基于深度学习的驾驶行为识别模型线上训练系统设计与实现

Online Training System for Driving Behavior Recognition Model Based on Deep Learning

为降低深度学习技术门槛,提高图像识别模型构建效率,设计并实现了一套驾驶行为识别模型线上训练系统,系统自底向上划分为数据存储层、功能模块层、用户接口层,设计了三类通用型深度学习模型,引入“模型状态”概念,实现状态迁移机制,提供了驾驶行为识别模型创建、训练、部署及测试“一站式”全流程功能.通过测试,系统所构建的模型具有收敛速度快,识别准确度高的优点,应用价值较高.

In order to lower the threshold of deep learning technology and improve the efficiency of image recognition model construction,this paper designs and implements an online training system for driving behavior recognition models. The system is divided into data storage layer,functional module layer,and user interface layer from the bottom up. The system designs three types of general deep learning models,introduces the concept of“model state”,realizes the state transition mechanism,and provides a“onestop”full-process function of driving behavior recognition model creation,training,deployment and testing. Through testing,the model built by the system has the advantages of fast convergence speed and high recognition accuracy,and has high application value.