AI-Based Model for Home Waste Separation Using Raspberry Pi 5 AI Kit

Effective domestic waste separation is also very critical to enhancing recycling and minimizing environmental pollution. Manual sorting, however, is labor-intensive, prone to errors and not practical to be widely adopted. In this paper, a new AI-controlled system of waste separation at home is presented, the computational power of the Raspberry Pi 5 AI Kit is used. It is a system consisting of a large-resolution camera, a conveyor belt system, and an edge deployment-optimized YOLOv5s convolutional neural network (CNN). The model is trained on the WasteNet dataset, with 25, 000 annotated images of five waste types (plastic, paper, glass, metal, and organic) and the classification accuracy of the model is 95.2%, with an average precision of 95.0%, and the time required to make an inference is 45 ms per frame. On average, the system consumes 5.2 W of energy and is therefore a cost-effective and energy-efficient system that can be used to manage household waste. It is possible to compare it with previous work and reveal that this technology is more effective regarding accuracy and latency and can be used as the effective tool to prompt environmentally friendly waste separation.