基于深度学习的工业车辆驾驶行为识别  

Industrial Vehicle Driving Behavior Recognition Based on Deep Learning

作　　者：李俊杰[1] 邓海勤 高志勇 张勇[1] Li Junjie;Deng Haiqin;Gao Zhiyong;Zhang Yong(Beijing University of Posts and Telecommunications, Beijing 100876, China;AIdong Super AI Co., Ltd., Beijing 100013, China)

机构地区：[1]北京邮电大学,北京100876 [2]爱动超越人工智能科技(北京)有限责任公司,北京100013

出　　处：《信息通信技术》2019年第1期26-32,38,共8页Information and communications Technologies

基　　金：国家重大专项No.2014ZX03004002

摘　　要：文章提出一种基于深度学习的工业车辆驾驶行为识别的方法。该方法对工业车辆在实际工厂环境中行驶的特点进行分析,将三轴加速度传感器和三轴角速度传感器采集到的数据进行预处理,根据处理结果将数据送入深度神经网络训练,完成对工业车辆驾驶行为的识别。系统先对样本数据使用数据插值、标准化处理等方法进行预处理,通过数据增强算法减少过拟合的影响,再基于长短期记忆网络(LSTM)处理时间序列数据,构建出CNN+LSTM的深度网络模型,用于驾驶行为的识别。测试结果表明,所提模型识别整体准确率可达96.51%,能够准确地识别出工业车辆行驶的状态。This paper proposes a method based on deep learning to identify the driving behavior of industrial vehicles. The method analyzes the characteristics of industrial vehicles driving in the actual factory environment, preprocesses the data collected by the three-axis acceleration sensor and the three-axis angular velocity sensor, and sends the data to the deep neural network training according to the processing result to complete the identification of the driving behavior of industrial vehicle. The system first uses data interpolation and normalization processing preprocess the sample data, reduces the influence of overfitting through the data enhancement algorithm, and then constructs the depth network model of CNN+LSTM based on the longshort- term memory network (LSTM) suitable for processing time series data for identification of driving behavior. The test results show that the overall accuracy of the model identification mentioned in the article can reach 96.51%, which can accurately identify the state of industrial vehicles.

关 键 词：深度学习 驾驶行为识别 CNN+LSTM 工业车辆 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]