摘要
基于K-SVD字典学习算法(K-singular value decomposition,K-SVD)的压缩感知技术应用在林区微环境监测站中,可极大地减少数据传输数量,从而降低监测站的使用能耗,延长监测站的使用寿命。本文选用空气温度作为实验对象,验证算法的可行性,并与前人提出的基于离散傅里叶变换基(Discrete fourier transform,DFT)的压缩感知方法进行对比实验。实验结果表明,当稀疏度k相同时,K-SVD算法的平均稀疏化误差始终小于DFT算法,且误差分布范围更加集中;当稀疏度和压缩率均相同时,K-SVD算法的平均重构误差也始终小于DFT算法,且误差分布范围更加集中。在林区微环境监测站中,K-SVD算法具有更好的稀疏表示性能以及重构性能,在降低相同系统能耗的同时,也降低了数据传输的误差。
The application of compression sensing technology was put forward based on K-SVD dictionary learning algorithm in forest microenvironment monitoring station. It would greatly reduce the number of data transmission, thus reducing the use of monitoring stations and prolonging the service life of monitoring stations. The air temperature data was used as the experimental object,and it verified the feasibility of the algorithm,and compared with the previous proposed compression sensing technique based on the discrete Fourier transform base. The experimental results were as follows: when the sparsity was the same,the average sparsity error of the learning dictionary based on K-SVD algorithm training was always less than that of the DFT dictionary when the original signal was sparsely represented,and the error distribution range was concentrated and basically lower than the median line of the DFT dictionary.When sparsity and compression ratio were the same, the average reconstruction error of learning dictionary was always smaller than that of DFT dictionary,and the error distribution range was more concentrated. To sum up,in the forest micro-environment monitoring station,the dictionary trained by the K-SVD algorithm had better sparse representation performance and reconstruction performance,which can reduce the operation power of the monitoring station while reducing the error of data transmission.
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2018年第S1期365-371,共7页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家重点研发计划项目(2017YFD0600901)
北京市科技计划项目(Z161100000916012)
北京市共建项目
关键词
微环境监测站
压缩感知
K-SVD
字典学习
稀疏表示
microenvironment monitoring station
compressed sensing
K-SVD
dictionary learning
sparse representation