Traffic Sign Recognition System Based On EdLeNet_LM

With the progress of society and the rapid development of science and technology,people’s living standards are constantly improving,and most people in China have begun to gradually accept car consumption.In addition,the improvement of living standards makes people pay more attention to the quality of life,and cars have become a way of travel for many people.Cars make life more convenient and faster,improve the quality of life and work efficiency,but more and more cars cause traffic jams and accidents to become more and more serious and more frequent,making traffic safety face a severe test and Smooth roads reduce efficiency.For drivers driving on unfamiliar roads,traffic signs that play the role of driving play a key role.At the same time,abnormal detection of traffic sign recognition models is also required,which is related to driver safety.In this paper,on the basis of summarizing a lot of relevant research at home and abroad,deep learning is used for the research of traffic sign recognition,backdoor attack and anomaly detection.The main tasks are as follows.(1)Based on EdLeNet network,it is improved to EdLeNet＿LM network model,and the proposed Leaky Mish activation function is used.Image pre-processing techniques such as image graying,normalization and data enhancement are performed on the GTSRB data set,and the pre-processing models are compared.The traffic sign recognition accuracy rate of the EdLeNet＿LM network model is 98.54%,which is 0.68% higher than the EdLeNet network.(2)For the research on backdoor attacks and anomaly detection,the backdoor is injected when training the EdLeNet＿LM model,and anomaly detection is performed through reverse engineering triggers,and the anomaly index is used to prove whether the model is infected.(3)Design and implementation of traffic sign recognition system.Through requirements analysis,the system is designed,and finally the system is realized.This system takes the deep neural network backdoor attack as a study to carry out anomaly detection on the traffic sign recognition model.The system is divided into three parts,traffic signs,identification and backdoor attacks.Under the Flask framework,the trained model is called to realize the traffic sign recognition system.