基于ViT的中欧班列集装箱Logo图像分类识别研究

Research on Classification and Recognition of Container Logo Image of China-Europe Train Based on ViT

摘要: 随着“一带一路”、长江经济带等国家战略的叠加实施，中国与西方的贸易也呈直线上升的趋势。伴随着中欧班列高质量运行，中欧班列集装箱的分类统计也成为中国对外贸易的一大重点。集装箱上代表着各供应商的logo图像在运输途中会产生褪色破损情况从而给识别增加了很多的难度。因此，公司为了有效识别各个标识，提出了基于ViT的logo图像分类模型，使用激活函数GELU代替传统的RELU。实验表明，改进后的模型可以很好地分类识别，准确率达到98%，且优于其他的分类模型。

Abstract: With the overlapping implementation of national strategies such as the “the Belt and Road” and the Yangtze River Economic Belt, China’s trade with the West is also on the rise. With the high- quality operation of the China-Europe train, the classified statistic of the containers of the China- Europe train has also become a major focus of China’s foreign trade. The logo image representing each supplier on the container will be faded and damaged during transportation, which adds a lot of difficulties to the identification. Therefore, in order to effectively identify each logo, the company proposed a logo image classification model based on ViT, using the activation function GELU instead of the traditional RELU. The experiment shows that the improved model can classify and recognize well, with an accuracy of 98%, and is superior to other classification models.