Convolutional neural network non-local information construction method

Abstract

The invention provides a convolutional neural network non-local information construction method. According to the invention, the method includes extracting convolutional features of an image by usinga common convolutional neural network; calculating a pixel-level global attention graph of the image by adopting a self-attention mechanism; constructing a global graph structure in the image throughthe attention graph; adopting a graph convolutional neural network to extract non-locality graph features on a global graph structure in the image; fusing the image features and the convolution features of the image by adopting a matrix multiplication method, and inputting the fused features into a subsequent network. According to the method, local features under a fixed receptive field of the image can be extracted, non-local features can also be extracted, errors of the convolutional neural network in image feature extraction are reduced, and image generation and classification effectivenessis remarkably improved.

Description

technical field [0001] The invention belongs to the technical field of convolutional neural networks, and in particular relates to a method for constructing non-local information of convolutional neural networks based on a self-attention mechanism and graph convolution. Background technique [0002] With the advent of the era of big data, neural networks have been applied to various fields of artificial intelligence, such as: image recognition, automatic translation, unmanned cars, etc. Among them, the convolutional neural network occupies an increasingly important position in computer vision, and has become an important method for extracting image features. The convolution operation in the convolutional neural network uses convolution filters for parameter sharing and sparse connections between layers to extract convolutional features under a fixed receptive field. These convolutional features will be input into the subsequent network to complete specific tasks, such as fa...

AI Technical Summary

Problems solved by technology

[0004] The invention provides a method for constructing non-local information of a convolutional neural network, which overcomes that the image features extracted by the existing convolutional neural network are local features in a fixed receptive field, and the non-local information is missing, resulting in large errors and low effectiveness of the convolutional neural network. Lowering the problem

Images