关键词: 光谱学, 运动鞋底材料, 化学计量学, 径向基函数神经网络

Abstract: Sports shoe sole is a kind of common material evidence in the scene of a case. Based on chemometrics and radial basis function neural network, Fourier transform infrared spectral data of 45 different brands of sports shoe sole materials were collected. After preprocessing, the infrared data were divided into four categories according to the material composition. Firstly, the principal component analysis was used to reduce the dimension of the infrared spectral data and extract the principal components. Then, the samples were grouped by using the two-step clustering method, and the clustering effect was evaluated by using the discriminant analysis method. Finally, based on the two-step clustering results, RBF neural network algorithm was constructed to train the infrared data of samples, and the identification and classification model of sports shoe sole evidence was further established. 78.6% of the samples were randomly selected as the training set and 21.4% as the test set. Results show that the classification accuracy of the training set reaches 100%, while the classification accuracy of the test set reaches 83.3%. The accuracy of RBFNN model was verified by drawing the ROC curve and calculating the offline area of each group of samples to evaluate the diagnostic value. The rapid and non-destructive identification of sports shoe sole materials was thus realized. This classification model has good operability, accurate and reliable results.

Key words: spectroscopy, sports shoe sole material, chemometrics, radial basis function neural network