Few-Shot Learning Based on Multi-Scale Fusion of Covariance Metric Matrix
For the image classification problem with limited training samples,the existing deep neural network-based models have exposed many problems.For example,traditional algorithms based on metric learning usually rely on the pre-defined distance measurement,which makes it difficult for the model to capture the differences among classes when there are too many classes to be classified.Moreover,they only focus on the relational calculation of first-order statistics and the utilization of high-level semantic information,and ignore the low-level but abundant visual features.To address above issues,in this paper,a multi-scale fusion al-gorithm is proposed based on the covariance metric matrix.The second-order statistics under different scale is used to update model parameters.Experimental results show that the proposed model can effectively improve the accuracy of few-shot image classifica-tion,which indicates its promising value in real-world applications.
国家科技期刊平台
NSTL
万方数据
