Image classification method based on self-attention mechanism

Abstract

The invention provides an image classification method based on a self-attention mechanism, and the method comprises the steps of constructing a Transform model containing the self-attention mechanism, and adding a classifier unit for an image classification task; processing the public data set ImageNet, and adjusting the original picture to a proper size; dividing the adjusted picture into sub-pictures with fixed sizes, connecting the sub-pictures, and performing dimension adjustment to obtain a picture embedding vector; performing two-dimensional position coding to obtain a two-dimensional position coding vector, and connecting the two-dimensional position coding vector with the picture embedding vector to serve as model input; and sending the connected vectors to a Transform model, extracting picture features, converting the vectors output by the model into probability representation through a classifier unit during final decoding, and completing image classification. By using the self-attention mechanism, the global information, namely the picture features extracted by the traditional convolutional neural network, can be effectively extracted from the picture, and the picture classification can be effectively completed based on the extracted features.

Description

technical field [0001] The invention relates to an image classification technology, in particular to an image classification method based on a self-attention mechanism. Background technique [0002] Image classification is a pattern classification problem. Its goal is to divide different images into different categories and achieve the minimum classification error. Its typical method is to extract the features of the image, and assign the classification label to the image based on the features. Image classification tasks have experienced decades of development from traditional methods to deep learning-based methods. The current method is mainly based on the convolutional neural network structure to extract image features, and on the basis of the network, operations such as deepening the number and depth of the model and improving the convolution method are adopted. The change of methods has made the performance of basic image classification tasks close to saturation, and t...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the inadequacy of the attention mechanism in the prior art to replace the convolutional network, and the incomplete attempt to introduce the advantages of the Transformer model into the image field, the technical problem to be solved by the present invention is to provide a self-attention mechanism based Image classification method, exploring the application of Transformer structure in image classification tasks

Images