Filter construction method based on cyclic neural network

Abstract

The invention discloses a filter construction method based on a cyclic neural network, belonging to a digital filtering technology, which mainly solves the problems of simple model, low complexity, poor universality and inability to learn of the existing filtering technology. Firstly, the filtering input and output and the filtering order are determined. Then a multi-level joint residual filteringnetwork is constructed with the cyclic neural network as the basic unit, and a training data set of the joint residual filtering network is constructed by widely collecting data. Then, the training data set is used, a neural network training method is adopted, each level of filtering network unit weight parameters is trained and optimized in sequence from top to bottom until the whole network istrained. Finally, according to the filtering output requirements, output results of the correlated layer residual filtering network structure are selected as filtering results of the whole filtering network to output, which can achieve filtering. The filter construction method has the advantages of wide application range, many adaptable scenes, good practical effect and so on and can be used to solve the filtering problems in various fields.

Description

technical field [0001] The present invention relates to digital filtering technology, more specifically, the present invention relates to a filter construction method based on cyclic neural network, which is suitable for target state estimation and other filter estimation problems in target tracking. Background technique [0002] Digital filtering is a data processing technique that removes noise and restores real data. Digital filters can be divided into two parts: classic filters and modern filters. The classical filter assumes that the useful components in the input signal and the desired filtered components are located in different frequency bands, and the noise is filtered through a linear system, including high-pass filters, low-pass filters, band-pass filters, and band-stop filters. Many different types, but not suitable for filtering where the signal and noise spectra alias into each other. The modern filter realizes the filter estimation of the signal by treating ...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to construct a general-purpose, high-efficiency, and trainable filter, aiming at overcoming the defects of existing filtering technology, and solving the problems of simple model, low complexity, poor adaptability and inability to learn in existing filtering technology

Images