Abstract: In order to improve the efficiency and accuracy of authenticity identification, an advanced method based on the improved frequency channel attention network (IFcaNet) for cigarette carton blanks was proposed. The method segments the image of cigarette cartons into three sub-images, and depth feature extraction of each sub-image was conducted by FcaNet. Through the self-attention module, effective interaction between the sub-image features was achieved, enhancing the expression ability of the carton features. Through max pooling and average pooling, the salient features and average features of the carton images were obtained, and the integration of the two types of data provided the model with more comprehensive image information. Finally, the concatenated vectors obtained from the max pooling and average pooling were inputted into a fully connected layer for binary classification. Experimental results showed that IFcaNet outperformed existing control models in terms of classification accuracy and model generalization ability, demonstrating its potential for authenticity identification of cigarettes.