Cost-Effective Automated Plastic Sorting Using InGaAs Photodetectors and Machine Learning: A Multi-Attribute Detection Approach for Enhanced Recycling Automation

Research at a glance

Design of the sorting machine This study presents the development of an intelligent plastic sorting system that integrates Indium Gallium Arsenide (InGaAs) photodetectors with machine learning algorithms to address critical challenges in plastic waste management within developing countries. The proposed system employs near-infrared (NIR) spectroscopy in combination with a simplified hardware architecture to enable automated classification of commonly encountered plastic types. A prototype was implemented using ESP8266 and Arduino Uno microcontrollers, halogen bulb illumination, and Support Vector Machine (SVM) algorithms for classification. Experimental evaluation conducted on 300 plastic samples across three categories (PET, HDPE, PP) demonstrated an average classification accuracy of 95%, with a processing speed of approximately 10 seconds per item. The total cost of 48,500 LKR (≈150 USD) highlights a substantial cost reduction compared to conventional commercial NIR-based systems while maintaining comparable performance levels. Key contributions of this work include the demonstration of cost-effective halogen illumination as a viable infrared source and the integration of embedded machine learning for real-time classification. The findings connect the gap between laboratory-scale NIR plastic sorting solutions and practical deployment in small-to-medium enterprises, offering a low-cost pathway towards circular economy practices and sustainable waste management in resource-constrained environments.