Research On Automatic Image Mosaic Techniques Of Aerial Hyperspectral Remote Sensing Images

This paper to aerial hyperspectral remote sensing images automatic mosaic techniques as the research object, hyperspectral remote sensing images stitching techniques and other traditional image mosaic techniques comparison, in addition to the geographical space of stitching, also taking into account the spectral values of the match. So the research of this paper is focused on the theoretical analysis and experimental verification of the spatial mosaic and spectral splicing techniques of hyperspectral remote sensing images.In the research of hyperspectral remote sensing images segmented space techniques in the process, by reading the analysis of existing technical literature, special for hyperspectral remote sensing images, in addition to achieve spatial mosaic, but also to achieve spectral matching. Therefore, a method based on feature points matching is used to match the feature points of hyperspectral image, which is used to match the feature points of the image to be matched. The matching process consists of coarse matching and fine matching. Rough matching using correlation coefficient matching method, fine matching method using least squares. After the end of the match, using the weighted fusion algorithm, to achieve the image space mosaic. The method is not easy to be disturbed in the image space mosaic process, and the error probability is greatly reduced, and the working time is reduced, and the working efficiency is improved.In the selection of feature point matching algorithm, Moravec feature point extraction algorithm, Harris corner detection algorithm, through the comparison of the two algorithms, Moravec feature point extraction algorithm can extract more number of feature points, according to the repeatability of overlapping image feature points higher, so the Moravec algorithm for feature extraction.After the feature points matching is introduced, the nearest neighbor algorithm is used to compute the integer value of the point matching. Because each image pixel coordinates are integer values, but after by geometric transformation model transformation of the coordinates of the point of not necessarily exactly integer values, which need to transform the coordinates of the points of the adjacent four coordinate points with integer coordinates value and choose from the recent a coordinate point integer as a transformed coordinate point approximate integer coordinate values. The algorithm is simple and efficient, especially for smaller transform to the adjacent pixel gray level values, but the experience of the gray value transform large can use the bilinear interpolation method or cubic interpolation method, by seeking after the removal of the transform coordinates of points around the more integer coordinates value to calculate approximate values of integer coordinates.A lot of spectral bands of hyperspectral remote sensing images, the number of bands have dozens or even thousands, while each individual spectrum bands for each pixel of the value is not the same, even if is two pieces of mosaic image to be overlapping region of the same name point to the spectral value is not necessarily the same. According to the characteristics of hyperspectral remote sensing image band, this paper uses each band image singly spliced, segmented space also apply to single band, so in a band as an example, the first to achieve single band of segmented space, in the unity of two pieces of mosaic image pixel coordinate system, the spectrum of overlapping pixel values of spectral matching.Spectral stitching in this paper, the use of single band image stitching, the integration of all bands of the method. Therefore, taking into account the time of the algorithm calculation, using the most simple and convenient method. According to the same wavelength of the same name in the overlapping area, the spectral value is different, first to find out the relationship between the spectral value of the same name on the single wave segment, and the fitting value of the spectral value of the same name is obtained according to the relationship. In this way, we can obtain the fitting spectrum curve of the pixel by fitting the spectral value of the overlapping area on all bands.In the resolution of hyperspectral remote sensing image space information and spectral information in the splicing problem, after fusion of spectral image and the original image of RGB color composite image were pixel level fusion is realized automatic image stitching. Experiments show that the method can effectively realize the high spectral image mosaic.