Classification of diverse plastic samples by LIBS and Raman data fusion

Abstract

The plastic production and usage in the world is steadily increasing. This leads to increased amounts of plastic waste. Most of the waste could be potentially recycled, but only 14 % of plastic waste is recycled. In order to increase the share of recycling in plastic waste management, the recycling process should be completely automated. The problematic part of sorting is being solved by either manual (labor-intensive) or spectroscopy-based (still in development) methods. In this work, we propose the data fusion of Laser-Induced Breakdown Spectroscopy (LIBS) and Raman spectroscopy as a fast, robust, and reliable way to sort/classify any potential polymer material. The sample set of this work consists of several types of polymers in clear, colored, and even mixture versions. So far, no LIBS/Raman classification works involved all these categories in one experiment. Additionally, the low and medium level of data fusion is discussed, and the performance is compared. By using LIBS and Raman data fusion method and both linear and nonlinear chemometric techniques, increased accuracy reaching more than 98 % in the classification of investigated plastic samples was achieved, which was a significant improvement when compared with singular methods classification accuracy.The plastic production and usage in the world is steadily increasing. This leads to increased amounts of plastic waste. Most of the waste could be potentially recycled, but only 14 % of plastic waste is recycled. In order to increase the share of recycling in plastic waste management, the recycling process should be completely automated. The problematic part of sorting is being solved by either manual (labor-intensive) or spectroscopy-based (still in development) methods. In this work, we propose the data fusion of Laser-Induced Breakdown Spectroscopy (LIBS) and Raman spectroscopy as a fast, robust, and reliable way to sort/classify any potential polymer material. The sample set of this work consists of several types of polymers in clear, colored, and even mixture versions. So far, no LIBS/Raman classification works involved all these categories in one experiment. Additionally, the low and medium level of data fusion is discussed, and the performance is compared. By using LIBS and Raman data fusion method and both linear and nonlinear chemometric techniques, increased accuracy reaching more than 98 % in the classification of investigated plastic samples was achieved, which was a significant improvement when compared with singular methods classification accuracy.