Research On Compression Method For Hyperspectral Images Based On Anomaly Signature Protecttion

In the 80s of the 20th century, the development of Hyperspectral Remote Sensing was one of the great technical breakthroughs in Earth observation field. Sensors can obtain hyperspectral data which has continuous spectral bands information and cover the entire visible spectral range in close proximity to infrared (0.4-2.4 microns). The large amount of data, the abundant and rich information have brought tremendous challenges for hyperspectral data storage and transmission therefore, the research of an efficient compression technique is required. Comparing with common remote sensing image, Hyperspectral data is not to simply accumulate data, but multiply the information, thereby to precisely identify the micro difference between targets on the earth. There is only a little Anomaly Characteristic in huge hyperspectral data used for precise target identification and surface features classification, so for future application and making storage and transmission easier, it is important and meaningful to correctly exact these Anomaly Characteristic and protect, compress them. Based on the above background, the research in this thesis will focus on compression method for hyperspectral images based on anomaly signature protection.Firstly, we studied on the principle of the Hyperspectral imaging, and then analyzed its features involving spatial correlation, spectral correlation and spectral feature. With the knowledge learned above, against the traditional feature extraction method defects such as the noise sensibility, and unable to exhibit the time frequency domain properties of the unstable signal, we learned a new spectral analysis and feature extraction method in wavelet transform domain. Based on wavelet transform spatial frequency localization and multi-resolution, we analyzed Anomaly characteristic in each high frequency components. Through many experiments and changing the wavelet decomposition series, we gained the location correspondence between Anomaly signature and high frequency information. According to the ration computation of the frequency correspondence, we obtained the best wavelet transform decomposition series in analyzed appointed Anomaly signature, found and extracted the location of Anomaly signature.Secondly, we studied on the hyperspectral image compression based on spatial and spectral combining. On first step, we compared three typical combination of 3-D decorrelation, involving structure form, energy distribution and coding performance. On the basis of unsymmetrical characteristic of hyperspectral 3D data and complication of the Karhunen-Loeve (K-L) transform computation, we learned a combined method of wavelet transform for 2-D spatial dimension plus less complicated K-L transform for spectral dimension, to remove hyperspectral image redundancy information. We adopted some data to compute the covariance matrix and then make K-L less complicated. We did the experiment with different adopted data under different compression ratio and proved the efficient of the method.Finally, the compression method based on key information protection is studied. The existing lift method has a limitation such as unhandy lifting-up coefficient in the region of interest and increasing transmission data. In order to eliminate the limitation mentioned above, we learned a new lift coefficient controllable interleaved bit-plane lifting method, and this method achieved flexible control on image compression quality through drawing into two references. Changing references and doing more experiments has proved the advantages of the new method. Combined with Chapter 2, extracting the location information in each decomposition level of wavelet transform domain configure 3-D mask of region of interest, at the end, using tree structure coding method realized the protection and compression of region of interest.