Multi-task defense model construction method for infrared image countermeasure attacks

Abstract

The invention provides a multi-task defense model construction method for infrared image countermeasure attacks. The method comprises the following steps: determining a training data set and a test data set according to an obtained infrared image data set; and establishing a target network model and an additional network model, training the target network model and the additional network model, constructing a multi-task defense model according to the parameters of the optimal target network model and the optimal additional network model obtained by training, performing training, and obtainingthe optimal multi-task defense model after the training is completed. Parameter sharing is carried out between the target network and the additional network, so that the target network has the noise reduction and disturbance rejection capabilities of the additional network on the basis of not sacrificing the original classification precision; in this way, the additional network and the target classification network perform information interaction in a joint training mode instead of independently training the additional network, the network application generalization is high, and the robustnessof the target network is improved.

Description

technical field [0001] The invention relates to a method for constructing a multi-task defense model facing infrared image counterattacks, and belongs to the technical field of image processing. Background technique [0002] Infrared image is an image formed by measuring the infrared difference between the target itself and the background by using a detector to obtain different thermal infrared rays. It has good target detection ability and better robustness. However, infrared images have many natural shortcomings. The imaging area of ​​infrared weak and small targets is generally relatively small, and the infrared target imaging has no definite shape. There is no texture, shape information, etc. on the infrared image, and the imaging information is less; infrared imaging is affected by atmospheric attenuation. Influenced by factors such as complex background and clutter coverage, the infrared image data itself has problems such as large noise, low contrast, large non-unifor...

AI Technical Summary

Problems solved by technology

[0019] By comparing the exploration process of different attack methods and defense methods, due to the competitive relationship between the two, the lagging behind of one party may lead to the stagnation of the research of both parties, so this invention explores a new defense strategy for the image-oriented adversarial attack problem

And based on the above defense algorithms and strategies, most defense methods use adversarial samples as research objects or training data, but in actual scenarios, the defender may not be able to obtain the information of the attacker, so the new defense strategy proposed by this invention can be used in different Improve the robustness of the target network itself without destroying the structural framework of the target network using adversarial samples

Images