The Impact of Data Augmentation Strategy Based on an Improved CNN on Traffic Sign Recognition in Complex Environments
TL;DR
This study takes the GTSRB dataset as the research object, designs a multi-dimensional scene-based data augmentation strategy, constructs an end-to-end recognition process combined with a lightweight improved CNN, and verifies the effect of the strategy through controlled experiments.
Abstract
The robustness of traffic sign recognition is the core guarantee for the safety of autonomous driving. Most existing studies focus on the innovation of Convolutional Neural Network (CNN) model architecture, yet neglect the optimization at the data level and lack targeted data augmentation schemes adapted to complex scenarios. Aiming to improve the generalization ability of CNN for traffic sign recognition in complex scenarios, this study takes the GTSRB dataset as the research object, designs a multi-dimensional scene-based data augmentation strategy, constructs an end-to-end recognition process combined with a lightweight improved CNN, and verifies the effect of the strategy through controlled experiments. The results show that this strategy can significantly improve the model's recognition performance in complex scenarios, with a recognition accuracy of 98.86% and a recall rate of 98.35% on the GTSRB test set, without sacrificing operational efficiency. It has a prominent optimization effect on low-frequency categories, with an improvement range of 14.93%. This study provides an efficient data-driven solution for the practical application of traffic sign recognition technology and makes up for the lack of data-level optimization in existing research.