Research On Photon Counting Fluorescence Microscopy Imaging Technology Based On Compressive Sampling

Single pixel imaging technology based on Compressed Sensing theory require only one point detector without space-distinguished ability,complete the multidimensional imaging to the object that be measured,has high sensitivity and inexpensive advantages.Fluorescence micro imaging has higher signal noise ratio and stronger specificity than conventional optical microscopy imaging technology.Single pixel fluorescence micro imaging combined single pixel imaging technology and fluorescence microscopy imaging technology,providing a new method for microscopy imaging.However,sampling and reconstruction of single pixel imaging results in a slow imaging speed,in order to solve this problem,deep learning neural network models suitable for compressed sampling are proposed.Among which,the binarized sampling reconstruction network has the advantages of less resource consumption,fast reconstruction speed and can be deployed on terminal devices to improve system integration,suitable for fast fluorescence microscopy imaging technology.In this context,this paper designed and built a photon counting fluorescence microscopy imaging system based on compressive sampling,using programmable logic of ZYNQ-7000 All Programmable So C accomplish precise micromirror deflection control and photon counting,using the platform embedded ARM processor real-time deployment of deep learning rebuilding networks,implement fluorescence microscopy imaging to objects,and research on various factors affecting imaging quality during imaging process.The main research contents and results are as follows:(1)On the basis of single pixel imaging system,change the pulse laser as a lighting source,design excitation optical path and transmitted optical path according to the principle of fluorescence micro imaging,digital micromirror device(DMD)as a structural light modulator,single photon counting module avalanches photodiodes(APD)as a weak fluorescence detector,built a photon counting fluorescence microscopy imaging system based on compressive sampling.(2)According to the system function,self-developed core control circuit based on ZYNQ platform,implement DMD deflection control,photon counting and neural network reconstruction function integrated in software and hardware collaborative platform,has high integration,real-time and inexpensive advantages.For achieving fast compressive sensing reconstruction,applying deep learning neural network to reconstruction process.Double-full-connected layer reconstruction network for binary sampling is deployed on the platform processing system,real-time transfer reconstruction images to PC.In order to verify the system core circuit,a simple single pixel micro-imaging system is built to experiment and test.(3)Complete fast compressive imaging of fluorescent balls.Design comparison experiment,research the impact of reconstruction algorithms,compressed sampling rate,and the DMD deflection frequency on imaging quality,experimental results confirmed the reconstruction network based on ARM processing can reconstruct microscopy images.In order to reduce the influence of fluorescence attenuation on imaging results,with the normalization method,each measurement value is the ratio of two adjacent frames count value,using traditional compressive sensing algorithms to reconstruct,the result shows that the method can eliminate effects and improve imaging quality.