Research On Hyper-spectral Image Compression Of Spectrometer In Unmanned Aerial Vehical

Hyper-spectral remote sensing developed rapidly since the 80's of the 20th century. It can get the image data of many bands with continuous spectrum and narrow spectral regions in visible and near infrared, short waveinfrared and thermal infrared range. With hyper-spectral image, many object that can't be recognized in multispectral image can be detectd .The unmanned aerial vehical remote sensing system can get high resolution image in real time with lower cost. So it is widely used in military reconnaissance, resource investigation, agriculture, meteorology and environmental assessment. But the increased data volume of hyper-spectral image present special challenges in the acquisition, transmission, analysis and storage process. So it has applicable value to design a compression scheme that suitable for realization in VLSI, with low complexity and good performance. In this paper, with the background of the 863 project , compression of hyper-spectral image is researched, the main contents and results are:1. The analysis of the correlation in hyperspectral remote sensing images.Contrast Experiment on correlation of hyper-spectral image and ordinary image is tesred at first.The results show the correlation is almost equal among different bands,so we can use the same method to get rid of the correlation for every band; The spatial correlation coefficient of hyperspectral remote sensing images is lower than ordinary image ,so spatial decorrelate can't get ideal compression effect ;The spectral correlation of hyperspectral remote sensing images is much larger than the ordinary image's, the higer of Spectral Resolution,the stronger of spectral correlation. So the most important task in compression is spectral decorrelation.2. The coding scheme based on bit-plane. The results of bit-plane analysis shows the spectral correlation is larger in higher bit-plane.To improve the comprssion effect; we can use different measures for higher and lower bit-plane.Two schemes are propsed in this paper.In lossless scheme, DPCM was used in spectral decorrelation of higher bit-plane, then an static code strategy was applied to residual image.Lower bit-planeis are divided with quadtree. Compared with the inter-spectral bit-plane and 3-D bit-plane algorithm, the encode time is decreased by 13.1% and 6.0%, the decode time is decreased by 12.7% and 6.3%.In lossy scheme, improved JPEG-LS was used in spectral decorrelation of higher bit-plane, the PSNR reached up to 35 DB when the compression ratio is about 1:8.An effective strategy is designed in comprssion ratio control, the error is abobt 3.7% between expectation ratio and actual ratio. Lower bit-planeis is removing mean after divided with quadtree.In view of the poor compression ratio of lower bit-plane, a triangle division method is proposed. By this method, the encode and decode time is decreased by 16.54% and 10.58% respectively, similarity of reconstructed image is increased bu 0.08%.3. The coding scheme based on second order different predictive.After analyse the performance of common predictive coding scheme,a novel compression scheme that based on second order difference predictive was proposed, It use MED predictor to remove the spatial correlation and second order difference predictor to remove spectral correlation. Then a Unified predictor is designed based on the weight of predictive error. Compared with the optimal linear prediction, the entropy of residual image is decreased by 3%, the encode and decode time are decreased by 5% and 16%. Compared with the inter-spectral LOCO-I, the entropy of residual image is decreased by 4%.