A Multi-Scale Feature Attention Image Recognition Algorithm

Abstract

The accuracy of small-sample image classification depends on the ability of neural network models to extract image representations from sample data. A small-sample image classification system based on attention mechanisms and meta-learning is suggested in order to extract more comprehensive information from pictures. The approach may extract richer multiscale features from pictures and enhance classification outcomes through meta-learning because the attention mechanism of multiscale features can concentrate on the data in the sample feature space. In order to demonstrate the efficacy of the suggested strategy, tests are conducted on the two industry-standard datasets miniImageNet and tieredImageNet for both 5-way 1-shot and 5-way 5-shot tasks. The results are compared with the best existing methods.