A road sign recognition attack defense method based on multi-objective road optimization

Abstract

The invention discloses a street sign recognition confrontation defense method based on multi-objective road optimization, comprising: (1) constructing a training set, a test set and a confrontation set; (2) constructing a street sign classifier, and using the training set and the test set to compare The classifier is trained to obtain the street sign recognition model; (3) Construct the street sign attack model according to the particle swarm optimization algorithm, and use the street sign attack model to generate confrontation samples; (4) Correct the confrontation samples, apply the confrontation samples to the physical scene, and then collect the confrontation samples. The physical image of the sample; (5) Input the physical image of the adversarial sample into the street sign recognition model, and screen high-quality adversarial samples; (6) Add the high-quality adversarial samples to the training set, and use the training set to retrain the street sign recognition model , to realize the optimization of the adversarial defense of the street sign recognition model; (7) use the street sign recognition model optimized by the adversarial defense to recognize the street sign image, so as to realize the adversarial defense of the street sign recognition.

Description

technical field [0001] The invention belongs to the fields of computer vision, intelligent transportation, and deep learning, and in particular relates to a street sign recognition attack defense method based on multi-objective road optimization. Background technique [0002] With the rapid development of machine learning in recent years, computer vision technology and autonomous driving have gradually become one of the key technologies in people's lives. Road sign recognition is based on computer vision technology, which extracts information from traffic sign images in the objective world and processes the information so that machines can understand and recognize traffic signs in the image. With the unremitting efforts of researchers, road sign recognition technology has gradually matured, and the recognition accuracy has been continuously improved. Sometimes it can even compete with humans, and it has been successfully applied to intelligent transportation, intelligent dri...

AI Technical Summary

Problems solved by technology

However, in real life, this kind of attack means that the internal parameters of the road sign recognition model need to be obtained in advance, which is quite difficult

Therefore, the street sign attack based on the black box model can realize the attack on the street sign recognition by simulating the physical scene without knowing the internal parameters of the street sign recognition model, which will cause greater harm

Images