Computational Signal Pixel Imaging Based On Compressive Sensing

New imaging system of high-resolution and low-cost is an efficient and reliable way to obtain the scene information.Besides,it is the main method for obtaining high quality,multi-spectral and multi-dimensional images.Meanwhile,the new imaging system is of great significance to many differnet fields,such as industrial,military and medical.Traditional imaging methods mainly rely on large-scale array sensors.When the imaging band expand to a wider range,there will appear some problems such as low resolution,bulky,high power consumption and low efficiency.Therefore,it is necessary to explore a new imaging system to effectively solve the above problems.The combination of compressive sensing(CS)and computational imaging(CI)provides a new way to solve the new system imaging problems.Recently,CS theory has made a breakthrough development in the field of signal processing.The theory includes three parts: the sparsity of signal,the observation matrix and the signal reconstruction.The imaging system based on CS theory can reduce array sensor,cost of data and the lower requirements of the transmission channel.The main theory of computational imaging is modulating,capturing and finalizing the spatial scene information through coding aperture,and can obtain the image with high spatial resolution and high signal to noise ratio(SNR).In this paper,according to the requirements and characteristics of the new imaging system.Firstly,we model and simulate the imaging process of whole scenario based on the theory of CS and CI.We use the digital micromirror device(Digital Micromirror Device,DMD)to realize the holographic observation of the scenario,and information is sampled by the single pixel detector.Secondly,aiming to solve the shortcomings of traditional single-pixel camera(SPC),such as poor noise suppression and high complexity of system construction,we proposed a single pixel camera based on fiber collection.Besides,we analyze the system factors such as system noise,object reflectivity,the effective area of the detector which is related to the image quality.In addition,the image reconstruction algorithm is extended and improved.At the same time,the model which is trained based on convolution neural network(CNN)can achieve good recovery accuracy.Lastly,the imaging results at different wavelengths(R,G,B,SWIR)are analyzed and discussed.The results show that the imaging system can be extended to different bands and applied to many different applications.In conclusion,the new imaging system has the characteristics of high resolution,high SNR,less data and lower cost.It is of great significance to solve the bottleneck problems in the imaging technology.