Introduction

The lack of comprehensive waste sorting facilities is an issue in Indonesia. While recycling efforts have been initiated, the absence of proper sorting mechanisms hampers their effectiveness. As a result, recyclable materials often end up in landfills, exacerbating environmental degradation and wasting valuable resources. Informal waste pickers play a crucial role in waste management in Indonesia. However, their working conditions are often substandard, and they face social stigmatization and health risks due to their activities. Integrating these informal waste pickers into formal waste management systems and improving their working conditions could enhance waste collection and sorting processes. In this work, we developed a robotic sorting system capable of separating items into 8 categories which are designated for the 7 plastic recycling categories and 1 extra category for general purpose.

Software

For the software, we used and modified the YOLOv5 machine learning model for real-time image analysis and used Python as the primary programming language. We also used the Arduino IDE to program our ESP32 microcontroller. C++ was used to program the ESP32.

Hardware

This project required a variety of different parts and technology, but the main ones include an ESP32 microcontroller, a stepper motor and driver and a 24V power supply. We used a stepdown voltage converter to turn the 24V to 5V as that was the maximum voltage the ESP32 could handle. The driver was necessary to help control the speed and smoothness of the stepper motor’s turns.

YOLOv5

The Yolov5 model follows a one-stage object detection approach, meaning it directly predicts bounding boxes and class probabilities from an input image, without the need for a separate region proposal network. A backbone network, a neck network, and a detection head comprise the model architecture. The backbone network, typically based on a convolutional neural network (CNN), extracts features from the input image at multiple scales. These features capture various levels of detail and are subsequently passed to the neck network. The neck network fuses the multi-scale features extracted by the backbone network to generate more contextually rich representations. This fusion of features enhances the model's ability to detect objects at different scales and aspect ratios. The detection head receives the fused features and performs the final prediction. It predicts bounding box coordinates (x, y, width, height) and class probabilities for each potential object in the image. Non-maximum suppression (NMS) is then applied to remove redundant detections and retain the most confident and accurate bounding boxes.

Result and Conlusion

The future of plastic sorting through AI promises significant advancements in efficiency and accuracy. AI-powered systems, equipped with computer vision and machine learning algorithms, can analyse and classify different types of plastics with greater precision than traditional manual sorting methods. By leveraging vast amounts of training data, AI models can learn to recognize and differentiate between various plastic materials, colours, and shapes. This enables automated sorting processes that are faster, more reliable, and more cost-effective. With ongoing research and development, AI-driven plastic sorting systems have the potential to revolutionize waste management, promoting increased recycling rates, reduced contamination, and a more sustainable approach to plastic waste.