摘要
由于香烟目标较小,图像分辨率较低,目前传统的目标检测算法不足以支撑对香烟的检测,为警戒驾驶员在驾驶时吸烟出现的安全问题,提出一种驾驶员吸烟检测算法。结合SSD目标检测模型,在其骨干网络中引入CSP架构;利用反卷积操作,将有效特征层进行融合;在GAM注意力机制上改进其通道子模块,结合最大池化和平均池化,抑制与香烟不相关的特征。模型在自制驾驶员吸烟数据集上的识别mAP达94.93%,具有较好的鲁棒性以及泛化能力。
Due to the small size of cigarette targets and low image resolution,traditional object detection algorithms are currently insufficient to support the detection of cigarettes.To alert drivers to safety issues related to smoking while driving,a driver smoking detection algorithm was proposed.Combining the SSD object detection model,a CSP architecture was introduced into its backbone.Deconvolution operations were used to fuse effective feature layers.The GAM attention mechanism was improved by combining maximum pooling and average pooling to suppress features unrelated to cigarettes.The recognition mAP of the model on the self-made driver smoking dataset reaches 94.93%,and it has good robustness and generalization ability.
作者
王晓强
李科岑
李雷孝
王鑫鹏
杨锦帆
WANG Xiao-qiang;LI Ke-cen;LI Lei-xiao;WANG Xin-peng;YANG Jin-fan(College of Information Engineering,Inner Mongolia University of Technology,Hohhot 010080,China;College of Data Science and Application,Inner Mongolia University of Technology,Hohhot 010080,China)
出处
《计算机工程与设计》
北大核心
2024年第11期3337-3344,共8页
Computer Engineering and Design
基金
内蒙古自然科学基金项目(2023MS06021)
2023年自治区直属高校基本科研业务费基金项目(JY20230065)
内蒙古自治区科技计划基金项目(2020GG0104)
内蒙古自治区科技成果转化专项资金基金项目(2020CG0073、2021CG0033)
内蒙古自治区重点研发与成果转化计划基金项目(2022YFSJ0013)
内蒙古自治区高等学校青年科技英才支持计划基金项目(NJYT22084)
内蒙古自治区研究生科研创新基金项目(S20210195Z)。
关键词
驾驶员吸烟检测
跨阶段局部网络
特征融合
注意力机制
多尺度检测
目标检测
深度学习
driver smoking detection
cross stage partial network
feature fusion
attention mechanism
multi-scale detection
object detection
deep learning
