基于稀疏滤波的无线传感器网络数据融合

Data fusion in wireless sensor network based on sparse filtering

在无线传感无器网络中,为了提高数据融合算法的性能,设计了一种基于深度学习的二值化层叠稀疏滤波模型(BSSFM),BSSFM将若干稀疏滤波器层叠并对权值参数进行二值化从而能快速有效的提取高维数据特征,之后将BSSFM和分簇协议相结合提出数据融合算法BSSFMDA,BSSFMDA首先在汇聚节点逐层训练BSSFM并对网络分簇,簇节点利用BSSFM进行数据特征提取,之后簇首将分类融合后的特征发送至汇聚节点。仿真实验表明,和SOFMDA等算法相比,BSSFMDA在模型训练时间、特征提取速度、正确率以及节点能耗等方面的表现均更加优异。

In order to improve the performance of data aggregation used in wireless sensor network , Binarized Stacked Sparse Filtering Model （ BSSFM） based on the deep learning is designed .BSSFM stacks several sparse fil-ters and binaries the weight parameters to extract the high-dimensional data features quickly and efficiently , and then a data fusion algorithm called BSSFMDA is proposed , which combines BSSFM and wireless sensor network clustering routing protocol.BSSFMDA trains BSSFM layer by layer in sink node and generates clusters in network , node data features will be extracted in cluster nodes and classified in cluster heads , and then cluster heads send the features got by fused in the same class to sink node .The simulation results show that compared with other algo-rithms such as SOFMDA, BSSFMDA has a better performance in training time of model, feature extraction speed, accuracy and the node energy consumption .