基于D-ResNeXt骨干网络的小样本图像分类算法  

Few-shot Image Classification Algorithm Base on D-ResNeXt Backbone Network

作　　者：杨红菊[1,2] 翟艳峰 YANG Hongju;ZHAI Yanfeng(School of Computer and Information Technology,Shanxi University,Taiyuan 030006,China;Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education,Shanxi University,Taiyuan 030006,China)

机构地区：[1]山西大学计算机与信息技术学院,山西太原030006 [2]山西大学计算智能与中文信息处理教育部重点实验室,山西太原030006

出　　处：《山西大学学报（自然科学版）》2024年第4期761-766,共6页Journal of Shanxi University(Natural Science Edition)

基　　金：国家自然科学基金(61976128);山西省回国留学人员科研资助项目(2022-008)。

摘　　要：小样本图像分类目前是人工智能领域中非常重要的方向之一,其中基于度量学习的方法具有简洁高效的特点。针对目前图像分类中特征提取阶段所使用的骨干网络问题,现有工作大多使用传统残差网络,受数据集的影响,对类内差异大的图片特征提取效果不佳。ResNeXt为传统残差网络ResNet的升级版本,优化了在特征提取阶段准确度不高,误差较大的问题。根据其网络特点,本文设计出一种适用于小样本模型的网络变体,运用其变体作为骨干网络,提高其特征提取能力,同时结合两种注意力模块,进一步提升对图像类内相似性以及类间差异性的识别效果,减少无关因素影响,有效提升整体分类精度。Few-shot image classification is currently one of the most important directions in the field of artificial intelligence.In this area,the method based on metric learning is concise and efficient.To address the problem of the backbone network used in the feature extraction stage of current image classification,most existing works use traditional residual networks,which extracts poorly the features of images with large intra-class differences as the method is influenced by the dataset.ResNeXt is an upgraded version of the traditional residual network ResNet,optimizing the problem of low accuracy and large errors in the feature extraction stage of the traditional network.According to its network characteristics,this paper designs a network variant suitable for small sample models,which uses its variant as a backbone network to improve its feature extraction ability,and combines two attention modules to further improve the recognition effect of intra-class similarity and inter-class variability of images,reduce the influence of irrelevant factors,and effectively improve the overall classification accuracy.

关 键 词：小样本学习 图像分类 注意力机制 度量学习 残差网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]