Cooperative immune defense method orienting to attack of multiple fighting pictures

Abstract

The invention discloses a cooperative immune defense method facing attack of multiple fighting pictures. The method includes the following steps: 1) according to normal pictures and the fighting pictures generated by existing attack, fighting samples train fighting sample sub-classifiers of different structures, and optimization is performed by utilizing a genetic algorithm; 2) pictures used for detection are detected by utilizing the multiple fighting sample sub-classifiers, and whether the pictures are fighting samples and belong to types of the fighting samples is jointly judged; and 3) ifa novel attack method is found during a detection process, novel fighting sample classifiers are trained by utilizing the normal pictures and novel fighting samples; after the training is completed, the pictures that are judged to be normal are judged again; and the whole fighting sample classifiers are retrained after the certain time. According to the method, the novel fighting sample classifiers are specially generated for the novel fighting samples, the defense time on the novel fighting samples is shortened, and the defense effect on the novel fighting samples is improved.

Description

Technical field [0001] The invention belongs to the field of deep learning security technology, and specifically relates to a collaborative immune defense method for multiple types of resisting image attacks. Background technique [0002] Deep learning is a branch of machine learning. Since it was proposed in 2006, it has received extensive attention and research from academia. Deep learning mainly imitates the human brain and abstracts low-level features into higher-level attributes or features through multi-layer perceptrons to discover distributed feature representations of data, which has more powerful feature learning and feature expression capabilities. Among them, Convolutional Neural Network (CNN) is widely used in image classification, target detection, image semantic segmentation and other fields, and has achieved a series of breakthrough research. [0003] However, with the widespread application of deep learning models in image recognition, the anti-interference and at...

AI Technical Summary

Problems solved by technology

The security problem of deep learning in image recognition can also be counted as anomaly detection to a certain extent, but unlike ordinary detectors, deep learning detectors require longer training time

Images