Accurate Characterization of Mixed Plastic Waste Using Machine Learning and Fast Infrared Spectroscopy
journal contributionposted on 12.10.2021, 19:34 by Stas Zinchik, Shengli Jiang, Søren Friis, Fei Long, Lasse Høgstedt, Victor M. Zavala, Ezra Bar-Ziv
We present a combination of convolutional neural network (CNN) framework and fast MIR (mid-infrared spectroscopy) for classifying different types of dark plastic materials that are commonly found in mixed plastic waste (MPW) streams. Dark plastic materials present challenges in fast identification because of the low signal-to-noise ratio. The proposed CNN architecture (which we call PlasticNet) can reach an overall classification accuracy of 100% and can identify the constituent materials in a multiplastic blend with 100% accuracy. The fast MIR system can collect spectral data at a rate up to 400 Hz, and the CNN model can reach prediction speeds of 8200 Hz. Therefore, this method provides an avenue to be able to characterize MPW in a real-time high-throughput manner.
Read the peer-reviewed publication
mixed plastic wasteconvolutional neural networkcollect spectral dataclassifying different typesdark plastic materialsreach prediction speedsproposed cnn architectureoverall classification accuracyfast mir systemfast infrared spectroscopyinfrared spectroscopyfast mirfast identificationconstituent materialstime highthroughput mannernoise ratiomultiplastic blendmethod provideslow signalcommonly foundcnn modelcall plasticnetaccurate characterization8200 hz400 hz