Soft threshold defense method for remote sensing image confrontation sample

Abstract

The invention discloses a soft threshold defense method for remote sensing image confrontation samples, which comprises the following steps: storing correctly classified remote sensing images and confrontation samples in the same type of verification set, and deleting remote sensing images which cannot be correctly classified in the verification set; taking the original image as a positive sample, taking an adversarial sample as a negative sample, and reclassifying the stored images in the verification set; obtaining a new data set in combination with the classified output confidence; training a logistic regression model on the new data set; obtaining a threshold value of output confidence through decision boundaries of the original image and the adversarial sample; and comparing the output confidence of the current input image with the defense soft threshold, and judging whether the current input image is a confrontation sample or not. According to the method, the attack of the adversarial sample in the remote sensing image scene classification problem can be effectively defended, and the torpoxic rate of the convolutional neural network is reduced to 0.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image classification, and in particular relates to a soft threshold defense method of remote sensing image confrontation samples. [0002] technical background [0003] Convolutional neural network (CNN) has excellent feature extraction ability and high accuracy, and has become a general technique for object recognition in the field of remote sensing. It is widely used in remote sensing fields such as disaster management, forest monitoring and urban planning. Well-performing CNNs can have high financial benefits. However, many studies have shown that CNNs are very vulnerable to adversarial samples, which are carefully generated and imperceptible, which can make the model predict wrong results with high confidence. Adversarial examples have become the most concerned security issue for CNNs in practical applications. Adversarial examples demonstrate the vulnerability of convolutional neural...

AI Technical Summary

Problems solved by technology

However, these algorithms require model retraining and are computationally intensive

The second is to detect only

Images