Composite neural network model and modeling method thereof

Abstract

The invention discloses a composite neural network model and a modeling method thereof. The composite neural network model comprises a full-connection sparse modeling module, an input mapping single-layer sensor layer, a dictionary learning single-layer sensor layer and a feedback path module. The full-connection sparse modeling module is used for completing sparse modeling of a single sample in each round of iteration of the system; wherein the input mapping single-layer sensor layer is used for generating continuous external input required by the full-connection sparse modeling module, the dictionary learning single-layer sensor layer is used for realizing learning and optimization of a sparse dictionary, and the feedback path module is used for assisting the full-connection sparse modeling module to improve the sparse quality of modeling. The method has the advantages that through a full-connection working mechanism of the Hopfield neural network and a weight learning mechanism of the sensor neural network model, integrated data-driven sparse modeling and dictionary learning are realized, and a modeling result with better globality and better sparse performance is obtained.

Description

technical field [0001] The invention relates to the technical field of composite neural network models, in particular to a composite neural network model and a modeling method for high-sparse quality sparse modeling. Background technique [0002] As an important method of information compression and feature extraction in the field of signal processing, compressed sensing (sparse decomposition) technology has been widely researched and applied in recent years. This kind of method can express the signal as a small number of sparse decomposition dictionaries. The linear combination of the structure coefficient and the dictionary atom realizes the compressed storage and transmission of the signal, and it can also be used as an effective means of signal analysis and feature extraction. Existing compressive sensing sparse modeling algorithms, such as K-SVD and its improved algorithms, often use l 1 Norm or other approximate sparsity constraint term as l 0 An optimal convex appro...

AI Technical Summary

Problems solved by technology

Although these methods can obtain sparse modeling results and dictionary learning results, they are limited by the solution algorithm. The sparse constraint items of the solution paradigm can only be used as a constraint condition rather than an optimization object to participate in the solution process. The sparsity limit It is often not sufficient, and the globality of the solution process will also be affected. The commonly used greedy solution algorithm will introduce a variety of problems that lead to the degradation of modeling quality, and it is difficult to obtain dictionary learning results that are close to the true components of the training sample data set.

Images