Research And Implementation Of Neural Network Attack And Defense Technology For License Plate Recognition

The license plate recognition system is widely used in daily life due to its convenience and recognition accuracy.License plate recognition systems generally use deep neural networks,its speed and accuracy have been greatly improved with the rapid development of deep learning.However,in recent years,researchers have found that although the recognition effect of deep neural networks seems like very good,they are still fragile.Deep neural networks can easily be attacked by carefully constructed adversarial examples,leading to errors in the license plate recognition system.This paper studies the security of the license plate recognition system based on the deep neural network.Through the research and implementation of offensive and defensive technology,it is hoped to improve the robustness of the license plate recognition system.In this paper,we use deep neural network countermeasures that have been popular in recent years to explore the license plate recognition system.Use Hyper LPR open source license plate recognition library to build a license plate recognition system.Improve the license plate detection method of the Haar cascade classifier in the Hyper LPR system,add the YOLO target detection algorithm-one of the deep learning research results in recent years,and build and train the neural network model.The trained YOLO recognition method has faster detection and higher accuracy than the Hyper LPR recognition method.On this basis,the license plate attack research on the built Hyper LPR license plate recognition system,combined with FGSM,BIM,PGD and other attack algorithms,is used to attack the license plate system and generate countermeasure data.At the same time,related algorithms(such as adding momentum,performing random transformation operations,etc.)are added to the attack to improve the migration of the attack effect.The verification shows that the generated adversarial examples are not only effective for the Hyper LPR license plate recognition system,but also effective for the Face++,Baidu and other license plate recognition systems,with a certain degree of migration.Finally,in order to display the above research results more conveniently,this paper builds an offensive and defensive system based on license plate recognition,and displays the Hyper LPR license plate recognition system,the license plate recognition system based on the YOLO model,and the offensive and defensive algorithms for the license plate recognition system through the interactive mode of the website.This paper improves the robustness of the license plate recognition system through the offensive and defensive confrontation research of the license plate recognition system,which can slow down the destructiveness of the confrontation attack,which has positive significance for the research of offensive and defensive confrontation.