Solution for Ill-posed Msplit Model Regularization of Multi-source Heterogeneous Data Fusion

Abstract Traditional methods of heterogeneous data fusion need prior information, although the method based on M split estimation is proposed in the absence of prior information, it ignores the ill-posed condition. In this paper, we analyze dynamic changes of ill-posedness state of the model resulting from data classification, and by constructing a scheme for consistency checking of parameters estimation, we develop a dynamical detection method for the regularization of ill-posed model. An example is presented to demonstrate that the ill-posed state of M split model is unstable, and the ill-posed model cannot be regularized through traditional approaches using fixed regularization parameters, the disturbance effect of the ill-posed M split model on parameters estimation and data classification is also demonstrated, and the efficiency of the proposed method for heterogeneous data fusion based on ill-posed M split model is verified.