Towards a robust object classifier for autonomous vehicles by feature synthesis

Ge Jin

doi:10.1117/12.2619787

22 December 2021 Towards a robust object classifier for autonomous vehicles by feature synthesis

Ge Jin

Proceedings Volume 12058, Fifth International Conference on Traffic Engineering and Transportation System (ICTETS 2021); 120583Z (2021) https://doi.org/10.1117/12.2619787
Event: 5th International Conference on Traffic Engineering and Transportation System (ICTETS 2021), 2021, Chongqing, China

Abstract

As the AVs’ perception system, there are many different types of sensors in the car to provide data support for the automatic driving system's judgment. The judgment system, which is made up of convolutional neural networks, is then lacking in robustness. When confronted with input data such as adversarial samples and malicious tampering, it will provide some options that significantly deviate from the correct answer. In this article, we first use FGSM as an adversarial sample generation method, and then used the generated adversarial samples to successfully disrupt the system's results. After that, the adversarial sample data was then added to the original data set and trained in the neural network. We successfully trained a classifier with high robustness after incorporating hyperparameters and feature fusion.

Citation Download Citation

Ge Jin "Towards a robust object classifier for autonomous vehicles by feature synthesis", Proc. SPIE 12058, Fifth International Conference on Traffic Engineering and Transportation System (ICTETS 2021), 120583Z (22 December 2021); https://doi.org/10.1117/12.2619787

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available