A fast and non-destructive classification method for paper express document bags was estab-lished.63 paper express document bag samples were analyzed by hyperspectral method.Smoothing,noise reduction,and multiple scattering correction were used to eliminate background interference from hyperspectral analysis.According to the significant differences in the sample spectra,the 63 paper ex-press document bag samples were classified into three categories.Meanwhile,competitive adaptive re-weighting sampling algorithm(CARS)and random forest algorithm(RF)were used to extract the feature spectral data of the samples,and a one-dimensional convolutional neural network(1D-CNN)model was established.The original classification labels were analyzed and verified by combining the feature wave-lengths extracted by the two algorithms,and the unknown samples were identified by prediction.The model recognition accuracy of CARS-1D-CNN and RF-1D-CNN were 84.21％and 94.73％,respective-ly.By comparison,it was found that RF-1D-CNN could realize more effective classification of paper ex-press document bags.The proposed method is simple and rapid,with a small sample size and no damage to sample,which can provide scientific basis for the identification of physical evidence in express docu-ment bags.
NETL
NSTL
万方数据
