基于自旋图结构权重分配的图信号突变点检测技术

Technology based on self-spin graph structure weight allocation for graph signal abrupt change points detection

对于不规则的、复杂的信号进行分析处理,新兴的图信号处理技术已经取得了研究突破。在无线传感器网络的信号突变点检测场景中,图信号处理技术也已被证明是有效的。在图信号突变点检测系统的基础上,文中着眼于检测结果的精确性和稳定性问题,首先研究了已有的不同图拉普拉斯矩阵和图的边权重定义方式,发现较采用组合拉普拉斯矩阵和维度倒数定义权重的方法有一定程度的性能提升;进而研究并新定义了一种自旋图结构并讨论了最佳的权值分配方案,通过实验证明了新方案能够显著减小检测系统的延迟,进一步提高突变点位置检测精度;最后与经典的DSP方法进行比较,并研究了图信号多次突变检测的情况,进一步验证了改进方案的优越性。

The emerging filed of research on the graph signal processing technology has made breakthroughs for analysis and processing of irregular and complex signals.In the signal abrupt change point detection scenario in wireless sensor networks,graph signal processing techniques have also proven to be effective.Based on the basic detection system of the signal abrupt change point detection,this paper focuses on the accuracy and the stability of the detection results.Firstly,the existing graph Laplace matrices and the method for defining the edge weights are discussed.It is found that there is a certain degree of the performance improvement over the traditional method by using the combined Laplace matrix and the weight defined by reciprocal of dimension.Then,a new self-spin graph structure is defined and an optimal weight assignment scheme is discussed.It is proved by experiments that the new scheme can significantly reduce the delay of the detection system and improve the accuracy of the position of an abrupt change point.Finally,compared with the classic DSP methods,the superiority of the improved scheme is verified by studying multiple abrupt change point detection.