Small sample image classification method and system based on self-supervision enhancement

Abstract

The invention discloses a small sample image classification method and system based on self-supervised enhancement, and the self-supervised learning and small sample learning are both used for relieving the dependence of a model on label data. On the basis of a small sample learning method based on a graph neural network, a self-supervised enhanced small sample learning method is provided in combination with self-supervised learning, a self-supervised learning task for cutout position prediction is designed, random cutout is performed on all samples of each small sample classification task, and after sample features are extracted, the position of an image block to be cut out of each sample is predicted through a full connection layer. According to the method, the matting position prediction task and the small sample classification task are jointly trained to enhance the effective representation extraction capability of the model, so that the classification result of the model is improved, and the effectiveness of the method is proved through comparison and ablation experiments on the miniImageNet.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a small-sample image classification method and system based on self-supervised enhancement. Background technique [0002] In recent years, with the continuous development of deep learning algorithms, artificial intelligence has been widely used in many fields, such as medical care, transportation, and industrial manufacturing. Behind the prosperity and development of artificial intelligence is the continuous deepening of neural networks and the continuous increase in the amount of data required, which has led to the continuous increase in the cost of data collection and manual labeling. Not only that, the existing deep neural network models often have poor generalization. For example, a good cat and dog classifier is trained with a large number of cat and dog pictures, but if you want to use it for bird recognition, you need a lot of bird recognition. picture...

AI Technical Summary

Problems solved by technology

However, in classification tasks, the models with the highest accuracy are deep networks trained on large-scale data sets. Annotated datasets of scale are indispensable

Therefore, a bottleneck of the existing deep learning-based small-sample classification model is how to train a limited-scale network to extract features that are as rich in semantics as possible.

Images