Few-shot Image Classification Algorithm Based on Multi-scale Attention and Residual Network

Current few-shot image classification algorithm cannot sufficiently extract features, the generalization ability of the model is weak, and the classification accuracy is low. In response to this problem, a few-shot image classification algorithm was proposed with multi-scale attention and residual connection based on Relation Network in this paper. The introduction of multi-scale attention can extract more important image features, and the residual connection in the model can transfer the shallow feature information to the deep, thereby enhancing the generalization ability of the model. Compared with Relation Network, the image classification accuracy of our method was significantly improved on the MiniImageNet and Omniglot dataset. Experimental results show that the introduction of multi-scale attention and residual connection can effectively improve the accuracy of few-shot image classification.