Defense-GAN-based traffic sign identification method

Abstract

The invention discloses a Defense-GAN-based traffic sign recognition method, and the method specifically comprises the steps: collecting a traffic sign image, carrying out the preprocessing of the image, and dividing the image into a training set, a test set, and a confrontation set; training the deep neural network by using the training set and the test set to obtain a traffic sign recognition model; acquiring a traffic sign confrontation sample picture similar to the original noiseless traffic sign image in distribution by using a confrontation set training model; inputting the confrontation sample picture into a traffic sign classification model for evaluation, and obtaining a high-confidence confrontation sample picture; adding the obtained high-confidence confrontation sample pictures into a training set to re-train the traffic sign recognition model; and identifying the traffic sign picture by using the traffic sign identification model after confrontation and defense optimization. The method can learn migration among any classification models, is low in training cost, is wide in application range, is high in processing speed, and is suitable for advanced auxiliary driving, unmanned driving and other applications.

Description

technical field [0001] The invention belongs to the technical field of traffic sign recognition, in particular to a traffic sign recognition method based on Defense-GAN neural network. Background technique [0002] In recent years, with the rapid development of machine learning and deep learning, computer vision technology and automatic driving have gradually become indispensable key technologies in people's lives. Traffic Sign Identification (Traffic Sign Identification), also known as traffic sign target detection, is a technology that uses convolutional neural network (CNN) training picture sets in deep learning technology to judge the traffic sign category in images or video sequences. There are many features that can be Improving the accuracy of convolutional neural networks requires practical testing of combinations of these features on large datasets and theoretical proof of the results. Most CNN-based object detectors are only suitable for recommender systems. Impr...

AI Technical Summary

Problems solved by technology

[0009] Aiming at the defects of existing deep learning image classification, the present invention provides a traffic sign recognition method based on Defense-GAN neural network

Images