Abstract: This study introduces IFcaNet, an advanced method for authenticating cigarette packaging. It segments the image into three sub-images, employs FcaNet for detailed feature extraction, and leverages self-attention to boost feature representation. By integrating salient and average features through max and average pooling, the model gains comprehensive image insights. The concatenated features are then classified using a fully connected layer. Experimental results show that IFcaNet has superior classification accuracy and generalization over existing models, highlighting its effectiveness in counterfeit detection.