Abstract:

Plastic beverage bottles are a type of common material evidence at the scene of the crime. Fourier-transform infrared spectroscopy （FTIR） was used to detect 41 different brands of plastic beverage bottles. After pretreatment， the samples were divided into polyethylene terephthalate and polyethylene. The infrared characteristic peaks of each sample in each category were different. For the largest number of samples， the spectral data dimension was reduced， and the principal components were extracted by the principal component analysis. The samples were further grouped by a K-means clustering method. Finally， the clustering results were used as dependent variables to construct a neural network algorithm to train the data to predict the classification of samples. With the help of random number generator， 86.5 % of the samples were selected as the training set and 13.5 % of the samples were selected as the test set. The results indicated that the accuracy of the training set and test set reached 100 %. Meanwhile， the accuracy of the K-means clustering results was verified， and a fast classification model for plastic beverage bottles was established. This classification model has good operability with accurate and reliable results. This work provides a reference for the police grassroots to handle cases.

Key words: spectroscopy, Fourier transform infrared spectroscopy, plastic beverage bottle, chemometrics, artificial neural network