EARTH

[2801-14-03] Design of Intelligent Classifier Realized with the Aid of Data Information Processing Method and Laser Induced Breakdown Spectroscopy for Sorting of Black Plastics

○Sang-Beom Park¹, Sung-Kwun Oh², Woo Zin Choi¹ (1. The University of Suwon, 2. The University of Suwon(Corresponding author))

Chairman：Kazutoshi Haga （Akita University）, Makoto Harita (Harita Metal Co lyd)

Keywords:Radial Basis Function Neural Networks (RBFNNs), Laser Induced Breakdown Spectroscopy (LIBS), Dimensional reduction algorithm, Slope of spectrum, Spectrum peaks

These days, lots of plastic wastes including black plastics are produced in various industrial fields. As a result, the recycling of plastic wastes is emerging as key issues for solving environmental problems. In many countries including Korea, Extended Producer Responsibility (EPR) system is enforced as rule to reuse the plastic wastes to prevent environmental pollution. Especially in Korea, plastic wastes sorting system using near infrared (NIR) spectroscopy is being run. However, classifying the black plastics are difficult problem because colored black absorbs near infrared and doesn’t transmit enough signal to spectrometer.
So in this paper, Radial Basis Function Neural Networks (RBFNNs) classifier is introduced for sorting black plastic wastes. Using Acrylonitrile Butadiene Styrene (ABS), Polypropylene (PP) and Polystyrene (PS) materials, computer simulation to construct intelligent classifier are carried out for sorting plastic wastes. With the aid of Laser Induced Breakdown Spectroscopy (LIBS), spectra are acquired from 400 samples per each material. To improve processing speed of the proposed classifier, preprocessing algorithms such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and fuzzy transform are used as dimension reduction algorithm.
Through the slope of spectrum, characteristic of each material is analyzed and then the proposed classifier is applied. Also, the spectrum peaks of chemical elements such as carbon, hydrogen and nitrogen are considered as the input variables of the proposed classifier. By means of Fuzzy C-Means (FCM) clustering method as well as preprocessing algorithm, the proposed intelligent classifier based on fuzzy inference method is effectively designed. This study demonstrates that the proposed classifier shows more excellent classification rate when compared with the conventional Support Vector Machine (SVM) classifier.