Abstract: Aiming at the long-tail distribution problem of fine-grained image classification task, a multi-scale feature Transformer based fine-grained image classification method is proposed to protect the underlying and deep features and optimize the long-tail distribution. First, a hybrid data sampling method is designed to obtain the ternary data for optimizing the representation learning, long-tail distribution and fine-grained features. Then, the Transformer multi-scale feature optimization method is designed to optimize the feature learning process by the bottom feature comparison learning method and the deep feature balance learning method, respectively, to improve the category confusion and fine-grained feature extraction, and to increase the attention to the tail category while protecting the head category feature learning. Simulation results show that the proposed method can effectively improve the impact of the long-tail distribution in fine-grained image classification tasks, optimize the feature distribution, and improve classification accuracy.

Key words: Transformer , fine-grained image classification , fine-grained feature , long-tail distribution