基于流形正则的堆叠胶囊自编码器优化算法  

Stacked Capsule Autoencoders Optimization Algorithm Based on Manifold Regularization

作　　者：王鲁娜 杜洪波[1] 朱立军[2] WANG Luna;DU Hongbo;ZHU Lijun(School of Science,Shenyang University of Technology,Shenyang Liaoning 110870,China;School of Information and Computing Science,Northern University for Nationalities,Yinchuan Ningxia 750021,China)

机构地区：[1]沈阳工业大学理学院,辽宁沈阳110870 [2]北方民族大学信息与计算科学学院,宁夏银川750021

出　　处：《广西师范大学学报（自然科学版）》2023年第2期76-85,共10页Journal of Guangxi Normal University:Natural Science Edition

基　　金：国家自然科学基金(11861003);辽宁省教育厅高等学校基本科研项目(LJKZ0157)。

摘　　要：针对堆叠胶囊自编码器存在检测性能慢、不能更好挖掘图像局部特征的问题,本文提出基于流形正则的堆叠胶囊自编码器优化算法。采用Scharr滤波器对堆叠胶囊自编码器模型中的图像进行重建,加强图像目标检测的精度,并在损失函数中引入流形正则项,从而加强对原始数据空间局部特征的提取,最终使用基于流形正则的堆叠胶囊自编码器学习参数,选择出更加具有区别性的特征。在MNIST和Fashion MNIST数据集上的实验结果显示,该优化算法相比于原网络结构,图像分类准确率分别提高了0.26和9.23个百分点,且模型训练速度也得到较大提高。To solve the problem that stacked capsule autoencoder has slow detection performance and cannot better mine local features of images,an optimization algorithm of stacked capsule autoencoders based on manifold regularization was proposed.Firstly,by using the Scharr filter to reconstruct the image in the stacked capsule autoencoders model,the accuracy of image target detection was enhanced.Then,a manifold regular term was introduced into the loss function to enhance the extraction of local features in the original data space.Finally,the stacked capsule autoencoders based on manifold regularization was used to learn parameters to select more discriminative features.Experiments results on MNIST and Fashion MNIST datasets show that compared with the original network structure,the accuracy of image classification is improved by 0.26%and 9.23%,respectively,which greatly improves the training speed of the model.

关 键 词：深度学习 图像分类 堆叠胶囊自编码器 流形正则 滤波器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]