基于图像变形网络的小样本图像分类算法研究  

Research on Image Deformation Network for Few-shot Learning Image Classification Algorithm

作　　者：杨飚[1] 周芷晴 

机构地区：[1]北方工业大学城市道路交通智能控制技术北京市重点实验室,北京100144

出　　处：《工业控制计算机》2022年第5期86-88,共3页Industrial Control Computer

摘　　要：传统的图像分类算法在数据集过小的情况下分类准确率不高,且传统的图像变形方法容易破坏数据主体语义信息。基于图像变形网络的小样本图像分类算法研究中,采用端对端的方式结合图像变形网络和小样本图像分类网络,通过加权融合训练图像和相似图像的方式实现了对原有数据集的有效扩充,利用数据增强提高了小样本图像分类的准确率。实验数据表明,提出的方法在mini-ImageNet数据集上对小样本图像分类网络的性能有较好的提升效果。The traditional image deformation method is easy to destroy the semantic information of the data subject,and the traditional image classification algorithm has a greater impact on the classification effect and classification accuracy when the data set is too small.In this paper,based on the few-shot learning image classification algorithm of image deformation network,the end-to-end method is used to combine the image deformation network and few-shot learning image classification network, and the original data set is effectively expanded by the weighted fusion of training images and similar images,using data enhancement to improve the accuracy of few-shot learning image classification.A large amount of data shows that this method has a good effect on improving the performance of the few-shot learning image classification network on the mini-ImageNet data set.

关 键 词：数据增强 图像变形 小样本学习 图像分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]