Research on the problem of image multi-classification

Abstract

Image classification is an important research topic in the computer vision area. The classification task of small data volume, large number of classes, smaller differences between classes and larger differences within classes has always been a challenge. And this article aims to study and discuss such problems, using three methods: ResNets, EfficientNet and MoCo V2 for experiment and continuous optimization, and got the following results: The accuracy rate of ResNets classification is 0.43, of EfficientNet classification is 0.47, and of MoCo V2 classification, it is 0.62.