Guideboard identification confrontation defense method based on genetic algorithm

Abstract

The invention discloses a guideboard identification confrontation defense method based on a genetic algorithm. The method comprises the following steps: (1) constructing a training set, a test set anda confrontation set; (2) constructing a guideboard classifier, and training the guideboard classifier by using the training set and the test set to obtain a guideboard recognition model; (3) constructing a guideboard attack model according to a genetic algorithm, and generating an adversarial sample by utilizing the guideboard attack model; (4) correcting the adversarial sample, applying the adversarial sample in a physical scene, and collecting a physical image of the adversarial sample; (5) inputting the physical image of the confrontation sample into a guideboard identification model, andscreening to obtain a high-quality confrontation sample; (6) adding high-quality confrontation samples into the training set, and training the guideboard recognition model again by using the trainingset to realize confrontation defense optimization of the guideboard recognition model; and (7) identifying the guideboard image by using the guideboard identification model after confrontation defenseoptimization to realize confrontation defense of guideboard identification.

Description

technical field [0001] The invention belongs to the fields of machine learning, computer vision, and intelligent traffic safety, and in particular relates to a road sign recognition confrontation defense method based on a genetic algorithm. Background technique [0002] With the rapid development of machine learning, deep neural networks (DNNs) have become one of the most prominent technologies of our time, due to their extremely high accuracy in completing artificial intelligence tasks that require highly abstract features, the application of deep neural networks It is becoming more and more widespread, and it has played a very good role. With the efforts of researchers, the recognition accuracy of street sign recognition technology based on deep neural network has been continuously improved, and it has even surpassed human beings. had a profound impact. [0003] Although road sign recognition has achieved good performance today, however, some recent studies have shown th...

AI Technical Summary

Problems solved by technology

Reflected in the real scene of road sign recognition, imagine that when an autonomous car driving in a busy urban area is affected by adversarial examples, it misrecognizes road signs and suddenly speeds up, which will cause serious consequences

Road sign recognition also has hidden dangers in other security-oriented scenarios

[0004] At present, image attacks on street sign recognition are mainly based on the white-box model, but there are the following challenges in this attack: (1) The attack on the white-box model needs to obtain the internal parameters of the opponent's street sign recognition model

(2) A single disturbance may be so small that it is difficult to be captured by the camera

(3) The changing environment of road signs may cause disturbance failure

(4) Disturbance may be distorted during printing

Therefore, to solve the above problems, generating robust street sign adversarial samples based on the black-box model will cause more harm

Images