利用深度学习进行毫米波图像违禁物体识别  

Using Deep Learning to Identify Prohibited Object in Millimeter-wave Image

作　　者：张健[1] 王卫民[1] 唐洋 ZHANG Jian;WANG Weimin;TANG Yang(School of Computer Science,Jiangsu University of Science and Technology,Zhenjiang 212003)

机构地区：[1]江苏科技大学计算机学院,镇江212003

出　　处：《计算机与数字工程》2020年第7期1760-1765,共6页Computer & Digital Engineering

摘　　要：毫米波人体成像技术是目前全球安防领域的先进技术,已在美国、英国等国机场用于旅客人身安检。但毫米波图像中违禁物体的高效识别仍然是一个亟待解决的难题,这极大地限制了毫米波成像技术在机场旅客筛选中的应用。文章将深度学习技术应用于毫米波图像,自动判别毫米波图像中是否藏有违禁物品,以及违禁物品藏在身体哪些部位。利用卷积神经网络(Convolutional Neural Network,CNN)对人体的毫米波图像提取特征,然后通过长短期记忆网络(Long Short-term Memory,LSTM)融合多视图图像特征,最后通过多路sigmoid分类器得到人体17个部位分别藏有违禁物品的概率值。实验中,在与训练集和验证集没有交集的测试集上的损失函数值为0.03,经设定阈值后,人体17个部位识别正确率为99.76%,说明了方法的有效性。The millimeter-wave human body imaging technology is currently the advanced technology in the global security field,and has been used for passenger security inspection in airports in the United States,the United Kingdom and other countries.However,the efficient identification of prohibited objects in millimeter-wave image is still a difficult problem to be solved,which greatly limits the application of millimeter-wave imaging technology in airport passenger screening.The article applies deep learning techniques to millimeter-wave image,automatically discriminating whether millimeter-wave image contain prohibited items,and where the prohibited items are hidden in the body.The Convolutional Neural Network(CNN)is used to extract the features of the millimeter-wave image of the human body,and then the multi-view image features are merged by the Long Short-term Memory network(LSTM).Finally,the probability of the existence of forbidden object which hidden in 17 parts of the human body in the image can be obtained via multi-channel sigmoid classifier.In the experiment,the loss function value on the test set that does not intersect with the training and validating set is 0.03.After the threshold is set,the recognition rate of the 17 parts of the human body is99.76%,which verifies the effectiveness of the method.

关 键 词：人身安检 毫米波图像 卷积神经网络 长短期记忆网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]