摘要
在公共安全检查领域中,研究毫米波图像目标检测的快速性和精准性的方法具有非常重要的实际应用价值。提出了基于Faster R-CNN深度学习的方法检测隐藏在人体上的危险物品。该方法将区域建议网络(region proposal network,RPN)和VGG16训练卷积神经网络模型相结合,接着通过在线难例挖掘(online hard example mining,OHEM)技术优化训练所提出的网络模型,从而构建了面向毫米波图像目标检测的深度卷积神经网络。实验结果证明所提的方法能高效地检测毫米波图像中的危险物品,并且目标检测的平均精度高达约94.66%,检测速度约为6帧/s,同时对毫米波安检系统的智能化发展有着极其重要的参考价值。
In the field of public security inspection,the method of studying the rapidity and accuracy of millimeter-wave image object detection is important with a practical value.A method based on faster R-CNN deep learning to detect dangerous objects hidden in the human body was proposed.In the method,first,the region proposal network and the VGG16 training convolutional neural network model was combined.Second,in the online hard example mining technology,the network model proposed by the training was optimized.Finally,a deep convolutional neural network for image object detection in millimeter wave was constructed.Results show that the proposed method could detect efficiently the dangerous objects in millimeter wave images,the average accuracy of object detection reached 94.66%,and the detection speed was 6fps,which provided an important reference for the intelligent development of security inspection system in millimeter wave.
作者
程秋菊
陈国平
王璐
管春
CHENG Qiu-ju;CHEN Guo-ping;WANG Lu;GUAN Chun(College of Optoelectronic Engineering,Chongqing University of Post and Telecommunications,Chongqing 400065,China)
出处
《科学技术与工程》
北大核心
2020年第13期5224-5229,共6页
Science Technology and Engineering
基金
重庆市科委面上项目(D021D2019027)。
