Research On Image Processing Algorithms For Millimeter-Wave Security Imaging

In recent years,the issue of public security has been the focus of the international community.Due to the significant limitations of traditional security screening methods such as metal detectors and X-ray scanners,efforts are being made to research various hazardous materials detection techniques,including infrared imaging,optical detection,and millimeter wave imaging.Millimeter wave has high penetrability to fabrics and no ionizing radiation.Millimeter-wave imaging technology can realize non-contact detection of human body.This thesis investigates several key techniques in the millimeter wave security system under the background of human body security.According to the theory of millimeter-wave near-field imaging,the distant field approximate model is established to analyze the millimeter wave echo signal,and the relation between reflection coefficient of target point and echo signal in two-dimensional and three-dimensional space are obtained.The stationary phase method is used to solve the two relations respectively,and the expression of the reflection coefficient of the target point under two conditions is obtained.According to the above results,the simulation is carried out,and the millimeter-wave security imaging is realized.The characteristics and noise reduction effects of different noise reduction algorithms are investigated,and according to the image noise characteristics of millimeter-wave security imaging,a quadratic hybrid domain noise reduction combined with variance-stabilized transform is proposed.The level set segmentation is used to filter the noise in the background area of the image,and the noise pattern is transformed by the variance-stabilized transform.On this basis,the improved threshold function is used to denoise the image by wavelet threshold denoising.The method is utilized to conduct a denoising experiment on millimeter-wave security screening imaging results,and the image noise is significantly reduced,indicating that this algorithm is clearly superior to traditional denoising algorithms.In view of the actual needs of the millimeter wave security system,this thesis further investigates the application of target detection technique in the millimeter-wave security system.Establishment of target detection sample library is achieved through Ka-band millimeter-wave near-field imaging experiment,and the detection effect of the YOLOv5 algorithm on millimeter-wave images is verified.On this basis,this thesis carried out a lightweight transformation of the YOLOv5 algorithm,using the Shufflenetv2 network as a feature extraction module,which further improved the efficiency of the algorithm while reducing the number of parameters.