무인 페트병 수거 장치를 위한 딥러닝 기반 페트병 분류 기술 개발 = Development of Deep Learning-based PET Bottle Classification Technology for Unmanned PET Collection Devices

다국어 초록 (Multilingual Abstract)

The recycling of PET bottles faces challenges due to impurities like labels, caps, and internal foreign substances, which degrade material purity and impede high-quality recycling. Therefore, accurately identifying contaminated PET bottles during the classification stage is crucial for enhancing recycling efficiency. This study develops an AI-based detection model specifically designed to identify foreign substances in PET bottles and assesses its applicability to unmanned PET bottle collection systems. We experimentally analyze performance differences from three perspectives: image augmentation strategies, labeling methods, and various transfer learning–based YOLO model versions. These analyses consider the unique characteristics of the domain, where the focus is solely on PET bottles within a consistent imaging environment. Based on our findings, we propose effective directions for developing AI models optimized for PET bottle classification and contamination detection.