Designing of IoT-based Smart Waste Sorting System with Image-based Deep Learning Applications

Abstract

With an increase in population, there is an exponential increase in the amount of waste produced. This waste contains a high percentage of plastic that can be recycled. It is therefore necessary to classify and separate different types of waste. In order to minimize the environmental impact of improper waste disposal, we propose a robotic automation system based on deep learning techniques to help ensure proper waste separation in the recycling categories. The ResNet-50 has been used to classify the waste. The model was trained in a TrashNet dataset and a local image collection containing approximately 5,326 images of four different categories of waste. The experimental accuracy was 98.81%. We have developed a Smart Bin with computer vision and IoT that can separate waste automatically. The Pi camera captures multiple images of the waste when the motion sensor is triggered, and then sends the images to the Deep Learning model, which then returns the output (PET, plastic, metal, and trash) to the Raspberry Pi. Based on the output generated, the waste is automatically moved to its respective bin using a motorized sliding tray to the appropriate container. A smart university social enterprise engages students in earning points by sorting out the amount of waste to be used for university redemption.