Research And Realization On Registration Technology Of Narrow Overlapping Imaging Spectrometer Data

As the images of remote sensing are widely used in the fields of feature classification, pattern recognition, environmental monitoring, military reconnaissance, marine remote sensing, resource exploration, etc, the image processing of remote sensing has become a research hotspot. At the same time, multi-view hyperspectral imager has been widely applied because that people demand higher resolution and wider viewing angle. But multi-view imager has different detectors, which can lead to some spectrum and brightness differences between different view images. And multi-view images need to be jointed together for a whole image in the application.Therefore, for the airborne double view push-broom-type hyperspectral imager, this thesis studied spectrum registration, brightness registration and image registration of hyperspectral image data. And on this basis, the data processing software system of push-broom-type hyperspectral imager had been completed. In this thesis, the main research works can be summarized as follows:(1) Registration data preprocessing algorithm was introduced, including the integrity analysis, data format rearranging, dark current elimination, format conversion and radiation correction, and the corresponding image processing results were given.(2) After radiation correction processing, the dead pixel detection and elimination algorithm was studied. Based on the analysis of data characteristics of push-broom-type hyperspectral image, this thesis put forward an algorithm of dead pixel detection and elimination based on dark current and forecast. Experiments showed that the algorithm could effectively eliminate the dead pixel of the continuous band and continuous pixels to improve the quality of the image.(3) The spectrum registration and brightness registration of push-broom-type hyperspectral image algorithm was studied deeply. On the basis of the research and realization of the spectrum registration of hyperspectral image data, combined with the data characteristics of narrow overlap in push-broom-type hyperspectral image, this thessis proposed an adaptive brightness registration algorithm. Using the histogram of image, hardware parameters and statistical characteristic, we calculated the brightness registration coefficient, corrected image brightness and removed brightness difference. Experiments showed that this algorithm can effectively avoid the brightness registration error caused by data characteristics and eliminated the differences between the image brightness perfectly.(4) After spectrum registration and brightness registration, this thesis studied deeply the technique of image registration and mosaic. On the basis of the study in the image registration algorithm based on the latitude, longitude and mutual information, considering narrow overlapping of push-broom- type hyperspectral image data and certain corresponding relationship between the left the right view image, this thesis proposed an improved image registration algorithm based on the latitude, longitude and mutual information. Using the latitude and longitude information of image after geometric correction, the left and the right view image were matched coarsely. And then the scope of calculating mutual information was determined and the image was registered. Experiments proved that the improved algorithm can get better effect.(5) On the basis of previous works, the software debugging and system testing of the push-broom-type hyperspectral imager data processing were completed. And the results of experimental data after dealing with the software system were given.