Research On New Algorithm For Lossless Compression Of Hyperspectral Images On Satellite

Hyperspectral images are three-dimensional data cubes with rich spatial and spectral information.Hyperspectral imaging instruments can identify the material and physical state of an object by measuring the energy of electromagnetic waves emitted and reflected by the object in a specific spectral band.Therefore,it is widely used in mineral exploration,environmental monitoring,military camouflage target recognition,and other fields.After nearly thirty years of development of hyperspectral remote sensing technology,with the continuous improvement of imaging spectral resolution and spatial resolution,the volume of hyperspectral image data has also increased.Because the downlink transmission bandwidth of the satellite link is limited,the data must be compressed before being transmitted.Therefore,in order to save data storage and transmission costs,it is of great significance to study the lossless compression method of hyperspectral images.This paper first analyzes the correlation characteristics of hyperspectral images,summarizes the calculation formulas of the inter-spectral and intra-spectral correlation coefficients,and analyzes the mechanism and code implementation methods of Huffman coding and unbounded precision integer arithmetic coding in detail,which provided technical support for subsequent simulation experimentsThis paper starts with eliminating redundant information of hyperspectral intra-spectral inter-spectral spatial information,and proposes a lossless compression algorithm for hyperspectral images based on KLT transform and FLIF algorithm.KLT transform is used to remove the redundant information of inter-spectral images.The MANIAC decision tree of the FLIF algorithm can dynamically update the context content and adjust the number of contexts in real time during compression.The design of multi-context environment modeling greatly improves the compression performance.Simulation experiments show that on the public AVIRIS data set,the proposed method achieves a compressed bit rate result close to the current highest level C-DPCM-RNN algorithm in related field.In addition,according to the characteristics of the KLT transformed image,this paper proposes an acceleration method for flexible switching of multiple compressors,which shortens the compression time to one-fifth that before optimization.For certain in-orbit satellites,the satellites are equipped with fixed hyperspectral instruments,and the main types of ground objects that can be observed are relatively fixed.Therefore,this paper proposes a learnable multilevel hyperspectral image lossless compression network.The inter-spectral and intra-spectral laws of the image are learned by training from the captured data.This network extracts feature maps at three scales: large,medium,and small.The redundant information of the image is eliminated step by step.This paper established a joint probability distribution model for hyperspectral images.Combined with unbounded precision integer arithmetic coding technology,the hyperspectral image is losslessly compressed.For on-board compression tasks in typical scenarios,this method can maintain more stable compression performance.