采用深度卷积神经网络方法的毫米波图像目标检测

Millimeter Wave Image Object Detection Using Deep Convolutional Neural Network Method

通过收集大量的毫米波图像并建立相应的人体数据集进行检测,提出基于Faster R-CNN深度学习的方法检测隐藏于人体上的危险物品。该方法将区域建议网络和VGG19训练卷积神经网络模型相结合,构建了面向毫米波图像目标检测的深度卷积神经网络。为了提高毫米波图像的处理能力,采用Caffe深度学习框架在图形处理单元上进行训练和测试。实验结果证明了基于Faster R-CNN深度卷积神经网络的目标检测方法能有效检测毫米波图像中的危险物品,并且目标检测的平均准确率约94%,检测速度约为6 frame/s,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。

Through collecting a large number of millimeter wave(MMW)images and establishing corresponding human data sets for detection,a method based on Faster R-CNN deep learning is proposed to detect concealed-object in a human body.The method combines the Region Proposal Network and the VGG19 training convolutional neural network model to construct a deep convolutional neural network for MMW image object detection.In order to improve the processing power of MMW images,the Caffe deep learning framework is used to train and test on the graphics processing unit.The experimental results show that the object detection method based on Faster R-CNN deep convolutional neural network can effectively detect dangerous objects in MMW images,the mean average precision of object detection is about 94%,and the detection speed is about 6 frame/s.It has extremely important reference value for the intelligent development of MMW security inspection system.