摘要
多源传感器数据具有非线性、高维度的特征,因此难以准确分类,直接进行数据融合后的噪声较大,可用性降低,为此提出一种多源传感器数据层叠分类降维融合方法。设计基于深度学习的层叠自动降维分类器(SAESM),将SAESM和Softmax分类器结合在一起,在簇内完成源传感器数据特征提取并区分数据属性类别。针对不同类别数据分类后构成的集合,分配一个可以代表数据类别的簇首节点,统一传输给汇聚节点。汇聚节点对簇首节点整合的信息表进行参数融合处理,完成多源传感器数据融合。实验分析结果表明:针对多源传感器数据特征提取分类正确的样本数量较高,融合后噪声数据量得到有效降低。
It is difficult to classity multi-source sensor data accurately which have nonlinear and high-dimensional characteristics.After direct data fusion,the noise is large and the usability is reduced.Therefore,a multi-source sensor data cascade classification dimension reduction fusion method is proposed.A cascade automatic dimensionality reduction classifier(SAESM)based on deep learning is designed.SAESM and Softmax classifier are combined to extract the data features of source sensors and distinguish the data attribute categories in the cluster.A cluster head node that can represent the data category is allocated to the collection composed of different categories of data and information of all the sensor nodes is integrated by the head node and transmitted to the sink node uniformly.The sink node performs parameter fusion processing on the information table integrated by the cluster head node to complete multi-source sensor data fusion.The experimental results show that the number of correctly classified samples with according to multi-source sensor data feature extraction is high,and the amount of noise data is effectively reduced after fusion.
作者
叶成景
郭海涛
陈红玲
杨叶芬
YE Chengjing;GUO Haitao;CHEN Hongling;YANG Yefen(College of Robotics,Guangdong Polytechnic of Science and Technology,Zhuhai Guangdong 519090,China;School of civil engineering and transportation,South China University of technology,Guangzhou Guangdong 510640,China)
出处
《传感技术学报》
CAS
CSCD
北大核心
2022年第8期1117-1122,共6页
Chinese Journal of Sensors and Actuators
基金
广东省教育科学“十三五”规划2019年度高校哲学社会科学专项研究项目(2019GXJK272)
广东科学技术职业学院2021年度校级科研项目(XJJS202104)。
