Development of Autonomous River Trash Collector System for Plastic Bottle Waste Segregation Using Machine Learning-Based Object Detection

Abstract

Plastic pollution in rivers is a critical environmental challenge that significantly contributes to marine pollution and jeopardises aquatic ecosystems, biodiversity, and human health. This study developed an autonomous river trash collector system that integrates a machine-learning-based object-detection framework. By leveraging advanced sensors and real-time machine vision, the system effectively detects and collects plastic debris, including bottles, preventing further pollution in rivers, lakes, and ocean streams. The trash collector enhances waste-management efficiency and promotes environmental sustainability. The object detection module was evaluated using three TensorFlow Object Detection models: Faster R-CNN, EfficientDet, and MobileNet SSD. Due to computational intensity, MobileNet SSD is used for implementation, as it provides an optimal balance between accuracy (F1 score 91.79% and precision 90.48%) and real-time processing efficiency. The system employs two bounding boxes: a detection box activated at a 75% confidence threshold for plastic bottles, and a blue Region of Interest (ROI) box for object tracking and counting. By merging the smart object detection with sustainability, the proposed solution offers a scalable and impactful strategy to mitigate plastic pollution, protect water resources and raise public awareness of the critical need for responsible plastic waste management.