An improved identification method based on Bayesian regularization optimization for the imbalanced proportion plastics recycling using NIR spectroscopy

Abstract

Near-infrared (NIR) spectroscopy is an efficient and non-destructive method for the identification and classification of mixed plastics. In the identification process of NIR spectroscopy, the dataset proportion of each type of plastic obtained is imbalanced due to the difficulty of obtaining or special application environments. When the backpropagation neural network (BPNN) identification model identifies samples with imbalanced proportions, it may misidentify plastic categories with small proportions, or even fail to identify them. Considering this, this study proposes an improved BPNN identification method based on Bayesian regulation optimization. To illustrate the performance of the proposed model, NIR spectroscopy data from 200 samples of plastic-containing additives were analyzed for four plastics: acrylonitrile butadiene styrene, polyamide, polypropylene, and polycarbonate/acrylonitrile butadiene styrene blend. The spectral data was preprocessed by Savitzky-Golay smoothing and multivariate scatter correction. Competitive adaptive reweighted sampling method was used to extract information from the spectral data. The identification ability of the proposed model was evaluated using accuracy, recall and precision determined through macro and micro average The experimental results show that the overall accuracy of the proposed method to identify imbalanced small proportion plastics is improved by 7.7% on average compared with the method using the BPNN identification model.