摘要： 针对深度学习中使用少量样本完成模型训练的小样本学习问题，构建一种双路的特征聚合网络，并提出一种新的综合损失函数对网络模型的参数更新过程加以控制。通过综合损失函数，特征聚合网络可将样本映射到更具代表性的特征空间中，从而获得更优的特征分布。实验结果表明，与MN、PN等方法相比，该特征聚合网络能有效降低特征空间的复杂性，提升整体模型的分类效果。

关键词: 深度学习, 小样本学习, 特征提取, 度量学习, 特征聚合

Abstract: To use a small number of samples to complete the model training in deep learning,this paper proposes a two-way feature aggregation network and gives a new comprehensive loss function to control the parameter update process of the model.By integrating the loss function,the feature aggregation network can map the sample to a more representative feature space,thereby obtaining a better feature distribution.Experimental results show that compared with MN,PN and other methods,this feature aggregation network can effectively reduce the complexity of the feature space and improve the classification effect of the overall model.

Key words: deep learning, small sample learning, feature extraction, measure learning, feature aggregation

中图分类号: 

• TP18