Deploying YOLOv5 and CNN on Edge Computing for Real-Time Classification of Household Waste in Autonomous Robotic Systems

Household waste mismanagement remains a persistent environmental challenge, particularly in developing countries such as Indonesia, where waste generation continues to rise due to rapid urbanization and population growth. In 2023, over 25 million tons of waste were produced, yet only half was properly managed. With household waste accounting for nearly half of the total, effective sorting systems are essential to improve recycling rates and reduce environmental harm. This study proposes an automated household waste classification system using a Convolutional Neural Network (CNN)-based YOLOv5 model integrated into a Jetson Nano-powered platform mounted on an omni-directional robot. The system employs a Logitech C270 camera and additional sensors to detect and classify waste into categories including plastic, paper, metal, and organic materials. Datasets were collected directly using the robot’s camera, then labeled and trained through Roboflow. Experimental results show that increasing dataset volume and variation significantly improves classification accuracy. The proposed approach demonstrates high performance in real-time waste detection, offering a promising solution to optimize waste management using deep learning and embedded systems.