tracking and cleanup resource allocation. This paper presents a computer vision-based framework for automated detection, classification, and density estimation of floating plastic waste in river and coastal water bodies, combining a fine-tuned YOLOv8 object detection model trained on a composite dataset of aerial drone imagery and fixed shoreline camera footage with a downstream density mapping module that converts per-frame detections into spatially and temporally resolved pollution density maps. The detection model is trained to distinguish seven plastic waste categories (PET bottles, plastic bags, styrofoam, multilayer packaging, fishing net debris, plastic containers, and miscellaneous plastic fragments) from natural floating debris such as leaves, wood, and algae mats that commonly cause false positives in aquatic object detection. Field validation conducted along a 12-kilometre stretch of the Adyar River in Chennai and a coastal stretch near Marina Beach demonstrates a mean detection accuracy (mAP@0.5) of 91.4%, real-time inference at 28 frames per second on edge GPU hardware suitable for drone deployment, and density mapping correlation of r=0.87 against manual ground-truth quadrant sampling, establishing the framework's viability as a scalable monitoring tool for river and coastal plastic pollution management authorities. Keywords — Plastic Pollution Detection, Computer Vision, YOLOv8, Marine Debris, River Monitoring, Aerial Drone Imagery, Object Detection, Environmental Monitoring, Density Mapping, Aquatic Ecosystem