Research on Online Classification and Recognition of Cigarette Brands Based on Convolutional Neural Network Combination Algorithm

Abstract

To explore the relationship between tobacco strip online near-infrared spectra and ciga⁃ rette brand identification， a cigarette brand classification method based on the ResNeXt18-CNN-LightGBM hybrid model is proposed. Firstly，the collected tobacco strip sample online spectral data are preprocessed，and the ResNeXt18 network model is used to extract initial features from the pre⁃ processed spectra. Then，the extracted features are input into a custom 3-layer CNN network model for secondary feature extraction. Finally，the features extracted by the CNN are fed into a LightGBM classifier for brand classification training. The results show that the classification accuracy of tobacco strip brand identification in the ResNeXt18-CNN-LightGBM model reaches 97%. Compared with tra⁃ ditional single chemometrics algorithms，the proposed cigarette brand classification method based on a convolutional neural network combination algorithm is simple，accurate，and stable. It can be ap⁃ plied to the online identification of cigarette brands in cigarette manufacturing，with significant impli⁃ cations for cigarette brand management，production quality evaluation，and cigarette quality control.