Font Size: a A A

Research On Miniaturized Hyperspectral Imaging System Based On Snapshot Compression

Posted on:2024-09-22Degree:MasterType:Thesis
Country:ChinaCandidate:D Y LiuFull Text:PDF
GTID:2542306944468694Subject:Communication Engineering (including broadband network, mobile communication, etc.) (Professional Degree)
Abstract/Summary:PDF Full Text Request
Hyperspectral image information has the advantages of atlas integration,and is of great significance in military,archaeological,food,and agricultural fields.Traditional hyperspectral image acquisition requires spatial scanning or wavelength scanning,resulting in low acquisition speed and excessive data volume,which affects the application of hyperspectral imaging.Coded aperture snapshot spectral imaging system(CAS SI)can obtain compressed spatial spectral information of targets at the speed of video,and use reconstruction algorithms to restore complete spatial spectral information,which has a broader application prospect.Traditional CASSI systems are divided into two categories:SDCASSI and DD-CASSI.SD-CASSI has a simple structure but limited spatial clarity,while DD-CASSI uses additional optics resulting in an increased size and cost.In response to the above issues,this paper has conducted a large number of related research and achieved the following results:1.A miniaturized hyperspectral imaging system is designed and developed,which is called Reflective coded aperture snapshot spectral imaging(R-CASSI).R-CASSI’s ingenious optical path design utilizes the reflective characteristics of the mask to multiplex the relay imaging module,making the entire optical system more compact.At the same time,only one prism is used to complete two dispersions,ensuring high spatial clarity of the system’s imaging.2.An encoder-decoder network with a 3D convolutional core,called U-net-3D,is proposed for the reconstruction of compressed spectral data.U-net-3D networks can capture spectral continuity prior information from hyperspectral images,and experimentally achieve the same level of reconstruction results as current state-of-the-art algorithms.At the same time,this paper also constructs a benchmark dataset for testing different compressed spectral reconstruction algorithms.3.A miniaturized hyperspectral imaging system prototype is developed,and optical components and installation hardware are designed to meet the three requirements of small size and low weight,flexible assembly,and avoiding stray light.The system is highly integrated,and functional tests such as outdoor acquisition and handheld acquisition is conducted.These efforts promote the practical development of miniaturized hyperspectral imaging systems.
Keywords/Search Tags:Snapshot Compressive Spectral Imaging, Compressive Sensing, Deep Learning, Coded Aperture Snapshot Spectral Imaging
PDF Full Text Request
Related items