基于空频域特征提取的小样本图像分类算法

Few-Shot Image Classification Algorithm Based on Spatial-Frequency Domain Feature Extraction

小样本学习的目的是使用极少的样本训练模型,并在有限的数据集上构建一种有效的模型,以实现对新样本的准确预测。关于小样本图像分类的研究大多只从空域的角度去提取图像的特征进行学习,且在计算相似性分数时采用单一的度量模式,极大地降低了图像分类的准确性。为此,提出了一种基于空频域特征提取的小样本图像分类算法网络(FENet),从空域和频域角度出发,提取图像特征,并结合图像到图像的度量与图像到类的度量方式,引入干扰因子,提高模型的鲁棒性和泛化性。在CUB-200-2011、Stanford-Cars、Stanford-Dogs 3个数据集上进行了大量的实验,结果表明,FENet在一定程度上能提升小样本图像分类的准确性。

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.