A Traffic Sign Detection Network based on Dynamic Focusing Loss and Lightweight Convolution

The booming development of unmanned technology has placed greater demands on traffic sign detection. In this context, speed and accuracy are simultaneously pursued, yet existing detection techniques often struggle with challenges such as small targets and complex backgrounds, particularly in real-time scenarios. To solve this problem, this paper proposes a new traffic sign detection model based on YOLOv8s with the following improvements. A Faster Neural Next Work (FasterNeXt) layer is used for small target feature extraction, which enhances the feature extraction capability of the network for small targets while guaranteeing lightweight. A lightweight global sparse convolution (GSConv) is used, which reduces the weight of the model and improves the generalization performance of the model. The optimization of low-quality labels in traffic sign detection using Wise-IOU version 3 (WIoUv3) significantly enhances detection accuracy. Tests on the CCTSDB2021 dataset and the TT100K show that our method performs better with fewer parameters than existing algorithms. The accuracy of our model in traffic sign detection improves by 1% to reach 98%, the Frames Per Second (FPS) increases by 3.37, and there is a reduction in the number of parameters. Our model can satisfy the requirements of assisting vehicle driving.