基于深度学习的交通事件检测系统设计  

Design of traffic event detection system based on deep learning

作　　者：屠康 马倩芳 Tu Kang;Ma Qianfang(ZTE Feiliu Information Technology Co.,Ltd.,Nanjing 210000,China;Nuctech Jiangsu Co.,Ltd.,Nanjing 210000,China)

机构地区：[1]中兴飞流信息科技有限公司,江苏南京210000 [2]同方威视科技江苏有限公司,江苏南京210000

出　　处：《无线互联科技》2024年第3期63-65,共3页Wireless Internet Technology

摘　　要：建立完善的交通事件检测系统,已经成为我国智能交通系统的重要组成部分。文章从数据和算法2个方面深入分析了交通事件检测场景特征,设计了一种基于深度学习的交通事件检测系统。文章提出了一种混合架构的联合学习网络,通过综合利用ViT和Swin Transformer的优势解决了图像多标签分类问题的挑战;设计了一系列数据增强方法,应对数据不平衡性对深度学习模型的影响,并有效缓解了模型过拟合问题。实验结果表明,该系统在交通事件检测中具有更好的准确性和泛化能力,已应用到多个实际项目并取得了良好的应用效果。The establishment of comprehensive traffic incident detection systems have become an important component of China’s intelligent transportation system.This paper analyzes the characteristics of traffic event detection scenes in depth from both data and algorithms,and proposes a traffic event detection system based on deep learning.A hybrid architecture joint learning network is introduced,addressing the challenges of multi-label classification in image data by comprehensively leveraging the advantages of ViT and Swin Transformer.A series of data augmentation methods have been designed to cope with the impact of data imbalance on deep learning models,and effectively alleviating the problem of model overfitting.The experimental results demonstrate that the system has better accuracy and generalization ability in traffic event detection.The system has been applied to multiple practical projects,and has achieved favorable application outcomes.

关 键 词：交通事件检测 深度学习 对抗生成网络 TRANSFORMER 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]