基于卷积神经网络的网络小样本数据特征提取  

Feature Extraction of Network Small Sample Data Based on Convolutional Neural Network

作　　者：张勇 林肖莹 ZHANG Yong;LIN Xiaoying(School of Computer Science,Minnan Normal University,Zhangzhou Fujian 363000,China)

机构地区：[1]闽南师范大学计算机学院,福建漳州363000

出　　处：《佳木斯大学学报(自然科学版)》2024年第12期30-32,36,共4页Journal of Jiamusi University:Natural Science Edition

摘　　要：针对现行的数据特征提取方法存在的错提率较高和可信度较低问题,研究应用基于卷积神经网络设计一种新的网络小样本数据特征提取方法。通过将网络小样本数据在Hibert空间中映射,调整数据相对熵,实现对数据的均衡处理,利用生成对抗网络对小样本数据扩充并修正,利用卷积神经网络对扩充后的数据特征提取,实现基于卷积神经网络的网络小样本数据特征提取。经实验证明,设计方法错提率不超过1%,可信度在0.9以上,可以实现网络小样本数据特征精准提取。In response to the problems of high false extraction rate and low credibility in current data feature extraction methods,this study applies a new network small sample data feature extraction method designed based on convolutional neural networks.By mapping small sample data in the Hilbert space and adjusting the relative entropy of the data,balanced processing of the data is achieved.Generative adversarial networks are used to expand and correct the small sample data,and convolutional neural networks are used to extract features from the expanded data,achieving feature extraction of network small sample data based on convolutional neural networks.Experimental results have shown that the error rate of the design method does not exceed 1%,and the reliability is above 0.9,which can achieve accurate feature extraction of small sample data in the network.

关 键 词：卷积神经网络 网络 小样本数据 特征提取 HIBERT空间 生成对抗网络 

分 类 号：TP387[自动化与计算机技术—计算机系统结构]