Research On Segmentation And Recognition Algorithm Of Targets For Passive Millimeter-wave And Terahertz Imaging

With the rapid development of China’s economy,military and other aspects,the status of great powers is rising and the international influence is rapidly improving.The threat of terrorism and unexpected public hazards is becoming more and more serious and the task of maintaining stability is more arduous.The development and study of a new practical security inspection technology is of great significance to the promotion of national security.Studies at home and abroad show that millimeter wave and terahertz passive imaging detection technology has broad and important application prospects in the field of security inspection,because of its high spatial resolution and good penetration.However,due to the limitation of production cost and technological process,the of signal-to-noise ratio millimeter wave and terahertz image acquired by the general practical system are relatively low.If the target is judged by the naked eye,it can not meet the basic requirements for the stability and speed of the security inspection system.Using millimeter wave and terahertz passive imaging "target segmentation and recognition" algorithm to mark targets,it can significantly improve the visibility of images,and further improve the reliability and efficiency of security inspection.Therefore,it is of great significance to study the algorithm of millimeter wave and terahertz passive imaging target segmentation and recognition for the practical application of the system.The study are based on the actual research projects and the main research content and results are as follows:(1)Analysing the radiation characteristics of terahertz frequency objects in millimeter wave and terahertz passive imaging,and discussing the evaluation method of the effectiveness of target segmentation algorithm.(2)Aimed at the traditional algorithms of segmentation are difficult to segment the target which is close to human brightness temperature.Estimated the image on the human body the regional distribution and by combining with CFAR(Constant False Alarm Rate)detection,the ATS-RG(,Adaptive Threshold Segmentation Algorithm based on Regional Growth)algorithm is proposed.Aimed at the problem that "the regional growth condition is difficult to set up",the edge detection results are used as the growth restriction conditions of regional growth.Based on the improved ATS-RG algorithm,we have achieved excellent experimental results in millimeter wave image processing.(3)A passive millimeter wave target identification sample library is set up.According to the processing results of target segmentation algorithm,the initial sample library for passive millimeter wave target recognition is established,and the sample library is expanded through rotation and scale changes.(4)Target recognition algorithm combining support vector machine and HOG(Histograms of Oriented Gradient)feature is studied.PCA(Principal Component Analysis)is applied to reduce the dimension of HOG feature.The target recognition algorithm based on convolution neural network is studied,and a convolution neural network architecture suitable for millimeter wave image recognition is built.In view of the principle of "we prefer false alarm rather than missed detection",the loss function of convolution neural network is optimized.The simulation results show that this method can effectively reduce the missed detection rate of dangerous targets under the condition of acceptable false alarm rate.