通信与信息技术

交通标志检测是能够确保智能驾驶安全的关键技术。针对交通标志在检测中像素面积小、精度低以及被遮挡的问题，提出一种基于YOLOv8改进的交通标志检测算法。首先，引入RCSOSA模块替换骨干网络中的C2f部分，利用重参数化技术分别在训练阶段和推理阶段提高模型的表达能力和推理效率、简化计算；然后，为捕捉到更细节的特征信息，引入上下文增强模块，突出微小目标在多尺度特征中的语义信息；最后，采用Wise-IoU取代原始的CIoU，对预测框的离群程度进行评估，以此为依据动态调整梯度增益，使模型具有更好的定位精度。实验结果表明，改进后的模型在中国交通标志检测数据集CCTSDB中较原模型在精确度P及平均精度均值mAP上分别提升了2.6%和1.1%，小目标检测精度提高了1.2%，检测速度为每秒68帧，满足实时检测的要求。

Traffic sign detection is a critical technology for ensuring the safety of intelligent driving. To address challenges such as small pixel areas, low accuracy, and occlusion in traffic sign detection, an improved traffic sign detection algorithm based on YOLOv8 is proposed. First, the RCSOSA module is introduced to replace the C2f component in the backbone network, leveraging re-parameterization techniques to enhance the model's expressiveness and inference efficiency during both the training and inference stages, while simplifying computations. Next, to capture more detailed feature information, a context enhancement module is added to emphasize the semantic information of small targets across multi-scale features. Finally, Wise-IoU is adopted to replace the original CIoU to evaluate the outlier degree of the predicted boxes, dynamically adjusting the gradient gain based on this evaluation to improve the model's localization accuracy. Experimental results show that the precision and mAP of the improved model are increased by 2.6% and 1.1% on the CCTSDB data set compared with the original model. The accuracy for small objects is improved by 1.2%, and the inference speed of 68 FPS is maintained, which meets the real-time detection requirements.