论文部分内容阅读
为了提高压缩数据收集对多样化传感数据的适应能力,同时抑制环境噪声对数据收集精度的影响,提出了一种优化字典学习算法来构造压缩数据收集中的稀疏字典。理论分析表明在压缩数据收集中由环境噪声导致的数据收集误差和稀疏字典的自相干程度正相关。为此在字典学习的过程中引入了自相干惩罚项来抑制环境噪声对数据收集精度的影响。该惩罚项还能减少字典学习过程中对训练数据的过拟合,从而进一步提高了该算法的稀疏表示能力。实验表明,该算法的稀疏表示能力高于同类字典学习算法,而且能有效地抑制环境噪声对压缩数据收集精度的影响。
In order to improve the adaptability of compressed data collection to diverse sensing data and to suppress the influence of ambient noise on data collection accuracy, an optimized dictionary learning algorithm is proposed to construct a sparse dictionary in compressed data collection. Theoretical analysis shows that the error of data collection caused by environmental noise in compressed data collection is positively correlated with the degree of self-coherence of sparse dictionary. Therefore, self-coherence penalty is introduced in the process of dictionary learning to suppress the influence of environmental noise on data collection accuracy. The penalty term can also reduce over-fitting training data in the process of dictionary learning, thereby further improving the sparse representation capability of the algorithm. Experimental results show that the proposed algorithm is more sparse than the same dictionary learning algorithm, and can effectively suppress the influence of ambient noise on the accuracy of compressed data collection.