The work aims to conduct rapid non-destructive inspection and differentiation of the physical evidence of edible oil barrels left on the scene of arson cases, and provide clues for the public security organs to investigate and solve the case. A total of 52 samples of 26 pairs of edible oil barrels and bungs were tested by the latest differential Raman spectroscopy technology. Firstly, the two types of samples were preliminarily classified through traditional spectral analysis. At the same time, combined with multivariate analysis, hierarchical clustering was performed to the bungs and K-means cluster analysis was carried out to the barrels based on the Gap statistic algorithm to determine the k value. The bungs and barrels were successfully divided into 3 categories, respectively. The classification results were consistent with the category of the known samples. Through this method, the rapid non-destructive inspection and differentiation of edible oil barrels can be performed, which provides certain help for the detection of arson cases where edible oil barrels are left on the scene.