INTELLIGENT PLASTIC BRAND AUDIT FOR EXTENDED PRODUCER RESPONSIBILITY INITIATIVES USING MACHINE LEARNING MODEL

Abstract

The ever-growing volume of plastic waste poses a significant threat to global ecosystems. Existing waste management systems often struggle with the identification and sorting of plastic waste due to limitations in the scalability and cost-effectiveness of smart technologies. One key aspect of plastic waste mitigation is to enhance extended producer responsibility (EPR) capacity through plastic waste auditing. While computer vision techniques have been explored for general waste sorting, there is a lack of research focused on automated brand identification within plastic waste. This paper proposes a novel Intelligent Plastic Brand Audit (IPBA) system leveraging TinyML with machine learning capabilities for resource-constrained edge devices. The system utilizes a lightweight Faster Objects More Objects (FOMO) model trained on user-generated labelled photos and data. The performance evaluation of FOMO for IPBA was performed on two hardware platforms: Arduino Nano BLE and ESP32-EYE-CAM. Across both configurations and with five plastic brand classes, the system achieves high accuracy, with a minimum F1 score of 93.5%. These results indicate the potential of IPBA to improve existing manual sorting systems and support circular economy initiatives. By facilitating automated brand identification in plastic waste, IPBA can enhance EPR programs and hold brands accountable for their waste footprints.