A Variational Bayesian Multisource Data Fusion Method Based on L2 Second Order Penalty Combined with Entropy Empirical Pool

Data fusion has become an important technology in all fields to cope with the current big data Era. Current data-level research algorithms have different degrees of flaws. The main purpose of this paper is to solve data sparsity and loss of information caused by L1 norm distance measurement in data-level variational Bayesian inference model fusion. Based on L2 norm second-order penalty, a variational Bayesian inference model is proposed. The input of the model is updated twice by combining an empirical module of entropy values. The Total Variation (TV) penalty factor is applied to keep the sparsity of signal mutations. The commonly used data level fusion algorithm is compared with the method proposed in this paper by using simulated biological signals. The experimental results show that the improved model method presented in this paper can effectively preserve the maximum amount of information and eliminate redundant information. Under the constraints of L2 norm model, the model has a good ability to prevent overfitting and a certain ability to generalize the model.