Traditional counterfeit handbag detection methods approach the problem as a direct image classification task. However, due to the very close similarity between genuine and counterfeit handbags, it is difficult to distinguish between the two types of handbags. This chapter describes a two-stage universal detection framework. In the first contrastive learning stage, a Siamese network is employed to discover subtle differences between the two classes. In the second transfer learning stage, the previously-trained backbone network is frozen, following which a classifier is added to implement the transfer from contrastive learning to classification. To enhance discrimination, a multilevel feature selection module is incorporated to refine features and a dimension reduction pyramid pooling module is employed to fuse multiscale features. Experimental results demonstrate that the proposed framework achieves the highest precision of 99.1% on three testing datasets, outperforming existing mainstream methods.