Identification of bio-plastics by NIR-SWIR-Hyperspectral-Imaging

Bio-plastics are characterized by the highest rate of growth in the plastics industry. In connection with the recycling chain they constitute the so-called “oxo-biodegradation” and drop-in problems. The present study tries to clarify possibilities of automatic recognition and sorting of conventional fossil-based plastics against similar “oxo-biodegradable” plastics and drop-ins by means of NIR-SWIR-Hyperspectral-Imaging (HIS). The spectral structures of the most important plastics (conventional fossil-based plastics and bio-plastics) have been incorporated in a database as references for different plastic types to be subject to identification by NIR-SWIR-HSI. In addition to widespread chemometrical methods (PLS-DA), artificial neural networks (ANN) and support vector machines (SVM) are estimated for classification. For “oxo-biodegradable” plastics it turns out that a decision tree is the most reliable procedure for identification. Different decision trees are passed to an industrial NIRSWIR-Hyperspectral-Imager for generating chemical images of different plastic mixtures. The mixtures consist of conventional fossil-based plastics and bio-plastics. The aim of the tests was to find bounds for sorting throughput and purity. Results of an industrial sorting trial are finally described