Research And Application Of Deblurring Algorithm For THz FMCW Defocused Image

Terahertz is commonly used in terahertz imaging because of its transmission ability and low single photon energy.Its imaging technology has been widely used in non-destructive testing,security and other aspects.However,compared with X-ray,terahertz has the advantages of high security,but because its wavelength is longer than X-ray,the resulting image is blurred.In order to solve the problem of defocus image deblurring,which is more obscure than the image in the focal plane,the text uses the image deblurring algorithm based on depth learning to improve the spatial resolution of the image under the experimental condition of THz FMCW imaging system.According to the image blur model,the blur reason of THz defocused image is studied.In this paper,the point spread function at the defocusing point in the imaging system is estimated by the way of measuring point source imaging,and the reason why blind deblurring is used instead of non-blind deblurring is explained.A convolutional neural network with high efficiency and high performance is designed.The problem that the network structure has no scalability is solved by using codec and cyclic structure,and the problem of gradient explosion and gradient disappearance is solved by using long jump connection.Using the fuzzy images of different defocusing distance of THz FMCW system in this paper,a targeted data set of THz defocusing image is constructed,which is expanded by various methods to ensure that the training results will not be under fitted.In order to prove the high performance of this method,the number of encoders,the number of cycles and the data set of defocusing distance are changed to verify the performance of the model.The qualitative and quantitative performance of the model is demonstrated by observing the training time,image results and no reference evaluation indexes.The advantages and practicability of this method are proved by experiments.Based on the Tkinter GUI Library of python,an image de blurring software is developed.The user interface of the software is simple,the operation is simple and fast,it is user-friendly,and it realizes common image processing functions.At the same time,the software is used to process the image which are not included in test set,and compared with the traditional deblur algorithm qualitatively and quantitatively,which proves the superiority of this method.