In this study, we investigate the application of the sensory fusion method to counterfeit label classification. Image sensing data were collected using a smartphone and processed through multilayer structural operations in various convolutional neural networks (CNNs) to extract features from different types of counterfeit label. The ReliefF algorithm filters out the top 10 important features of each CNN model. Canonical correlation analysis reorganizes the feature data into a small group of feature datasets. Finally, an efficient naive Bayes classifier and support vector machine methods were proposed to complete the classification of feature images. After experimental validation, the sensing feature fusion method proposed in this study demonstrated strong performance in counterfeit label recognition, achieving a maximum accuracy of 99.5238% per group. In addition, the fused feature dataset successfully achieved a high accuracy of 99.0496% under a high compression ratio of 1/50. In the case of wine counterfeit labels, the results showed that we can effectively fuse sensory features to enhance the processing speed while maintaining recognition accuracy.