基于云模型的变分自编码器数据压缩方法  

Variational autoencoder data compression algorithm based on cloud model

作　　者：郭秋燕 胡磊[1,2] 代劲 Guo Qiuyan;Hu Lei;Dai Jin(Information Center,The First Affiliated Hospital of Chongqing Medical University,Chongqing 400016,China;Medical Data Science Academy of Chongqing Medical University,Chongqing 400016,China;Chongqing Key Laboratory of Computational Intelligence,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆医科大学附属第一医院信息中心,重庆400016 [2]重庆医科大学医学数据研究院,重庆400016 [3]重庆邮电大学计算智能重庆市重点实验室,重庆400065

出　　处：《电子技术应用》2023年第10期96-99,共4页Application of Electronic Technique

基　　金：国家自然科学基金(61936001);2020年重庆医科大学智慧医学研究项目(ZHYX202005)。

摘　　要：图像数据解压缩问题是一类重要的数据处理问题,数据特征学习在数据压缩研究中有重要的研究价值。提出了一种基于云模型的变分自编码器特征表征模型,将云模型作为变分自编码器的先验分布,解决变分自编码器在特征表征上的局限性。变分自编码器的编码器部分负责构建数据的特征空间,通过在该空间中采样获得隐变量,完成数据压缩;解码器部分完成从数据特征到原数据的生成,即数据的解压。在人脸数据集上与原方法作实验对比,验证了该方法的正确性和有效性。Image data decompression is a kind of important data processing problem.Data feature learning has important research value in data compression research.This paper proposes a feature representation model of variational autoencoder based on cloud model,which takes the cloud model as the prior distribution of variational autoencoder and solves its limitation in feature representation.The encoder of the variational autoencoder is responsible for constructing the feature space of data,obtaining hidden variables by sampling in the space,and completing data compression.The decoder completes the generation from data features to raw data,that is the decompression of data.The correctness and effectiveness of the proposed method are verified by the experimental comparison with the original method on face dataset.

关 键 词：云模型 变分自编码器 隐变量空间 数据压缩 数据重构 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]