Classification of plastics by near infrared spectroscopy based on clustering algorithm

Studied a classification method of plastics by combining machine learning algorithm with near infrared spectral analysis technology. After SG filtering pretreatment and principal component analysis of spectral data, unsupervised clustering K-means algorithm, supervised clustering maximum likelihood estimation, Fisher discriminant and spectral Angle algorithm are used. PP, PET, HDPE, TPE, PLA, PBT, TPU, POM-M90, POM-GF20NC, TPB, PPS, ABS, PPO(natural), SAN, PPO(white), POM-F20 plastic particles were classified and discriminated. The classification results based on the near infrared spectral data of 16 kinds of plastic samples show that the K-means algorithm can distinguish POGF20NC, PLA and PPO from other plastic particles with the accuracy of 100%, 100% and80%, respectively. Fisher discriminant and maximum likelihood estimation method were used to identify POM-M90 and POM-F20 with an accuracy of 93%.The recognition accuracy of spectral Angle algorithm is 80% for PET, 47% for POM-F20, and more than 90% for other particles.