Abstract:

Middle infrared spectroscopy and chemometrics were adopted to identify the fragments of automobile bumpers. Two classification models， Fisher discriminant analysis and k-nearest neighbor algorithm， were established on the basis of the full-band spectral data， the spectral data of fingerprint region and the spectral data after the dimensionality reduction of principal component analysis of 52 vehicle bumper fragments. The comparison of classification results was also carried out. The results indicated that the classification model had higher classification accuracy when constructed by the principal component analysis that extracted the characteristic variables. The classification accuracy of polypropylene， polypropylene/talcum powders， polypropylene/talcum powders/calcium carbonate reached 92.3 %， and the classification accuracy of polypropylene/talcum powder obtained from 10 brand samples reached 88.9 %. This indicates an ideal classification result. In the two classification models constructed in this work， the classification rate of Fisher discriminant analysis model is much higher than that of the k?nearest neighbor algorithm model. It is believed that the k-nearest neighbor algorithm model is influenced by the imbalance of samples. The middle infrared spectroscopy combined with chemometrics can accurately distinguish the fragments of vehicle bumpers and therefore meets the requirement for rapid and nondestructive tests.

Key words: car bumper fragment, middle infrared spectrum, discriminant analysis, K-nearest?neighbor algorithm, classification