Image classification method and device based on relational network, equipment and storage medium

Abstract

The invention relates to an image classification method based on a relation network. The method comprises the following steps: obtaining a target image; then inputting the target image and the support set image into a trained image classification model to obtain the similarity between the target image and each category of image in the support set image, wherein the image classification model comprises an embedding module and a measurement module, the embedding module is a random deep network, and the measurement module comprises a convolution layer and a full connection layer which are connected with each other; and obtaining the category of the target image according to the maximum similarity. According to the method, a random deep network is adopted in an embedded module to replace a convolutional layer in a relational network, and the network can optimize the training process of a residual network by randomly removing some redundant layers, so that the network can deepen the number of layers and can prevent the problem of overfitting at the same time; therefore, more accurate support set image features and query set image features can be extracted, so that the category judgment of the query set is further improved.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to an image classification method, device, equipment and storage medium based on a relational network. Background technique [0002] In recent years, the unprecedented breakthroughs of deep learning in various fields have largely relied on the large amount of available labeled data, which requires a lot of cost to collect and annotate, which severely limits the ability to learn new categories. More importantly, these deep learning models are difficult to solve the problem of a small amount of labeled data. Therefore, the problem of small-sample learning based on relational networks has become a hot research topic in recent years. [0003] The purpose of small sample research is to design a relevant learning model so that the model can achieve fast learning and identify the category of new samples in only a small number of labeled samples. Existing research applica...

AI Technical Summary

Problems solved by technology

[0004] Using deep convolutional networks to extract image features is a critical step in the process of small-sample learning. However, using existing learning methods for small-sample learning tasks makes it difficult for deep convolutional networks to improve the classification accuracy of the model.

[0005] When training a deep neural network, there are problems such as gradient disappearance, information gradually decreasing when it flows forward, and training time is too long, which makes the training of a deep neural network very difficult.

In some tasks, if shallow neural networks are used, their structure is simple and easy to train but their expressiveness is poor; if deep neural networks are used, they have more redundant layers, although their expressiveness is better, they are more difficult to train

Images