Classification of plastics using laser-induced breakdown spectroscopy combined with principal component analysis and K nearest neighbor algorithm

• Published in: Results in optics 2021-08, Vol.4, p.100093, Article 100093
• Main Authors: Yan, Xiaotao, Peng, Xinying, Qin, Yuzhi, Xu, Zhiying, Xu, Bohan, Li, Chuangkai, Zhao, Nan, Li, Jiaming, Ma, Qiongxiong, Zhang, Qingmao
• Format: Article
• Language: English
• Subjects: Dimensionality reduction; kNN algorithm; Laser-induced breakdown spectroscopy; Plastics classification; Principal component analysis
• ISSN: 2666-9501; 2666-9501
• DOI: 10.1016/j.rio.2021.100093

•Using K-nearest neighbor(kNN) and principal component analysis(PCA) algorithm to support multi-target classification.•The loading plot of each principal component and the relationship between principal component and classification accuracy are provided.•On the premise of reducing the data dimension to 20 dimensions, the training time of classification model based on 1000 sample data is reduced from 368.99 seconds to 167.30 seconds, while the time for individual sample classification is reduced from about 0.11 seconds to about 0.02 seconds, and the classification accuracy reaches 99.6%. Plastics play an important role in manufacture and our daily life. In order to realize fast classifications of plastics products, this paper proposes a method using laser-induced breakdown spectroscopy (LIBS) combined with principal component analysis (PCA) and K_nearest neighbor algorithm (kNN) to achieve highly-accurate classification of plastic products with a small amount of data training. After dimensionality reduction by PCA, the higher the dimensionality reduction, the higher the average recognition accuracy of samples, but the rising trend tends to be flat. When the original data of each sample is reduced to less than 10 dimensions, the classification accuracy of classifying the same kind of samples produced by different manufacturers into different categories is significantly higher than that of classifying the same kind of samples produced by different manufacturers into one category. However, when the data is reduced to more than 10 dimensions, there is little difference between the two classification methods, when reduced to 20 dimensions, the average recognition accuracy is 99.6%. In the aspect of improving classification efficiency, after dimensionality reduction by PCA, the training time of the model is reduced from 369 s to 168 s, and the time of classifying a single sample is reduced from 0.1 s to less than 0.02 s. This work provides an effective method for rapid automatic classification in the process of plastic manufacturing and recycling.