Research On Target Detection And Recognition Algorithm For Passive Detecting Of Millimeter-Wave

Because passive millimeter wave security inspection technology have the advantages of non-contact,no radiation,and perspective imaging,and it can detect hidden dangerous objects hidden under clothing easily and effectively.So its application receives the high attention in important places such as airports,stations,stadiums,military camps,institutions,etc.Target detection and recognition algorithm is the core of the technology,and the performance of the algorithm directly determines the final functional effect of the relevant system.In recent years,the target detection and recognition algorithm based on deep neural network has far surpassed the traditional algorithm based on machine learning in performance.Therefore,it is an inevitable development trend and research hotspots to use the deep learning method to solve the problem of millimeter wave image target detection and recognition.Based on the actual research projects,this thesis studies and proposes corresponding solutions to the problems of low detection precision,small target missed detection and slow detection and recognition speed when using deep neural network in the targets detection and recognition on millimeter wave images.The main research contents are as follows.(1)In this thesis,the basic principles of radiation in the millimeter bands are analyzed and discussed.Based on these radiation characteristics,the basic imaging principles of millimeter-band passive imaging systems are discussed and the general functional unit structure of passive imaging systems is introduced.(2)Aiming at the problem that the detection precision of typical target detection algorithms of millimeter wave image is not high in recent years,the millimeter wave image target detection algorithm using deep convolutional neural network is studied.The deep neural network needs a large number of training sample sets,but the number of millimeter wave picture samples is relatively small.The Target Detection Algorithms Based on Transfer Learning(TDA-TL)is proposed,and the detection precision is promoted effectively.(3)Aiming at the problem of small target missed detection in target detection of millimeter wave images,the multi-scale features are utilized to detect the target in this thesis.According to the characteristics of small targets in millimeter wave images,the four different scale features are utilized respectively to detect the targets independently,and finally the detection results are combined to solve the problem of small target missed detection in this thesis.(4)In this thesis,for the requirement that the detection and recognition algorithm must run in real time in practical system applications,a method of classifying target after segmenting(CTAS)is proposed.The segmentation stage accurately segments the target by improving the maximum entropy segmentation algorithm;the classification stage designs a classification network with simple structure,sufficient feature extraction and optimization of convolution calculation method;The real-time operation is realized on the whole.The simulation experiment results and data of the project study show the effectiveness of the above methods and algorithms.