Remote sensing image classification network robustness improvement method based on self-supervised learning

Abstract

The invention provides a remote sensing image classification network robustness improvement method based on self-supervised learning, which not only utilizes labeled data, but also fully utilizes a large amount of non-labeled data existing in the remote sensing field, and effectively improves the robustness of a model by mining the information of images through a twin network. Feature extraction is conducted on the clean sample and the adversarial sample simultaneously by using a twin network to obtain feature vectors, and model training is completed by comparing and learning the feature vectors approaching the clean sample and the adversarial sample, so that the image has stable expression in a deep remote sensing image encoder network in an online network in the twin network, and therefore, the robustness is improved. According to the method, the robustness of the model to adversarial sample noise and natural noise is effectively enhanced, meanwhile, the classification effect of a clean data set is hardly influenced, and application is facilitated.

Description

technical field [0001] The invention relates to the cross field of deep learning and remote sensing, in particular to a method for improving the robustness of a remote sensing image classification network based on self-supervised learning. Background technique [0002] In recent years, neural networks have achieved breakthrough results in various fields such as computer vision and natural language processing. In the application of remote sensing image classification, neural networks inevitably act on various unknown remote sensing datasets containing a large number of different noises. Although these noises have no effect on human eye recognition, they can often induce deep neural networks to make wrong decisions. Judging, this poses a serious security threat to the application of neural networks in remote sensing image classification. [0003] Why the small noises that cannot be perceived by the human eye will cause the deep neural network to make completely wrong judgment...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide a method for improving the robustness of a remote sensing image classification network based on self-supervised learning, aiming at providing a method for improving the robustness of a remote sensing image classification model, and at the same time solving the problem that the existing technology does not make full use of remote sensing resources There is a large amount of unlabeled data in the database to improve the performance of the model. The specific technical solutions are as follows:

Images