基于潜在变量二元回归模型的多传感器数据融合

针对目前数据融合算法存在的置信度无法获取的问题,提出了一种基于潜在变量二元回归模型(LatentVariable Binary Regression Model)的多传感器数据融合算法。将每个传感器获取的特征值作为多变量回归模型中的相关变量,通过Gibbs抽样得到潜在变量的分布概率,确定多变量回归模型中的表征量作为融合结果,并以潜在变量的分布概率作为融合结果的置信度。基于实地采集的运动目标震动信号进行仿真实验,结果表明该融合方法拥有较好的识别效果,同时能够给出识别结果的置信度。其中错分类的结果具有较低的置信度,可以提醒观测者做进一步的观察。

Multi-sensors Data Fusion Based on Latent Variable Binary Regression Model

Aimed at the problem of that the existing data fusion algorithm can not get degree of confidence,a multi-sensors data fusion algorithm based on Latent Variable Binary Regression Model(LVBRM) is proposed.The features obtained by each sensor are considered as covariates of the multivariate regression model.The indicator of the multivariate regression model,which can be taken as fusion result,is determined by the probability distribution of latent variable from Gibbs sampler,and the probability distribution of latent variable is considered as the degree of confidence of fusion result. Based on real tremor signals of moving targets,experiment results indicate that the fusion method has a good recognition result,which is able to give the degree of confidence of recognition result.The wrong result has a low degree of confidence which remind the observer for further observation.