基于EfficientNetB0的司机分心检测研究  

Study of Driver Distraction Detection Based on the EfficientNetB0

作　　者：石鑫 吕广强 陈果 谢延景 董芳艳 陈科伟 SHI Xin;LV Guangqiang;CHEN Guo;XIE Yanjing;DONG Fangyan;CHEN Kewei(School of Mechanical Engineering and Mechanics,Ningbo University,Ningbo Zhejiang 315211,China;Institute of Intelligent Interactive Design and Manufacturing of Special Robots and High-end Equipment,Ningbo Institute of Technology,Zhejiang University,Ningbo Zhejiang 315100,China;Ningbo Guangqiang Robot Technology Co.,Ltd.,Ningbo Zhejiang 315000,China)

机构地区：[1]宁波大学机械工程与力学学院,浙江宁波315211 [2]浙大宁波理工学院特种机器人与高端装备智能交互设计制造研究院,浙江宁波315100 [3]宁波广强机器人科技有限公司,浙江宁波315000

出　　处：《佳木斯大学学报(自然科学版)》2024年第12期41-43,共3页Journal of Jiamusi University:Natural Science Edition

基　　金：宁波市国际合作项目(2023H007)和宁波市“揭榜挂帅”暨“科技创新2025”重大专项(2023Z180)。

摘　　要：提出了一个名为CBAM-EfficientNetB0的框架,旨在解决低参数条件下司机分心行为识别准确率低的问题。它集成了CBAM注意力机制,包括通道注意力模块和空间注意力模块。这使得网络更加关注重要的特征信息,从而提高了特征的区分度和表达能力。通过将模型的优化器转换为SGD,并获得优化的学习率和动量参数,提高了模型的识别准确率和收敛性。CBAM-EfficientNetB0在State Farm Distracted Driver Detection数据集上达到了96.8%的准确率。结果显示,与同类的框架相比,它在低参数条件下表现良好。The article proposes a framework named CBAM-EfficientNetB0,aiming to address the issue of low accuracy in distracted driver behavior recognition under low-parameter conditions.It integrates CBAM attention mechanisms,including channel attention modules and spatial attention modules.This enables the network to focus more on important feature information,thereby improving feature discriminability and expressiveness.By converting the model's optimizer to SGD and obtaining optimized learning rate and momentum parameters,the model's recognition accuracy and convergence are improved.CBAM-EfficientNetB0 achieves an accuracy of 96.8%on the State Farm distracted driver detection dataset.The results demonstrate its strong performance under low-parameter conditions compared to similar frameworks.

关 键 词：司机分心检测 EfficientNetB0 CBAM SGD 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]