Abstract:

Aiming at the requirement for fast， nondestructive and accurate inspection， a spectral data fusion technology based on infrared spectrum original data and derivative data was adopted to analyze a series of automobile lampshade samples. The infrared spectra were obtained and collected from 44 lampshade samples. The automatic baseline correction， peak area normalization and Savitzky?Golay algorithm were used to preprocess the infrared spectra， and the first derivative was carried out for the processed data. The classification model was constructed by an artificial neural network algorithm. In the radial basis function neural network model， the original spectral data， first?order derivative data and fused data were classified as combined with the principal component analysis， and their classification accuracies reached 81.2 %， 84.1 % and 90.9 %， respectively. In the multi?layer perceptron neural network model， the original spectral data， first?order derivative data and fused data were classified with the classification accuracies of 84.1 %， 86.4 % and 97.7 %， respectively. In addition， the classification accuracy for 12 brands in 44 lampshade samples also reached 97.7 %， indicating an ideal experimental result. The research results indicated that the spectral data fusion technology based on the infrared spectrum original data and derivative data could realize an accurate analysis for the automobile lampshade samples and therefore meet the requirements for fast， nondestructive and accurate inspection. This work provides a good reference for the examination of physical evidence in the field of forensic science.

Key words: automobile lampshade, spectral data fusion technology, derivate spectrum, artificial neural network, classification