Deep learning network based on group convolution feature topological space and training method thereof

Abstract

The invention discloses a deep learning network based on a group convolution feature topological space. The deep learning network comprises a convolution feature extraction layer, a group convolutiontopological layer and a deep feature recognition layer. The convolution feature extraction layer is used for extracting multi-channel CNN convolution features of the sample data and taking an extraction result as input of the group convolution topology layer; the group convolution topology layer is used for combining the extracted multi-channel CNN convolution features, forming group convolution according to group classification by using channel indexes, constructing a graph topological space, regarding each group convolution feature as a graph topological space node, automatically / manually constructing a graph topological space node connection rule, generating a Laplace matrix L, and taking the Laplace matrix L as input of a depth feature recognition layer; and the depth feature recognition layer is used for outputting group convolution feature topological space diagram features corresponding to the sample data according to the input Laplace matrix L. According to the method, graph topology space rules of CNN features under different channels can be given, so that the traditional CNN training and convergence speed is increased.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a deep learning network based on group convolution feature topological space and a training method thereof. Background technique [0002] Convolution Neural Network (CNN) is one of the representative algorithms of deep learning, which includes convolution calculation and neural network with deep structure. Most of today's image recognition methods use CNN to automatically extract image features, thus realizing the transformation from the experience-driven artificial feature paradigm to the data-driven representation learning paradigm. However, CNN also has the following disadvantages: (1) training requires a large number of samples, and the training speed is very slow; (2) the convolution kernel only extracts local features, and the final features used for recognition are the aggregation of local features, which lacks global features to some extent. spati...

AI Technical Summary

Problems solved by technology

However, CNN also has the following disadvantages: (1) training requires a large number of samples, and the training speed is very slow; (2) the convolution kernel only extracts local features, and the final features used for recognition are the aggregation of local features, which lacks global features to some extent. Spatial features

However, how to mine the value behind the topology of the graph and achieve rapid training and convergence around the topological spatial relationship of multi-channel CNN convolution features has not yet been seen.

Images