摘要
深度学习在图像识别任务中的表现依赖于数据集的大小,当样本稀缺时,模型难以获得优异的成绩。针对如何在少量数据的条件下训练出表现优越的识别模型这一问题,受生成对抗网络的启发,文章提出了一种基于多域数据扩充的小样本识别模型。该模型通过已有数据集训练生成模型,生成用于扩充数据集的伪样本。再利用这些扩充样本与真实样本协同训练小样本识别模型。实验结果表明,所提方法在识别准确率与模型训练的稳定性上有一定的提升。
The performance of Deep Learning in image recognition tasks depends on the size of the dataset.When the samples are scarce,the model is difficult to achieve excellent results.Aiming at the problem of how to train a superior recognition model under the condition of a small amount of data,inspired by the Generative Adversarial Networks,this paper proposes a Few-Shot Learning model based on multi-domain data expansion.The model generates a model through the training of existing datasets,and generates pseudo-samples for expanding the datasets.Then these expanded samples and real samples are used to train the small sample recognition model coordinately.The experimental results show that the proposed method has a certain improvement in recognition accuracy and stability of model training.
作者
陈琪
徐长文
董非非
李正
CHEN Qi;XU Changwen;DONG Feifei;LI Zheng(Jiangxi Earthquake Agency,Nanchang 330026,China)
出处
《现代信息科技》
2025年第3期61-67,共7页
Modern Information Technology
基金
江西省防震减灾与工程地质灾害探测工程研究中心、江西九江扬子块体东部地球动力学野外科学观测研究站2022年度开放基金项目(SDGD202217)。
关键词
小样本学习
多域
数据扩充
生成对抗网络
Few-Shot Learning
multi-domain
data expansion
Generative Adversarial Networks
