摘要
目前,各项新技术不断被应用于各个行业,对于人工智能技术在交通行业的应用方面,目前有许多车辆拥有的数据资源并未被充分利用,移动的车辆可以为人工智能网络模型训练提供丰富的数据资源。因此,以司机驾驶识别场景为例,提出一种基于联邦学习的司机驾驶状态识别的方法,利用联邦学习“数据不动,模型动”的特点,在保护车辆数据隐私的情况下,充分利用现有车载数据资源训练出性能好的模型应用于实际交通场景。
At present,various new technologies are constantly applied in various industries.For the application of artificial intelli⁃gence technology in the transportation industry,the data resources owned by many vehicles have not been fully utilized.Mobile ve⁃hicles can provide rich data resources for artificial intelligence network model training.Therefore,taking the driver driving recognition scene as an example,a driver driving state recognition method based on federal learning is proposed.Taking advantage of the character⁃istics of"data does not move,model does not move"of federal learning,and under the condition of protecting the privacy of vehicle data,make full use of the existing on-board data resources to train a model with good performance and apply it to the actual traffic scene.
作者
胡宇翔
高琛
谭北海
Hu Yuxaing;Gao Chen;Tan Beihai(School of Automation,Guangdong University of Technology,Guangzhou 510006;Guangzhou weichi Technology Co.,Ltd,Guangzhou 511455)
出处
《现代计算机》
2021年第35期41-46,共6页
Modern Computer
基金
桂林市科学研究与技术开发项目(20190214-3)。
关键词
深度学习
联邦学习
驾驶状态识别
deep learning
federal learning
driving status recognition
