Multi-Scale Feature Fusion for Incomplete Cigarette Code Recognition

A method for recognizing incomplete cigarettes with disordered backgrounds and incomplete characters based on multi-scale feature fusion networks is proposed.The method is able to train or apply in an end-to-end manner.Firstly,the multi-scale fusion features are extracted from the image using the feature extraction network.Then,a region optimization module(ROM)is proposed to optimize the extracted features.The recognition and localization network learns these optimized features to perform the recognition and localization task more robustly.Finally,a matching algorithm is used to match the recognition and localization results,and the final results are obtained.The experimental results show that,compared with the existing methods,the proposed method has an overall improvement in recognition accuracy and efficiency for the task of identifying complete cigarette codes.In the experimental of transcribing incomplete cigarettes code,proposed method is more efficient than the manual method.