基于动态聚类的奇异值分解降噪方法研究

Noise reduction in singular value decomposition based on dynamic clustering

针对奇异值分解降噪中吸引子轨道矩阵重构阶次难以有效确定的问题,提出了基于非监督动态聚类算法来确定矩阵有效重构阶次的新方法。该方法利用含噪声信号的奇异谱图中表征噪声的噪声平台平缓和集中的特性,通过向谱图纵轴投影,应用动态聚类合理确定噪声平台的边界,进而有效地确定奇异值分解降噪中矩阵的有效重构阶次。仿真结果表明,该方法有较好的降噪精度和算法稳定度,提高了算法的实用性。

The order of the optical reconstruction rank of the trajectory matrix is difficult to determine for noise reduction in singular value decomposition. A new method is proposed to solve for solving this problem based on unsupervised dynamic cluster. Since those singular values representing the noise is relatively centralized, through projecting the singular spectrum line to the ordinate axis, the border of noise value can be determined by dynamic clustering algorithm. And then the optical reconstruction rank can be determined. Simulation results show that this method not only has good noise reduction accuracy and stability, but also controls the complexity of algorithm effectively.