NTS-YOLO

NTS-YOLO：a nocturnal traffic sign detection method based on improved YOLOv5In this paper, a nighttime traffic sign recognition method "NTS-YOLO" is proposed, which consists of three main parts. Firstly, this paper adopts the unsupervised nighttime image enhancement technique proposed by Ye-Young Kim et al. Secondly, the Convolutional Block Attention Module (CBAM) attentional mechanism is introduced on the basis of the YOLOv5 network structure, and lastly, the Optimal Transmission Allocation (OTA) loss function is used to optimize the model's performance in the target detection task. With this approach, the accuracy of predicting the bounding box can be effectively optimized so that the model can predict the location of the target and the bounding box more accurately, thus improving the robustness and stability of the model in the target detection task.Other datasIn this paper, 599 nighttime images from the CCTSDB2021 dataset are referenced, of which 80% of the images (479 images) are used as the training set and 20% of the images (120 images) are used as the validation set. In view of the relatively small number of road sign types at night, 9170 daytime road scene images from the TT100K dataset are also referenced to increase the diversity of the data, which are divided into a training set (7208 images) and a validation set (1962 images) at a ratio of 8:2.Links to other publicly accessible locations of the data:CCTSDB2021：GitHub - csust7zhangjm/CCTSDB2021TT100K：http://cg.cs.tsinghua.edu.cn/traffic-sign/data_model_code/data.zipEnvironmentThe experimental environment consists of a high-performance computer configured with an Intel Core i7 processor, 32GB RAM, and an NVIDIA GeForce RTX 4060 graphics card. PyTorch 2.0.1 was chosen as the main deep learning framework, and CUDA technology was utilized to accelerate model training and inference to ensure computational efficiency and data processing power during the experiment.NTS-YOLO：一种基于改进YOLOv5的夜间交通标志检测方法 本文提出一种夜间交通标志识别方法“NTS-YOLO”，其主要包含三大核心模块。其一，采用Ye-Young Kim等人提出的无监督夜间图像增强技术；其二，在YOLOv5网络结构中引入卷积块注意力模块（Convolutional Block Attention Module，CBAM）注意力机制；其三，使用最优传输分配（Optimal Transmission Allocation，OTA）损失函数优化模型在目标检测任务中的性能。通过该方案，可有效优化边界框预测精度，使模型能够更精准地预测目标位置与边界框，进而提升模型在目标检测任务中的鲁棒性与稳定性。 其他数据集 本文引用CCTSDB2021数据集的599张夜间图像，其中80%（共计479张）作为训练集，剩余20%（共计120张）作为验证集。鉴于夜间道路交通标志类别相对匮乏，本文还引入TT100K数据集中的9170张日间道路场景图像以扩充数据多样性，该数据集按照8:2的比例划分为训练集（7208张）与验证集（1962张）。 该数据集的其他公开获取地址如下： CCTSDB2021：GitHub - csust7zhangjm/CCTSDB2021 TT100K：http://cg.cs.tsinghua.edu.cn/traffic-sign/data_model_code/data.zip 实验环境 本次实验搭载高性能计算机，配置为Intel Core i7处理器、32GB内存以及NVIDIA GeForce RTX 4060显卡。选用PyTorch 2.0.1作为核心深度学习框架，并借助CUDA技术加速模型训练与推理流程，以保障实验期间的计算效率与数据处理能力。