TOPIC

• Titile: Plastic Material Identification for Recycling Using 1D SWIR Spectral Data with AutoEncoder-Assisted XGBoost Classification

INFORMATION

• Producer: Ki-Hyun Hwang
• Published: IEEE Sensors Conference'25
• Paper: Link

IMAGE AND VIDEO

OVERVIEW

• This paper presents a hybrid machine learning framework for plastic classification using a custom-developed short-wave infrared (SWIR) spectrometer based on a 1D spectral image sensor. To address the challenge of misclassifying out-of-distribution (OOD) data in real-world recycling scenarios, the system integrates an AutoEncoder-based OOD detector with an XGBoost classifier. The AutoEncoder filters out anomalous samples by measuring reconstruction error, while the XGBoost model classifies in-distribution (ID) spectral data into predefined plastic types. Experimental results demonstrate that the proposed system achieves a high ID F1-score of 0.95 and improves OOD detection F1-score from 0.78 to 0.98 when the AutoEncoder is applied. The results confirm the effectiveness of the proposed architecture in enhancing classification reliability and robustness, making it suitable for deployment in automated plastic recycling processes.