The original remote sensing image usually has serious geometric deformation, which is generally divided into two categories: systematic and non-systematic. Systematic geometric deformation is regular and predictable, so mathematical formulas or models that simulate the deformation of the remote sensing platform and the internal deformation of the remote sensors can be applied to predict it. Non-systematic geometric deformations are irregular and can be the instability of the remote sensor platform height, latitude and longitude, speed and attitude, as well as changes in the curvature of the earth and air refraction, etc., which are generally difficult to predict.
? The purpose of remote sensing image geometry correction is mainly to eliminate the geometric deformation on the image, so that the image can be accurately correspond to the field in the spatial location relationship. The general correction mainly includes: system geometry correction, projection deformation correction and geometric fine correction. Among them, the system geometry correction and projection deformation correction are mainly made by the ground receiving station before providing the information to the user, and the geometric distortion of the image has been coarsely corrected according to the conventional processing program combined with the data about the running attitude, sensor performance index, atmospheric condition, solar altitude angle and other data received from the image at the same time. For the majority of image users to get the image data, the geometric correction to really do is geometric fine correction, that is, to ensure that the accuracy to meet the requirements of the conditions, the use of certain mathematical models will be converted to the image of the required projection, so that the image and other with the same geographic parameters of the same area of the data in the spatial location of the match.
? Geometric fine correction is usually performed using the polynomial method of correction. The mechanism of this method is to establish polynomial spatial transformations and pixel interpolation operations between different images through a number of control points to realize the alignment between remote sensing images and actual geographic maps, so as to achieve the purpose of reducing and eliminating geometric distortions of remote sensing images.
1.1.1 Selection of ground control points and polynomial correction model
1. Selection of ground control points
? The selection of control points is to establish the correspondence between the pixel points on the image and the corresponding points on the map or the corresponding points on the actual features, which requires that the number of control points should be sufficiently large and the accuracy should be guaranteed to a certain extent. The precision of control points and the difficulty of selection are closely related to the image quality, feature characteristics and image spatial resolution, which is the most important step in geometric correction. The principles of ground control point selection are as follows: ground control points have obvious and clear geographic signs on the image, such as road intersections, river forks, building boundaries, farmland boundaries; the features on the ground control points do not change with time to ensure that when two images or maps of different time periods are geometrically corrected, they can be recognized at the same time; when selecting the control points on the image that has not been topographically corrected, the control points should be selected at the same topographic height; the ground control points should be selected at the same topographic height; the control points should be selected at the same geometrical height. terrain height; ground control should be evenly distributed throughout the image, and there should be a certain number of guarantees.
2. Polynomial correction model
? After the control points are selected, the pixel coordinates or corresponding geographic coordinates of the control points on the reference image and the image to be aligned should be calculated respectively. Then select the appropriate coordinate transformation function (i.e., mathematical correction model) to establish the coordinate correspondence between the reference image and the image to be aligned, usually also known as the polynomial correction model. For general geometric corrections, a primary linear polynomial correction model can be used, and for higher accuracy requirements, a quadratic or cubic polynomial correction model can be used. The purpose of polynomial correction of an image is to reposition the pixel coordinates of the image to be aligned so that they correspond to the coordinates of the reference image.
1.1.2 Image Resampling
? The distribution of the relocated image elements in the original image is not uniform, i.e., the row and column numbers of the output image image element points in the output image are not or not all integer relationships. Therefore, according to the position of each pixel on the output image in the input image, the original image needs to be resampled according to certain rules, and the interpolation calculation of luminance values is performed to establish a new image matrix.
? Commonly used interpolation methods include:
? ① Nearest neighbor method, is the closest neighboring pixel value to the new pixel, such as the original image of a pixel of the luminance value assigned to the output image of the corresponding pixel with a shadow. The advantage of this method is that the output image still maintains the original pixel value, the process is simple, and the processing speed is fast. However, this method also has limitations, that is, this method can produce a maximum of half the pixel position offset, which may cause some features in the output image of the incoherence.
? ② Bilinear interpolation, which is a linear interpolation using the pixel values of four neighboring points, given different weights according to their distance from the interpolated point. This method has the filtering effect of averaging, and the edges are subjected to smoothing, thus producing a more coherent output image. The disadvantage is that it destroys the original image element values, which can cause some problems in the subsequent classification analysis for spectral recognition.
? (iii) Three times convolutional interpolation, which uses the 16 image element values around the interpolation point and interpolates them with a three times convolution function. This is one of the more complex of the three resampling methods, which enhances the edges of the image features and has the effect of equalization and clarity, but it still destroys the original pixel values and has a large amount of computation.
? Image resampling is not only an important step in geometric correction, but is also needed in some image processing, such as between images of different time periods and spatial resolutions, as well as with other data in GIS for alignment and compositing between different layers.
? There are many results of research on atmospheric correction at home and abroad, mainly using different correction models to deal with, mainly including the following methods:
? ① Image feature model method: this is a relative atmospheric correction method, which does not need to measure the actual atmospheric conditions and the actual ground spectra, but only uses the information contained in the remote sensing image, such as some vegetation index operations can partially eliminate the atmospheric effects, and the dark target method. Generally only applies to a small range, and there are different noise in the processed image, the effect is not very good.
? ② Statistical modeling method: that is, the use of remote sensing images of selected features on the gray scale value and the corresponding imaging time field measured feature reflectance spectral values, the establishment of a statistical model, the calculation of the amount of correction to correct the entire image. This method requires measured spectral data at the time of imaging, and cannot be used to calibrate historical images for which no measured data are available in the past and image data that cannot be measured under difficult field conditions.
? ③ Theoretical modeling method: the main use of atmospheric radiation transport theory to establish the equation, the establishment of atmospheric correction model to correct atmospheric interference. The method is based on a rigorous physical model, is an absolute atmospheric correction method.
1.3.1 Remote Sensing Image Fusion and Enhancement
1. Image fusion is the synthesis of two or more images into a new image by a specific algorithm. The information contained in multi-source remote sensing image data is cooperative and complementary, as well as the redundancy of image data. In order to utilize the information of the data more reasonably and effectively, remote sensing image fusion can make a set of image data with a certain spatial resolution, spectral resolution and temporal resolution respectively, all of which are included in a unified space and time, constitute a new set of spatial information, and be fused to form a new image, which makes up for the insufficiency of a single piece of information, achieves the mutual complementation of a variety of information resources, improves the visual effect of target recognition, and improves the The precision of comprehensive analysis.
2. Remote sensing image enhancement
? The purpose of remote sensing image enhancement is to highlight relevant subject information and improve the visual effect of the image. Commonly used image enhancement methods include: image contrast adjustment, image smoothing, image sharpening, multispectral image quadratic budget and so on.
1.3.2 Image Mosaics and Image Mapping
1. When the work area involves different scenic data, the image mapping process must be mosaic processing, mosaic processing is essentially a data overlap within the scope of the process of alignment and tonal adjustment. Should be based on the distribution of the map, selected in the center of the work area of an image as a mosaic of the reference frame, the other images as a benchmark in order to quasi-near to far mosaic.
(1) image geometry alignment
? To mosaic the image to be accurately aligned, so that they are under the same spatial coordinate system. Generally used between the images using control points for alignment, in addition to the use of homonymous points for alignment, that is, based on the two scenes of data on the same amount of data alignment to another scene of data in the process, so that the overlap of the two images geometrically more consistent parts.
(2) Adjacent image color matching
? For a certain method of color matching of adjacent images, so that the different phases of the image in color coordination with each other. In order to make the established color matching equation is more accurate, the selected for the adjacent two images hue matching, adjustment of the **** the same region should be as large as possible, select a representative region for hue matching. There are sometimes clouds and various noises on the remote sensing images, which should be avoided when selecting the matching region, otherwise it will have an effect on the matching equation, thus reducing the accuracy of the hue matching. Irregular polygons (rather than simple rectangles) are used to define the image area used to establish the hue matching equation. This avoids clouds and noise, but also obtains the largest possible, representative image hue matching region in order to equalize the luminance value and contrast of the output image after mosaicing. Neighboring image color matching process to meet the "first overall after the local, and gradually adjust the details of the features" principle.
(3) Resampling
? Resampling is the process of extracting low-resolution images from high-resolution remote sensing images. Resampling can improve the efficiency of image processing, common resampling methods are the nearest neighbor pixel method, bilinear interpolation and double triple convolution method. The resampling method should be applied accurately in the orthorectification process, and at the same time, it should be ensured that the pixel size of the image correction and the resampling method meet the requirements of the project mapping and the project itself.
4) Single view image processing
a. De-clouded
? The existence of satellite image clouds will have an impact on the interpretation of the image, you can choose the appropriate algorithm or replace the image with a different phase of the method of cloud processing.
b. De-shadowing
? The shadow area can be recognized by the human eye to confirm its range, and the local adjustment of brightness and contrast within the shadow area can remove the shadow. There will be some degree of difference in the hue and brightness values between the processed shadow area and the non-shadow area, so the whole image needs to be transitioned with appropriate brightness and contrast adjustments to achieve a good visual effect.
c. Partial Color Processing
? The general image to be processed is a multi-spectral selection of the R (red), G (green), B (blue) channel for the synthesis of the resulting color image. In the RGB color system, each channel has 0 to 255***256 kinds of brightness values, and the three channel values are mixed (256 × 256 × 256) will be able to produce about 16.77 million colors, should be adjusted through the brightness values of different channels to achieve the true color of the features.
d. Multi-scene image consistency adjustment
? When the color of a scene image is adjusted to the truest hue it is possible to bring the hue of other images closer to it, this process is more complex and requires repeated attempts to accumulate experience.
e. Marquetry
? (1) Set the appropriate feathering value. Walking the mosaic line should choose the appropriate feathering value, generally the same track image of the small differences between the feathering value should also be smaller than the feathering value of the different track image selection.
? (2) go mosaic line. Multi-image splicing should be made so that the image of good quality to cover the poor quality of the image, the new image to cover the old image, the splicing of geometric edges often produce a very obvious hard edges, take the mosaic line can be eliminated after the splicing of the image splicing line obvious problems, but should pay attention to the mosaic line to try to avoid roads, rivers and other features, if you can't avoid it, then it should be increased feathering value.
2. Image mapping principles
(1) The image must be rich in levels, color uniformity, moderate contrast, clarity, and no discoloration.
(2) The error in the plane position of the randomly selected feature points on the image map is not greater than +0.5mm, and in special cases, it is not greater than +0.75mm.
(3) The absolute value of the actual and theoretical sizes of the map contour line should not exceed the limit, the edge length of the spreading point map is 0.15mm, and diagonal is 0.20mm, and the edge length of the image map is 0.20mm, and diagonal is 0.30mm.
2.
(4) The production of color image map should choose three or more multi-spectral band images, the alignment error between the bands is not greater than 0.2mm, the image registration error is not greater than 0.3mm, the production of color remote sensing image map requires the selection of the full-color bands or the need to select a band of images.