Recognition of cigarette brand based on deep learning

In order to meet the demands of automatic collection of cigarette retail data, a method for cigarette brand recognition was proposed on the basis of deep learning. Firstly, design a detection model to detect the position and posture of cigarette packet image, and adjust the packet to right its image. Secondly, develop a feature extraction model to extract the feature of the image. Finally, generate a Euclidean distance threshold to search the cigarette feature with the highest similarity via the threshold from the cigarette feature database, and the cigarette brand associated with that feature was taken as the cigarette brand to be identified. The proposed method was tested in some cigarette retail stores in Nanning City of Guangxi Province, the results showed that the accuracy of the method for cigarette brand recognition was 98.0%, which indicated that the proposed method could meet the demands of automatic collection of cigarette retail data and had good generalization performance.