Research On Security Inspection Method Based On Terahertz Imaging

Security inspection equipment responsible for checking whether there are dangerous goods in passengers' luggage is an important part of stations and airports,and security inspection is extremely important to ensure people's safety and reduce injuries.At present,the security gate based on X-ray identification technology is widely used to detect the articles carried by passengers,but this method will cause some physical damage to human body,and the inspection speed is relatively slow.Therefore,this paper studies a dangerous goods security inspection method based on terahertz images and deep learning,which can effectively reduce the damage of security inspection equipment to human body and further improve the security inspection speed.Firstly,the experimental samples were scanned by terahertz time domain spectrometer(TPS Spectra 1000).Before scanning,TPI Spectra upper computer is used to configure parameters before scanning,and the scanning speed reaches 28 pixels per minute.The experimental equipment can scan samples with a size of 2 cm × 2 cm at most,and the scanning results with a resolution of 0.5 mm can be obtained Based on the research background of this paper,ceramic blade,metal blade and aluminum sheet gun models are selected as experimental samples,and terahertz spectral time domain data are scanned by terahertz pulse time domain spectrometer for imaging and subsequent image recognition.Secondly,according to the characteristics of terahertz images obtained after scanning.After collecting the data set,the reconstruction of THz dangerous goods image,by comparing the image reconstruction effects of SRRNET and SRGAN two high-resolution reconstruction methods at 4 times magnification factor,it can be seen that SRGAN algorithm has more complete information,sharper edge contour and better visual effect for THz image reconstruction.Finally,YOLOv4 detection algorithm is selected to train and test the self-made terahertz dangerous goods image data set.The reconstructed YOLOv4 data set algorithm and the unreconstructed YOLOv4 data set are tested on the same terahertz image,and the experimental results and detection performance of the two are compared.The m AP value of the final test result is 65.02%,and the detection speed is 23 milliseconds,which basically meets the design requirements.