Research On Simulation Image Training Set Construction For Passive Millimeter-wave Concealed Object Detection

With the frequent terrorist attacks and the emergence of new types of dangerous articles,people pay more and more attention to the safety checks of crowded places such as railway stations and airplanes.Passive millimeter-wave(PMMW)imaging has a great development prospect for the detection of concealed objects in security inspection due to its good penetration of clothing.In recent years,target detection based on deep learning have developed rapidly in the field of natural images.It is the current development trend to use deep learning to solve the problem of hidden object detection in PMMW imaging.Due to the technical difficulty of PMMW imaging system development and the high cost of devices,imaging needs a certain amount of manpower and time.At present,there is a lack of largescale PMMW image database,which easily leads to insufficient network training and poor feature expression ability.In order to solve the key problem of insufficient training set for concealed object detection,the brightness temperature calculation model of target scene is established based on PMMW radiation mechanism,and PMMW image in human security scene is simulated to build the training set required for deep learning network model,the main contents are as follows:Firstly,on the basis of the existing PMMW image simulation research,focusing on improving the accuracy and diversity of PMMW image simulation for human security inspection,this paper analyzes the typical objects in the scene,such as the dielectric properties of human skin,the influence of sweat,the radiation characteristics of different kinds of concealed contraband,and measures the transmission of different materials clothing,for the later construction of high-precision training set to lay the model and data foundation.The model and data obtained in this part also have positive significance for the identification of hidden objects in human security inspection.Then,according to the radiation modeling and simulation method in this paper,the real human body security scene is simulated to generate large-scale PMMW images.The experiment verifies that the simulation image has a high degree of authenticity,and then a training set is built through manual annotation and data enhancement.Finally,the PMMW imaging equipment MPSIR(Multi-Polarization Scanning Imaging Radiometer)is used to image the human body hiding metal knives,guns,spanners and flammable liquids.The hidden objects detection experiment was conducted through the SSD network after training with large-scale simulated images.Meanwhile,it was compared with the training set composed of a few measured images or a few simulated images.The experimental results show that the large-scale simulation training images can get better detection performance than the small-scale images,which not only shows that the large-scale data set is the basis of training a well-performing target detection network,but also verifies the feasibility and effectiveness of the simulation image training set constructed by the method in this paper.It can solve the problem of insufficient PMMW image data set to a certain extent,with low cost and high efficiency,and has a good research prospect.