Research On The Application Of Asymmetric Compressed Sensing In Infrared Remote Sensing Images

Infrared remote sensing image(IRSI)is an important part of satellite-borne multi-source remote sensing image and has been more and more widely used in military and civil fields.However,the big challenge existed in this research field is how to effectively improve the resolution of IRSI.Notably,the theory of compressed sensing(CS)provides a new way of acquisition for sparse signal and sparse signal.It can reduce the storage and transmission cost of data and improve the speed of image acquisition while ensuring the accuracy of reconstructed image.Inspired by the advantages of CS in signal acquisition,this paper applies CS to improve the resolution of infrared remote sensing imaging,and makes improvements for the current mainstream CS imaging framework.At the same time,in order to maximize the use of IRSI characteristics,the visible remote sensing image is fused in the under-sampled compressed sensing domain.The main tasks of this paper include:(1)An asymmetric CS imaging framework is proposed to improve the resolution of IRSI without increasing the number of infrared sensors.At present,the CS imaging framework of multi-pixel single exposure is mainly realized by partitioning the encoding mask.The smaller the partitioning,the higher the imaging accuracy,but too small partitioning will destroy the overall structure of the image.In this paper,the concept of macroblock partition is proposed on the basis of 2 ? 2 subdivision,and the original image is reconstructed by asymmetric construction of observation values,observation matrices and sparse dictionaries.Experimental results show that the proposed framework can effectively improve the visual effect of reconstructed images.(2)Based on asymmetric CS,a new image transformation method that can directly use common fusion methods in CS domain is proposed,which solves the problem of smoothness reconstruction existed in compressed sensing domain by using common image fusion methods.The method simulates the down sampling process of the image to approximately restore the structure information of the compressed sensing domain data,this strategy can be called as the local observation normalization transform,and then fuses the image by using the normal image fusion method.The experimental results show that this method can not only effectively solve the problem of image smoothness caused by direct fusion,but also achieve the same fusion effect of image reconstruction and fusion.The time costs of the whole fusion reconstruction process can be reduced by 50%.(3)We design an infrared remote sensing imaging simulation system based on asymmetric compressed sensing.The system is divided into four modules: data loading module,asymmetric imaging module,image fusion module and reconstruction and result module.Each of these modules completes the acquisition of remote sensing image information,asymmetric compression observation and sparse representation,image fusion in CS domain,reconstruction of observation results and performance evaluation,respectively.Using this system,we can visually and clearly display the different stages of asymmetric compressed sensing and compare the reconstructed results.