To achieve rapid classification and identification of common cosmetic paper packaging boxes in public security practice, a packaging material recognition method based on X-ray fluorescence spectroscopy (XRF) combined with deep learning algorithms is proposed. First, XRF technology is used for non-destructive testing of 61 paper packaging box samples from different cosmetic brands to analyze their elemental composition, followed by manual classification based on key characteristic elements. Subsequently, random forest, multi-layer perceptron (MLP), deep neural network (DNN), and the proposed MLP-DNN model are constructed. A total of 70% of the samples are randomly selected as the training set for model development, while the remaining 30% serve as the test set for validation. The system evaluates the performance of each model in the classification task and checks the classification performance through cross-validation. The experimental results show that the MLP-DNN model achieves a classification accuracy of 0.89 on the test set, significantly outperforming the random forest (0.85), MLP (0.68), and DNN (0.74), thereby demonstrating its superiority in classifying complex samples. The proposed MLP-DNN model provides a more efficient technical solution for the rapid identification of forensic evidence in public security applications.