Message transfer-based pattern recognition filtering method

Abstract

The present invention relates to the technical field of signal filtering, and particularly relates to a message transfer-based pattern recognition filtering method. The filtering method includes a first step of initializing parameters; a second step of performing iterative calculation, determining whether data number N is greater than 3, if not, directly outputting original data, if yes, performing forward prediction and backward smoothing simultaneously; a third step of performing forward and backward joint estimation, updating a mean value and a variance, and obtaining a joint estimation value of a state variable mean value; and a fourth step of determining whether an iteration number of times k is equal to K, if yes, indicating that iteration is completed, outputting a state variable estimation value; if not, indicating that iteration is not completed, updating two expressions shown in the specification, then executing the second step, that is, continuing iteration calculation. Adaptive recognition is performed on recognition parameters in a system dynamic model by using a method of maximizing expectations on a factor diagram, so that the problem of performance degradation of a standard Kalman filtering algorithm caused by modeling errors and system model change is solved.

Description

technical field [0001] The invention relates to the technical field of signal filtering, in particular to a pattern recognition filtering method based on message transmission. Background technique [0002] The classic Kalman filter can obtain good estimation performance. A necessary condition is to establish an accurate dynamic model and observation model, which requires a clear understanding of changes in objects and possible abnormal interference, and the ability to establish and objective system Accurately conform to the dynamic equation, but due to changes in the working environment and use conditions, the statistical characteristics of the noise are often not deterministic, and the observations that deviate from the ideal assumption or the dynamic model that deviates from the ideal assumption will inevitably bring unpredictable results to the dynamic filtering results. Imaginary biases may even make the Kalman filter diverge. [0003] In order to overcome this shortcom...

AI Technical Summary

Problems solved by technology

[0004] The invention provides a pattern recognition and filtering method based on message transmission, which uses the expectation maximization method on the factor graph to carr...

Images