Vehicle’s paint chips are one of the common physical evidences at the scene of criminal cases and traffic accidents. Identification and recognition of such physical evidences have been an important part of the research of public security technicians as they might contain some important information about the vehicles involved．Traditional instrumental analysis methods are cumbersome and cannot meet the demand for rapid nondestructive testing of physical evidence at the scene．Based on the advantages of nondestructive imaging and simple operation of spectral detection，a hyperspectral imaging technology combined with deep residual shrinkage network was explored to identify on-site car paint evidence，in order to improve detection efficiency and optimize detection means．The hyperspectral images of 54 car paint samples from 18 different car models were collected，and the images were fused with 10 × 10 pixels to form a reflection spectrum of 19 740 pixels．Combined with the characteristics of hyperspectral data，a targeted one-dimensional deep residual shrinkage network（1D-DRSN） recognition model was established in this paper．The results showed that 1D－DRSN model has 99.5% and 98.6% accuracy for the training and test sets，with the loss function values decreasing to 0.093 and 0.106，respectively．1D-DRSN was significantly better than 1D convolutional neural network（1D-CNN） model and three traditional machine learning methods，including support vector machine（SVM），random forest（RF） and logistic regression（LR）．The white car paint samples from different brands or models are distinguished by deep mining of hyperspectral data．This study was applied to solving the practical challenges，and it will be a powerful complement to the existing techniques and methods in the detection and classification of vehicle’s paint fragments from the scene.