Abstract: This study addresses the traceability requirements of hand cream plastic packaging evidence by designing a classification method that combines X-ray fluorescence spectroscopy (XRF) with deep learning. Non-destructive testing was conducted on 52 hand cream packaging samples using XRF. The obtained spectral data were standardized using z-score normalization, and K-means clustering was employed for classification. The optimal number of clusters was determined using the Elbow method and Silhouette analysis, with four clusters identified as the best choice. The reliability of the classification results was validated by constructing a Backpropagation Neural Network (BPNN) model and a BPNN model optimized with Particle Swarm Optimization (PSO). Experimental results showed that the BPNN model alone achieved an accuracy of 96.4286% on the training set and 91.6667% on the test set, while the PSO-BPNN model achieved 100% accuracy on both the training and test sets. The results indicate that the combination of X-ray fluorescence spectroscopy and the PSO-optimized BPNN model can effectively classify hand cream packaging with high precision, providing a more scientific approach for evidence traceability