Research Of Imaging Through Scattering Media By Single Photon Detection Based On Deep Learning

Scattering imaging is an image reconstruction technology to recover the target image through scattering medium.It has many applications prospect in the fields of atmospheric imaging,biological tissue imaging and underwater imaging.The scattering medium destroys the linear propagation path of photons in traditional optical imaging,and the propagation direction of photons will be randomly changed many times in the scattering medium.Not only the scattering medium will affect the imaging,but also the intensity of light will affect.Under weak light condition,the light shows more particle characteristics of light.The traditional detector can not deal with the detection under weak light condition.The single photon detector can record the information of a single photon.It is very suitable for weak light imaging.In recent years,deep learning has developed rapidly.It has been applied in many fields.Especially in image processing.In this paper,we explore the ultra-weak light scattering imaging technology,using a single photon detector to obtain data,and combining deep learning with scattering imaging.The main research includes the following points:(1)Underwater scattering imaging algorithm based on depth learning:In our experiment we use array single photon detector which can count single photon independently.In time domain different photons are distinguished.we proposed an underwater scattering imaging technology.It can reconstruct the target in turbid water and get the information of intensity and depth.We use convolutional neural network to learn the classification model of timeof-flight histogram,and then distinguish the background photon and target photon in time domain.Deep learning classification model can effectively reduce the background noise.In the experiment,the results of different scattering conditions are analyzed,and the imaging results of different methods are compared.Experimental results show that the target can be recovered from higher concentration of scattering water by combining time-dependent single photon counting technology and deep learning.(2)Based on the depth learning method of scattering imaging in extremely weak light: a depth learning network is proposed for imaging through scattering medium in weak light.Under the condition of weak light,the data detected by SPC is photon limited,insufficient sampling,and greatly affected by Poisson noise.In this case,the traditional scattering imaging method is no longer available.Deep learning is an effective way to solve this problem.Compared with the speckle under the condition of high photon flux,in the condition of weak light,because the speckle is not completely formed,only part of the information of the object can be obtained.Poisson noise also introduces the problem of detection randomness to the speckle.When we reconstruct the dataset,we detect the original training dataset many times to expand the number of datasets,which can improve the quality of the reconstructed scattering image.At the same time,we also discuss the method of data augmentation by using the strong light speckle simulation.Compared with other two popular deep learning network structures,our method has the best performance.In the experiment,it is shown that the image can still be restored under the condition of about one photon per pixel on average.