Research On Compression Of Multispectral Remote Sensing Image

With the orientation of high-resolution, wide coverage and multi-temporality in the development of remote sensing of Earth observation System, the remote sensing image data acquisition amount is growing drastically. However, the storage capacity and the channel downstream bandwidth of the satellite are very limited. So it is necessary to study how to reduce the transmission data rate and data storage in the compression algorithm of the sampled image, utilizing a variety of natural redundancies of the image signal.Nowadays the multi-spectral remote sensing is widely applied in the spacecraft with high ground resolution, thus the application of the image compression technology of multispectral remote sensing in a constrained environment of satellite onboard has always been a very important theoretical research topic in the field of image compression. This thesis provides in-depth theoretical analysis and applied research on the topic from multiple perspectives, with original methods proposed. The innovative research results are as follows:This thesis reviews the present domestic and overseas research status of remote sensing image compression onboard satellites, analyzes the data statistic features of the spatial and spectrum space of the multispectral&hyperspectral remote sensing image, improves the3D-SPECK algorithm according to the MHDC recommendation requirements of lossy compression for on board satellite that with zero sorting and simplified list structure it is quicker in speed and easier in hardware implementation, and proposes a new algorithm combining spectral band combination SPECK with spectral karhunen-Loeve Transformation, which improves the quality of reconstructed images over2compression rate by1to2dB compared to other similar algorithms.It is the first domestic thesis to study intensively the latest CCSDS MHDC lossless standard. It discusses the relationship of FL predictor and the LMS adaptive filtering, and studies its parameter settings and properties of multi-spectral compression. According to the multi-spectral remote sensing CCD push-broom imaging features, the author advances the V-scan prediction method that enables an increase up to19.98%in compression ratio of few spatial variation image with MHDC lossless algorithm.This thesis for the first time applies MHDC FL predictor to the compressed sensing multispectral image reconstruction and builds an applicable framework of compressed sensing of multi-spectral remote sensing. The author uses multispectral FL to predict residuals of multi-spectral signal of block compressed sensing, and then with SPL reconstructs them in multi-directional dual tree complex wavelets, so that a high-quality and quicker reconstruction of remote sensing multispectral image is achieved. In this way, memory consumption in compression sampling and amount of computation in reconstruction are greatly reduced, while the reconstructed image quality is improved by2dB or so compared to that using fast reconstruction algorithm alone or BP algorithm.On the basis of the above-mentioned work, LMS convergence control parameters of FL predictor and convergence control parameter settings of dual-threshold shrinkage SPL iteration with DTCWT are optimized through the theoretical study of the algorithm, according to the characteristics of remote sensing image scene. The author proposes an adaptive iterative residual reconstruction method within CS measurement domain prediction residual norm-2’s minimum constraints, enabling an average increase of0.3to1dB or so in the quality of reconstructed multi-spectral image compared to that without joint optimization and constraint iteration, and reducing the computation time and human interference.This study has been supported by the major projects of high-resolution earth observation system. The corresponding project was delivered in July2012.