Research On Super-resolution Algorithm And Object Contour Extraction Algorithm For Passive Terahertz Wave Imaging

Terahertz(THz)passive imaging uses the high penetration of terahertz frequency band,and collects the terahertz wave radiated by the target itself for imaging,which can realize the imaging of hidden objects,determine the types of hidden objects according to the imaging results,and realize the recognition of hidden objects.It is the research hotspot of security inspection technology and provides important technical support for international anti-terrorism and stability maintenance.The resolution of THz passive image acquired by THz passive imaging is limited.Before target detection and recognition,some image processing algorithms need to be used to process it,such as image denoising algorithm,image fusion algorithm,image super-resolution algorithm and image target contour extraction algorithm and so on.This thesis mainly studies image super-resolution algorithm(SR)and target contour extraction algorithm(CE)based on the actual project.For these algorithms,this thesis has carried out related theoretical analysis,method research and experimental simulation,mainly including:(1)Based on the basic theory of terahertz passive imaging technology,the radiation measurement principle of terahertz passive imaging is analyzed and studied,forming the technical route of the algorithm research in this thesis,providing theoretical support for the super-resolution algorithm and the target contour extraction algorithm of terahertz passive imaging image.(2)A super-resolution algorithm of terahertz passive imaging image based on denoising residual network is researched and proposed.In view of the loss of some image information caused by many degradation factors(such as down sampling due to the influence of antenna aperture,the loss of high-frequency details,blur due to noise,etc.),the traditional convolution neural network is not effective.Combining the filtering function of convolution operation and the inverse filtering function of deconvolution operation,it can realize the function of fuzzy elimination of convolution neural network.Compared with the way of deepening network layers to improve network performance,this method can speed up network training speed and the reconstruction effect is better than the way of deepening network layers.(3)Based on the active contour model,a contour extraction algorithm for terahertz passive imaging is researched and proposed.In order to solve the problem that the active contour model is sensitive to the initial contour position and poor convergence of the depression area,the local maximum variance between classes method is used to segment the terahertz passive image once,and then the contour is extracted on the basis of the threshold segmentation using active contour model.This method can effectively converge to the depression contour of the terahertz passive image and further improve the accuracy of contour extraction.The above work has been verified by actual experiment and simulation.The results show that these methods can effectively achieve the reconstruction of high-resolution image and the extraction of target contour.