Abstract:

To realize the data?based， visualized， non-destructive and high-efficiency identification of automobile lampshades as physical evidence frequently appearing in the judicial appraisal work， the PCA principal component analysis pre?processing method combined with a FDA-SVM （RBF） combination analysis was adopted to identify the physical evidence. An experiment and theoretical analysis was conducted for 173 sets of the Raman infrared spectroscopy data obtained from 18 brands “Aodi” and “Bieke” cars. With the help of Pearson correlation analysis and PCA principal component analysis， the characteristic Raman shift was selected， and the data classification models based on the Fisher discriminant analysis and SVM support vector machine were established. The results indicated that the FDA and SVM （RBF） models had a comprehensive discrimination accuracy of 97 % and 51.85 % for lampshade samples， respectively. The SVM model had an accuracy of 100 % for 8 brands of “Benchi” and “Bieke” cars. The FDA-SVM （RBF） model obtained from the complement of the FDA and SVM models can accurately distinguish the Raman infrared spectra of different brands of lampshades. This method is efficient and accurate， and it can provide some references for reducing the scope of investigation when using the lampshade material evidence identification in case investigations.

Key words: Raman spectroscopy, lampshade material evidence, support vector machine, fisher discriminant analysis