University of Twente Student Theses


Applying Heat Maps on a Traffic Sign Detection Case Study

Petrov, Aleksandar (2023) Applying Heat Maps on a Traffic Sign Detection Case Study.

[img] PDF
Abstract:The advances in deep neural networks (DNN) have enabled the de- velopment of some of the most sophisticated systems currently used in various industries. DNN systems are used in applications such as autonomous driving, where traditional software engineering is insufficient. DNNs, however, lack the explainability inherent in con- ventional software methods. A problem linked with such networks is the possibility of attackers introducing backdoors (attacks) that hinder the decision process of the models. This research explores the effect of visualization algorithms, such as Grad-CAM, on backdoor mitigation methods, specifically for models that classify traffic signs. The contribution of this paper is to show the explainability capabili- ties of heat maps in the context of trojaned traffic sign DNN models. Visualizing the network’s activations aims to solidify the work of the backpropagation mitigation research. To achieve that, we introduce a novel method of exploring individual feature maps’ activations, offering even more crucial detail in the network workings. This paper should aid the development of more robust DNNs for autonomous driving systems.
Item Type:Essay (Bachelor)
Audi AG, Germany
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page