Few-Shot Image Classification Algorithm Based on Spatial-Frequency Domain Feature Extraction
The purpose of few-shot learning is to train a model with very few samples and build an effective model on a limited dataset to achieve accurate prediction of new samples.Most studies on few-shot image classification only extract image features from the perspective of spatial domain for learning,and use a single measurement mode when calculating similarity scores,greatly reducing the accuracy of image classification.To this end,a few-shot image classification algorithm network(FENet)based on spatial-frequency domain feature extraction is proposed.From the perspectives of spatial and frequency domains,image features are extracted,and combined with image to image and image to class metrics,interference factors are introduced to improve the robustness and generalization of the model.A large number of experiments were conducted on the CUB-200-2011,Stanford Cars,and Stanford Dogs datasets,and the results showed that FENet can improve the accuracy of few-shot image classification to a certain extent.
NETL
NSTL
万方数据
