CN115063291A - Full-automatic geometric precise correction method for remote sensing image - Google Patents
Full-automatic geometric precise correction method for remote sensing image Download PDFInfo
- Publication number
- CN115063291A CN115063291A CN202210476021.5A CN202210476021A CN115063291A CN 115063291 A CN115063291 A CN 115063291A CN 202210476021 A CN202210476021 A CN 202210476021A CN 115063291 A CN115063291 A CN 115063291A
- Authority
- CN
- China
- Prior art keywords
- image
- longitude
- corrected
- latitude
- spots
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000012937 correction Methods 0.000 title claims abstract description 37
- 238000001914 filtration Methods 0.000 claims abstract description 13
- 238000006073 displacement reaction Methods 0.000 claims description 9
- 238000012952 Resampling Methods 0.000 claims description 4
- 230000001419 dependent effect Effects 0.000 claims description 4
- 238000012417 linear regression Methods 0.000 claims description 4
- 238000003908 quality control method Methods 0.000 claims description 4
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 7
- 238000005192 partition Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/80—Geometric correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a full-automatic geometric fine correction method of a remote sensing image, which comprises the steps of obtaining a plurality of control pattern spots from a reference image; respectively and sequentially storing the central longitude and latitude and data values of a plurality of control pattern spots into a longitude and latitude index file and a pattern spot data file; searching a target control pattern spot from a longitude and latitude index file and a pattern spot data file according to the longitude and latitude range of the image to be corrected, using part of the pattern spots, and performing rough matching based on a correlation coefficient to obtain the integral offset of the image; using all target control pattern spots to carry out fine image matching based on the correlation coefficient; filtering out gross errors of the same-name point pairs; and (5) performing geometric correction to obtain an image meeting the geometric positioning requirement. The characteristic significance degree of the image spots is controlled through the mean square error and the data range balance of the image spots, and stable and reliable control image spots are automatically generated; measuring the matching degree between the image spots through the correlation coefficient; the two phases are combined to realize the geometric fine correction of the remote sensing image with full automation, high precision and universality.
Description
Technical Field
The invention relates to a remote sensing image processing technology, in particular to a full-automatic geometric precise correction method for a remote sensing image.
Background
The applications of fusion, splicing, multi-temporal superposition analysis, change detection and the like of the remote sensing images all put high requirements on the geometric positioning precision of the images, and generally require that the error is within one pixel or even half of the pixel. The full-automatic and high-precision geometric precise correction has important significance for popularization and application of the remote sensing technology.
Patent document CN2021109222553 discloses a geometric fine correction method applicable to high-resolution remote sensing images, which adopts a hash algorithm to estimate the approximate position of an image to be processed in a reference base map, applies an image retrieval method in the field of computer vision to geometric fine correction of the remote sensing images, reduces redundant calculation in the matching process of the same name points, and improves the running speed; the Fourier Mellin transform blocks are used for image homonymous point matching, parameters and threshold values suitable for high-resolution remote sensing images are provided, and stable and accurate results are obtained. The method needs to estimate the approximate position of the image to be processed in the reference base map, and is not efficient and convenient; the used parameters and threshold values are only suitable for the remote sensing images of specific types, and the universality is deficient; the problem of automatic removal of gross error control points is not solved, and the final correction precision is difficult to ensure within one pixel or even half of the pixel.
Disclosure of Invention
In order to solve at least one technical problem in the background art, the present invention provides a full-automatic geometric precision correction method for remote sensing images.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a full-automatic geometric fine correction method for remote sensing images comprises the following steps:
acquiring a plurality of control pattern spots from the reference image;
respectively and sequentially storing the central longitude and latitude and the data value of the control pattern spots into a longitude and latitude index file and a pattern spot data file;
searching a target control pattern spot from the longitude and latitude index file and the pattern spot data file according to the longitude and latitude range of the image to be corrected, using part of the pattern spots, and performing rough matching based on a correlation coefficient to obtain the integral offset of the image;
using all target control pattern spots, carrying out fine image matching based on correlation coefficients, and carrying out preliminary quality control on the same-name pattern spots by setting correlation coefficient thresholds;
based on a least square regression method and an iteration method, coarse difference filtering of the homonymous point pairs is carried out;
and obtaining an image meeting the geometric positioning requirement based on geometric correction of a quadratic polynomial.
Further, the acquiring a plurality of control patches from the reference image comprises:
calculating the average value of all wave bands from visible light to near infrared from the reference image to obtain a full-color reference image;
re-projecting the panchromatic reference image to obtain a standard panchromatic image with the projection type and the spatial resolution consistent with the image to be corrected;
partitioning the standard panchromatic image, wherein the size of the partitioned blocks is determined according to the length and the width of the image to be corrected and the number of target control points of a single image to be corrected;
for each standard panchromatic image block, taking the central area of the block as a generation area for controlling the image spot;
moving the template of N x N pixel by pixel in the central area, wherein N is an odd number and is smaller than the side length of the central area; when the image is moved to a position, calculating the mean square error and the data range of the N × N image spots, and recording the coordinates, the mean square error and the data range of the central pixel image at the position;
after traversing, selecting the image spot with the largest data range in the first X image spots with the largest mean square error as the control image spot of the block; wherein X is a positive integer.
Further, the step of sequentially storing the central longitude and latitude and the data value of the control pattern spots as a longitude and latitude index file and a pattern spot data file respectively comprises:
calculating the longitude and latitude of a central pixel of the control pattern spot according to the image coordinates of the control pattern spot and the spatial reference information of the standard panchromatic image, and storing the N pixel values and the central longitude and latitude of the control pattern spot;
the longitude and latitude coordinate index file comprises three rows of data, wherein the first row is a storage serial number of the control pattern spot, the second row is the central longitude of the control pattern spot, and the third row is the central latitude of the control pattern spot;
and the image spot data file sequentially stores N pixel values for controlling the image spots in a binary format according to the storage sequence numbers.
Further, the searching a target control pattern spot from the longitude and latitude index file and the pattern spot data file according to the longitude and latitude range of the image to be corrected, using part of the pattern spots, performing rough matching based on a correlation coefficient, and acquiring the overall offset of the image comprises:
calculating the average value of all wave bands from visible light to near infrared from the image to be corrected to obtain a full-color image to be corrected;
calculating longitude and latitude coordinates of 4 corner points of the full-color image to be corrected, calculating a longitude and latitude range of the image to be corrected according to the longitude and latitude coordinates of the corner points, and searching M target control pattern spots located in the range from the index file according to the longitude and latitude range;
for each target control pattern spot, reading N pixel values from the data file according to the storage serial number of the target control pattern spot, and calculating the image coordinate [ x ] of the central pixel of the control pattern spot on the full-color image to be registered according to the central longitude and latitude coordinate and the space reference information of the full-color image to be corrected 1 ,y 1 ]；
And randomly selecting M coarse matching control patterns from the M target control patterns.
Further, for each of the coarse matching control patches, performing the following operations: controlling the central pixel image coordinate [ x ] of the pattern spot by the coarse matching 1 ,y 1 ]Performing pixel-by-pixel traversal search within the range of L pixels in radius as a center, wherein L is determined according to the maximum position deviation of the image to be corrected in a pixel unit, calculating the correlation coefficient of the N control image spots and the N image spots to be corrected when the image to be corrected traverses to a position, and recording the correlation coefficientThe position image coordinates and the correlation coefficient are recorded.
Further, the calculation formula of the correlation coefficient is as follows:
wherein p is i To control the ith pixel value of the pattern spot, q i Is the ith pixel value of the full color image to be corrected.
After traversing is finished, the position with the maximum phase relation number is taken as a matching position, and the image coordinate [ x ] of the matching position 2 ,y 2 ]And [ x ] 1 ,y 1 ]Forming a same-name point pair; correlation coefficient<Deleting the point pair of Tc; calculating the position deviation [ Delta x, Delta y ] of each remaining point pair]Taking two nearest point pairs in all position deviations as accurate matching point pairs, and calculating the average position deviation of the two accurate matching point pairs
As the overall deviation of the image to be registered.
Further, the step of using all the target control patches to perform fine image matching based on the correlation coefficient includes the following steps of:
for each of the target control patches, performing the following operations: the central pixel [ x ] of the target control pattern spot 1 ,y 1 ]Displacement of
Then, with the position as a center, performing pixel-by-pixel traversal search in a range with the radius of k pixels, wherein k is far smaller than L; calculating the correlation coefficient of the N-N control pattern spots and the N-N image pattern spots to be matched every time the image is traversed to a position, and recording the image coordinates and the correlation coefficient of the position; after traversing, taking the position with the maximum phase relation number as the optimal matching position, wherein the central pixel image coordinate [ x ] of the matching position 2 ,y 2 ]And [ x ] 1 ,y 1 ]Forming a same-name point pair; correlation coefficient<And deleting the point pairs of Tc to obtain an initial homonymous point pair set.
Further, the filtering out the gross error of the same-name point pairs based on the least square regression method and the iterative method comprises the following steps:
step 1, with x 1 ,y 1 As independent variable, with x 2 ,y 2 Performing a multiple linear regression of a least square method for the dependent variable, and calculating the overall error of each homonymous point pair according to a regression coefficient; calculating the maximum integral error E of M homonym point pairs max If E is max >T e Then, the homonymous point pairs corresponding to the maximum overall error are removed from the set; t is a unit of e Positioning longitude for the target geometry by taking pixels as units;
step 2, carrying out loop iteration on the step 1, recalculating the fitting coefficient and calculating the maximum overall error for the new homonymous point pair set until E max <＝T e Or number of remaining pairs of homologous points<When the value is 9; the rest homonymous point pairs are used as control points of the image to be corrected.
Further, equally dividing the image to be corrected into G regions, executing the steps 1-2 for each region, and taking the union of the remaining homonymous point pairs of each partition as the control points of the image to be corrected.
Further, the obtaining of the image satisfying the geometric positioning requirement based on the geometric correction of the quadratic polynomial includes:
fitting out the secondary [ x ] based on a quadratic polynomial method by using the control points after coarse filtering 1 ,y 1 ]To [ x ] 2 ,y 2 ]Using a resampling method to carry out geometric fine correction on the image to be corrected to obtain geometric positioning precision<＝T e An image of a pixel.
Compared with the prior art, the invention has the beneficial effects that:
the method controls the characteristic significance degree of the image spots through the mean square error and the data range balance of the image spots, is convenient for automatically generating stable and reliable control image spots, measures the matching degree of the image spots through the correlation coefficient, is simple and easy to operate, and has stronger universality.
Drawings
Fig. 1 is a flowchart of a full-automatic geometric fine correction method for a remote sensing image according to an embodiment of the present invention;
FIG. 2 is a processed reference image;
FIG. 3 is a diagram illustrating the distribution of control points on an image to be corrected;
FIG. 4 is a diagram showing comparison between the effects before and after correction.
Detailed Description
Example (b):
the technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
The image registration method and the image registration device have the advantages that the image registration is carried out on the basis of edge detection features such as SIFT operators in the prior art, the feature significance degree of the image spots is controlled through the mean square error and the data range balance of the image spots, stable and reliable control image spots are conveniently and automatically generated, the matching degree of the image spots is measured through correlation coefficients, and the image registration method and the image registration device are simple and easy to operate. Specifically, referring to fig. 1, the method for automatically geometrically and precisely correcting a remote sensing image according to the present embodiment includes:
101. acquiring a plurality of control pattern spots from the reference image; the reference image is a reference image with high geometric positioning precision;
102. respectively and sequentially storing the central longitude and latitude and the data value of the control pattern spots into a longitude and latitude index file and a pattern spot data file;
103. searching a target control pattern spot from the longitude and latitude index file and the pattern spot data file according to the longitude and latitude range of the image to be corrected, using part of the pattern spots, and performing rough matching based on a correlation coefficient to obtain the integral offset of the image;
104. using all target control pattern spots, carrying out fine image matching based on correlation coefficients, and carrying out preliminary quality control on the same-name pattern spots by setting correlation coefficient thresholds;
105. based on a least square regression method and an iteration method, coarse difference filtering of the same-name point pairs is carried out;
106. and obtaining an image meeting the geometric positioning requirement based on the geometric correction of the quadratic polynomial.
Therefore, through the steps, the characteristic significance degree of the image spots is controlled through the mean square error and the data range balance of the image spots, stable and reliable control image spots are conveniently and automatically generated, the matching degree of the image spots is measured through the correlation coefficient, and the method is simple and easy to operate. The invention has the advantages that the control pattern spots are prepared in advance, and the image re-projection and the image superposition are not needed to be carried out during the correction, so the processing time is greatly saved. By adopting a reasonable storage structure of the control pattern spots, the pixel values and the longitude and latitude coordinates of the pattern spots required to be controlled can be quickly positioned and read during geometric correction, so that the storage space of the control pattern spots is saved, and the overall efficiency of geometric accurate correction is improved. The quantity and the uniform distribution of the control points are ensured through image partition and blocking, and the quality of the control points is ensured through one-time control and one-time screening. The method has strong universality, shows good robustness in practical application, and can effectively overcome the influence of cloud coverage on the premise of enough control points. The method is not only suitable for geometric correction preprocessing of a ground application system, but also suitable for in-orbit geometric fine correction of remote sensing satellite images.
In an embodiment, the step 101 includes:
1. calculating the average value of all wave bands from visible light to near infrared from the reference image to obtain a full-color reference image;
2. re-projecting the panchromatic reference image to obtain a standard panchromatic image with the projection type and the spatial resolution consistent with the image to be corrected;
3. partitioning the standard panchromatic image, wherein the size of the partitioned blocks is determined according to the length and the width of the image to be corrected and the number of target control points of a single image to be corrected;
4. for each full-color image block, taking the central area of the block as a generation area for controlling the image spot;
5. and moving the template of N x N in the central area pixel by pixel, wherein N is an odd number and is smaller than the side length of the central area. When the image is moved to a position, calculating the mean square error and the data range of the N × N image spots, and recording the coordinates, the mean square error and the data range of the central pixel image at the position; wherein, the calculation formula of the mean square error is as follows:
wherein p is i Is the pixel value of the ith pixel, p mean Is the average of all pixels.
The formula for the calculation of the data range is:
RNG＝MAX{p 1 ,p 2 ,…,p N*N }-MIN{p 1 ,p 2 ,…,p N*N }
6. and after traversing, taking the image spot with the largest data range in the first 10 image spots with the largest mean square error as the control image spot of the block.
In one embodiment, the step 102 includes:
and calculating the longitude and latitude of the central pixel of the control pattern spot according to the image coordinates of the control pattern spot and the spatial reference information of the standard panchromatic image, and storing the N pixel values and the central longitude and latitude of the control pattern spot.
The storage structure for controlling the pattern spots comprises the following two files:
1. the longitude and latitude coordinate index file is convenient for quickly finding out a control pattern spot in a to-be-corrected image space range, and comprises three rows of data, wherein the first row is a storage serial number of the control pattern spot, the second row is the central longitude of the control pattern spot, and the third row is the central latitude of the control pattern spot;
2. and the data file sequentially stores N pixel values of the control pattern spots according to the storage sequence numbers in a binary format. Optionally, the N × N pixel values are linearly extended to a data range of 0 to 255, converted into byte types, and then written into the binary file.
In an embodiment, the step 103 includes:
1. calculating the average value of all wave bands from visible light to near infrared from the image to be corrected to obtain a full-color image to be corrected;
2. calculating longitude and latitude coordinates of 4 corner points of the full-color image to be corrected, calculating a longitude and latitude range of the image to be corrected according to the longitude and latitude coordinates of the corner points, and searching M target control pattern spots located in the range from the index file according to the longitude and latitude range;
3. for each target control pattern spot, reading N pixel values from the data file according to the storage serial number of the target control pattern spot, and calculating the image coordinate [ x ] of the central pixel of the control pattern spot on the panchromatic image to be registered according to the longitude and latitude coordinates of the center of the target control pattern spot and the spatial reference information of the panchromatic image to be corrected 1 ,y 1 ]；
4. Coarse matching, namely randomly selecting M coarse matching control patches from the M target control patches, and executing the following operations for each coarse matching control patch: controlling the central pixel image coordinate [ x ] of the pattern spot by the coarse matching 1 ,y 1 ]And performing pixel-by-pixel traversal search within a range with the radius of L pixels as a center, wherein L is determined according to the maximum position deviation of the image to be corrected in a pixel unit. And calculating the correlation coefficient of the N-by-N control pattern spots and the N-by-N image pattern spots to be corrected every time the image is traversed to a position, and recording the image coordinates and the correlation coefficient of the position. The correlation coefficient is calculated as follows:
wherein p is i To control the ith pixel value of the pattern spot, q i Is the ith pixel value of the full color image to be corrected.
After traversing is finished, the position with the maximum phase relation number is taken as a matching position, and the image coordinate [ x ] of the matching position 2 ,y 2 ]And [ x ] 1 ,y 1 ]A pair of homonymous points is formed. Correlation coefficient<The point pair of Tc is deleted. Calculating the position deviation [ Delta x, Delta y ] of each remaining point pair]Taking two nearest point pairs in all position deviations as accurate matching point pairs, and calculating the average of the two accurate matching point pairsDeviation of mean position
As the overall deviation of the image to be registered.
In an embodiment, the step 104 includes:
fine matching: for each of the target control patches, performing the following operations: the central pixel [ x ] of the target control pattern spot 1 ,y 1 ]Displacement of
And then, with the position as a center, performing pixel-by-pixel traversal search in a range with the radius of k pixels, wherein k is far smaller than L. And calculating the correlation coefficient of the N-by-N control pattern spots and the N-by-N image pattern spots to be matched every time the image is traversed to a position, and recording the image coordinates and the correlation coefficient of the position. After traversing, taking the position with the maximum phase relation number as the optimal matching position, wherein the central pixel image coordinate [ x ] of the matching position 2 ,y 2 ]And [ x ] 1 ,y 1 ]A pair of homonymous points is formed. Correlation coefficient<And deleting the point pairs of Tc to obtain an initial homonymous point pair set.
In one embodiment, the step 105 includes:
the automatic filtering steps of the gross error control point are as follows:
1. with x 1 ,y 1 As independent variable, with x 2 ,y 2 For dependent variables, a least squares multiple linear regression was performed, with the following formula:
x 2 ＝a 1 ·x 1 +b 1 ·y 1 +c 1 ·x 1 ·y 1 +d 1
y 2 ＝a 2 ·x 1 +b 2 ·y 1 +c 2 ·x 1 ·y 1 +d 2
and calculating the integral error of each homonymous point pair according to the regression coefficient, wherein the error calculation formula is as follows:
wherein, [ x' 2 ,y′ 2 ]Is a predicted value calculated from the regression coefficient.
Calculating the maximum integral error E of M homonym point pairs max If E is max >T e Then the homonymous point pairs corresponding to the maximum overall error are removed from the set. T is a unit of e The longitude is located for the target geometry in pixels, such as 1.0 or 0.5.
2. Step 1 is carried out in a circulating iteration mode, fitting coefficients are recalculated for the new homonymous point pair set, and the maximum integral error is calculated until E max <＝T e Or number of remaining pairs of homologous points<When it is 9. The rest homonymous point pairs are used as control points of the image to be corrected.
Preferably, the image to be corrected is equally divided into G × G regions, automatic filtering of gross error control points is performed for each region (step 1-2), and the union of the remaining homonymous point pairs in each partition is used as the control points of the image to be corrected.
Fitting out the secondary [ x ] based on a quadratic polynomial method by using the control points after coarse filtering 1 ,y 1 ]To [ x ] 2 ,y 2 ]Using proper resampling method to carry out geometric fine correction on the image to be corrected to obtain geometric positioning precision<＝T e An image of a pixel. The function mapping relationship is also applicable to geometric correction of secondary products derived from the image to be corrected, such as index products or classification products like NDVI.
The invention is further described below with reference to an application scenario example:
the method and the technical process of the invention are elaborated by taking the geometric precise correction of the 16-meter resolution image of the domestic high-resolution first-grade satellite in the range of Hunan province as an example.
Firstly, a Landsat8 OLI full-color image covering Hunan province is downloaded as a reference image, the image has high geometric positioning precision and 15 m spatial resolution, and is re-projected to Albers projection, and meanwhile, the spatial resolution is re-sampled to 16 m and is consistent with the resolution of the high-resolution first-order image. All processed images are stitched as shown in fig. 1.
The coverage area of a single high-resolution one-image is about 200km x 200km, in order to ensure that any one high-resolution one-image can obtain at least 400 control patches, the reference image is partitioned, the size of each partition is 600 pixels x 600 pixels, namely 9.6km x 9.6km, and for each partition, an ideal patch is searched within 300 pixels x 300 pixels of the central area of the partition. Moving a window of 19 pixels by 19 pixels in the target area pixel by pixel, calculating the mean square error SDE and the data range RNG of the pixel values in the sliding window at each position, and calculating the formula as follows:
RNG＝MAX{p 1 ,p 2 ,…,p N*N }-MIN{p 1 ,p 2 ,…,p N*N }
and taking 10 positions with the maximum SDE in all the positions, selecting the image spot at the position with the maximum RNG in the 10 positions as an ideal image spot, writing 19 x 19 pixel values of the image spot into a binary file, calculating the longitude and latitude coordinates of a central pixel of the image spot, and writing the storage serial number, the longitude and the latitude into another longitude and latitude index file.
And after all the blocks of the reference image are processed, generating 4480 control pattern spots in total, and finishing the preparation work of the control pattern spots of the whole Hunan province. The total size of the longitude and latitude index file and the image spot binary data file is less than 7M, and a 3G space is required for storing the reference image.
A high-resolution first-order image is downloaded from a Chinese resource satellite application center, the name of the image is GF1_ WFV2_ E112.4_ N27.6_20191211_ L1A0004466109.zip, after radiometric calibration, fast atmospheric correction and RPC orthorectification, a 16-meter resolution image of Albers projection is obtained, and the average value of all four wave bands is calculated to obtain an approximate full-color image. Calculating longitude and latitude coordinates of the four angular points to obtain a longitude and latitude range of the image as follows: the east longitude is 111.00421 degrees to 113.73613 degrees, and the north latitude is 26.36083 degrees to 28.83947 degrees.
The storage serial numbers and the central longitude and latitude of 4480 control patches are read from the index file, and 806 index information in the range is obtained by taking the longitude and latitude range of the image to be corrected as a selection condition. 80 pieces of index information are randomly selected from 806 pieces of index information to serve as the index information of the coarse matching control patch. And for each piece of coarse matching index information, quickly positioning a pointer from a binary file for storing control pattern data according to the storage sequence number and reading 19 × 19 control pattern data values. Calculating the row number [ x ] of the central pixel of the control pattern spot on the image to be corrected according to the longitude and latitude and the spatial reference information of the image to be corrected 1 ,y 1 ]If [ x ] 1 ,y 1 ]And if the pixel value of the image to be corrected, which is positioned outside the image to be corrected or at the position, is the background value, the matching is abandoned. Otherwise is in [ x ] 1 ,y 1 ]Matching sliding windows with the radius of 20 pixels as the center, and calculating the correlation coefficient of the control image spot and the image spot to be corrected at each position, wherein the calculation formula of the correlation coefficient is as follows:
taking the position with the maximum correlation coefficient as the best matching position [ x ] 2 ,y 2 ]Calculate [ x ] 2 ,y 2 ]And [ x ] 1 ,y 1 ]A displacement amount [ Δ x, Δ y ] therebetween]Calculating the difference between two displacement amounts, taking the two displacement amounts with the minimum difference, and calculating the average value of the two displacement amounts
As the overall displacement of the image to be corrected. Of the present embodiment
Is [2,6 ]]。
For each of the 806 pieces of index information, the following operations are performed: controlling patch data from storage according to storage sequence numberFast locate the pointer and read the 19 x 19 control spot data values. Calculating the row number [ x ] of the central pixel of the control pattern spot on the image to be corrected according to the longitude and latitude and the spatial reference information of the image to be corrected 1 ,y 1 ]To proceed with
After displacement, [ x ] is obtained 1 ′,y 1 ′]If [ x ] 1 ′,y 1 ′]And if the pixel value of the image to be corrected, which is positioned outside the image to be corrected or at the position, is the background value, the matching is abandoned. Otherwise is in [ x ] 1 ′,y 1 ′]And matching the sliding window within the range of 5 pixels in radius as the center, calculating the correlation coefficient of the control image spot and the image spot to be corrected at each position, calculating the maximum correlation coefficient of all positions, and if the maximum correlation coefficient is less than 0.65, discarding the dotted pair. Otherwise, record the row and column number [ x ] of the best matching position 2 ,y 2 ]And [ x ] 1 ,y 1 ]Forming a pair of homonymous points. After all the index information processing is completed, 201 groups of same-name point pairs are left.
Although the condition that the correlation coefficient is greater than 0.65 preliminarily controls the quality of the control points with the same name, it cannot be guaranteed that each point pair is correctly matched, and no matter how good the matching algorithm is, the mismatched point pairs are inevitable, so the gross error filtering needs to be performed on the remaining 204 groups of the point pairs with the same name, and the process is as follows: with x 1 ,y 1 As independent variable, with x 2 ,y 2 As dependent variables, multiple linear regression was performed:
x 2 ＝a 1 ·x 1 +b 1 ·y 1 +c 1 ·x 1 ·y 1 +d 1
y 2 ＝a 2 ·x 1 +b 2 ·y 1 +c 2 ·x 1 ·y 1 +d 2
substituting the coefficient obtained by regression into the formula to obtain [ x 2 ,y 2 ]Predicted value of [ x' 2 ,y′ 2 ]And calculating the integral error of each point pair, wherein the calculation formula is as follows:
calculating the maximum overall error E of all point pairs max If E is max >And 1, deleting the point pair with the maximum error. The above process is iterated until E max <1 or number of remaining pairs<9. After filtering, 93 sets of pairs of identical-name points remain, and the distribution of the pairs on the image to be corrected is shown in fig. 2. The integral error of the two pixels is within 1 pixel, and the two pixels can be used for geometric fine correction of an image to be corrected.
The geometric fine correction is carried out on the image to be corrected by using the homonymous point pair set, a quadratic polynomial fitting correction method and a bilinear interpolation resampling method, the attached figure 3 is a comparison schematic diagram of the effect before and after correction, the obvious dislocation phenomenon of the river before correction can be seen from the diagram, and the river after correction can be well corrected.
The above embodiments are only for illustrating the technical concept and features of the present invention, and the purpose thereof is to enable those skilled in the art to understand the contents of the present invention and implement the present invention accordingly, and not to limit the protection scope of the present invention accordingly. All equivalent changes or modifications made in accordance with the spirit of the present disclosure are intended to be covered by the scope of the present disclosure.
Claims (10)
1. The utility model provides a full-automatic geometric correction method of remote sensing image which characterized in that includes:
acquiring a plurality of control pattern spots from the reference image;
respectively and sequentially storing the central longitude and latitude and the data value of the control pattern spots into a longitude and latitude index file and a pattern spot data file;
searching a target control pattern spot from the longitude and latitude index file and the pattern spot data file according to the longitude and latitude range of the image to be corrected, using part of the pattern spots, and performing rough matching based on a correlation coefficient to obtain the integral offset of the image;
using all target control pattern spots, carrying out fine image matching based on correlation coefficients, and carrying out preliminary quality control on the same-name pattern spots by setting correlation coefficient thresholds;
based on a least square regression method and an iteration method, coarse difference filtering of the same-name point pairs is carried out;
and obtaining an image meeting the geometric positioning requirement based on geometric correction of a quadratic polynomial.
2. The method for fully automatic geometric refinement of remote sensing images of claim 1, wherein said obtaining a plurality of control patches from a reference image comprises:
calculating the average value of all wave bands from visible light to near infrared from the reference image to obtain a full-color reference image;
re-projecting the panchromatic reference image to obtain a standard panchromatic image with the projection type and the spatial resolution consistent with the image to be corrected;
partitioning the standard panchromatic image, wherein the size of the partitioned blocks is determined according to the length and the width of the image to be corrected and the number of target control points of a single image to be corrected;
for each standard panchromatic image block, taking the central area of the block as a generation area for controlling the image spot;
moving the template of N x N pixel by pixel in the central area, wherein N is an odd number and is smaller than the side length of the central area; when the image is moved to a position, calculating the mean square error and the data range of the N × N image spots, and recording the coordinates, the mean square error and the data range of the central pixel image at the position;
after traversing, selecting the image spot with the largest data range in the first X image spots with the largest mean square error as the control image spot of the block; wherein X is a positive integer.
3. The method of fully automatic geometric refinement of remote-sensing images according to claim 2, wherein said storing the central latitude and longitude and data values of a plurality of said control patches as a latitude and longitude index file and a patch data file, respectively, in sequence comprises:
calculating the longitude and latitude of a central pixel of the control pattern spot according to the image coordinates of the control pattern spot and the spatial reference information of the standard panchromatic image, and storing the N pixel values and the central longitude and latitude of the control pattern spot;
the longitude and latitude coordinate index file comprises three rows of data, wherein the first row is a storage serial number of the control pattern spot, the second row is the central longitude of the control pattern spot, and the third row is the central latitude of the control pattern spot;
and the image spot data file sequentially stores N pixel values for controlling the image spots in a binary format according to the storage sequence numbers.
4. The method of claim 3, wherein the searching for the target control patches from the latitude and longitude index file and the patch data file according to the latitude and longitude range of the image to be corrected, using some of the patches to perform coarse matching based on the correlation coefficient, and obtaining the overall offset of the image comprises:
calculating the average value of all wave bands from visible light to near infrared from the image to be corrected to obtain a full-color image to be corrected;
calculating longitude and latitude coordinates of 4 corner points of the full-color image to be corrected, calculating a longitude and latitude range of the image to be corrected according to the longitude and latitude coordinates of the corner points, and searching M target control pattern spots located in the range from the index file according to the longitude and latitude range;
for each target control pattern spot, reading N pixel values from the data file according to the storage serial number of the target control pattern spot, and calculating the image coordinate [ x ] of the central pixel of the control pattern spot on the panchromatic image to be registered according to the longitude and latitude coordinates of the center of the target control pattern spot and the spatial reference information of the panchromatic image to be corrected 1 ,y 1 ]；
And randomly selecting M coarse matching control patterns from the M target control patterns.
5. The method for fully automatic geometric refinement of remote-sensing images according to claim 4,
for each of the coarse matching control patches, performing the following operations: controlling the central pixel image coordinate [ x ] of the pattern spot by the coarse matching 1 ,y 1 ]And performing pixel-by-pixel traversal search within the range with the radius of L pixels as the center, wherein L is determined according to the maximum position deviation of the image to be corrected by taking the pixel as a unit, calculating the correlation coefficient of the N control pattern spots and the N image spots to be corrected every time the image to be corrected traverses to a position, and recording the image coordinates and the correlation coefficient of the position.
6. The method for fully automatically geometrically fine correcting remote-sensing images according to claim 5, wherein said correlation coefficient is calculated as follows:
wherein p is i To control the ith pixel value of the pattern spot, q i The ith pixel value of the full-color image to be corrected;
after traversing is finished, the position with the maximum phase relation number is taken as a matching position, and the image coordinate [ x ] of the matching position 2 ,y 2 ]And [ x ] 1 ,y 1 ]Forming a same-name point pair; correlation coefficient<Deleting the point pair of Tc; calculating the position deviation [ Delta x, Delta y ] of each remaining point pair]Taking two nearest point pairs in all position deviations as accurate matching point pairs, and calculating the average position deviation of the two accurate matching point pairs
As the overall deviation of the image to be registered.
7. The method for fully automatically geometrically fine correcting remote-sensing images according to claim 6, wherein said using all target control patches, performing fine image matching based on correlation coefficients, and performing preliminary quality control of the same-name patches by setting correlation coefficient threshold values comprises:
for each of the target control patches, performing the following operations: the central pixel [ x ] of the target control pattern spot 1 ,y 1 ]Displacement of
Then, with the position as a center, performing pixel-by-pixel traversal search in a range with the radius of k pixels, wherein k is far smaller than L; calculating the correlation coefficient of the N-N control pattern spots and the N-N image pattern spots to be matched every time the image is traversed to a position, and recording the image coordinates and the correlation coefficient of the position; after traversing, taking the position with the maximum phase relation number as the optimal matching position, and taking the central pixel image coordinate [ x ] of the matching position 2 ,y 2 ]And [ x ] 1 ,y 1 ]Forming a same-name point pair; correlation coefficient<And deleting the point pairs of Tc to obtain an initial homonymous point pair set.
8. The method for fully automatically geometrically fine correcting remote-sensing images according to claim 7, wherein said filtering out gross errors of homonymous point pairs based on least squares regression and iteration comprises the steps of:
step 1, with x 1 ,y 1 As independent variable, with x 2 ,y 2 Performing a multiple linear regression of a least square method for the dependent variable, and calculating the overall error of each homonymous point pair according to a regression coefficient; calculating the maximum integral error E of M homonym point pairs max If E is max >T e Then, the homonymous point pairs corresponding to the maximum overall error are removed from the set; t is e Positioning longitude for the target geometry by taking pixels as units;
step 2, carrying out loop iteration on the step 1, recalculating the fitting coefficient and calculating the maximum overall error for the new homonymous point pair set until E max <＝T e Or the number of the remaining homonymous point pairs<When the value is 9; the rest homonymous point pairs are used as control points of the image to be corrected.
9. The method of claim 8, wherein the image to be corrected is equally divided into G x G regions, and for each region, steps 1-2 are performed separately, and the union of the remaining pairs of homologous points in each region is used as the control point of the image to be corrected.
10. The method of claim 8, wherein said geometric correction based on a quadratic polynomial to obtain an image satisfying geometric orientation requirements comprises:
fitting out the secondary [ x ] based on a quadratic polynomial method by using the control points after coarse filtering 1 ,y 1 ]To [ x ] 2 ,y 2 ]Using a resampling method to carry out geometric fine correction on the image to be corrected to obtain geometric positioning precision<＝T e An image of a pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210476021.5A CN115063291A (en) | 2022-04-29 | 2022-04-29 | Full-automatic geometric precise correction method for remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210476021.5A CN115063291A (en) | 2022-04-29 | 2022-04-29 | Full-automatic geometric precise correction method for remote sensing image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115063291A true CN115063291A (en) | 2022-09-16 |
Family
ID=83196861
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210476021.5A Pending CN115063291A (en) | 2022-04-29 | 2022-04-29 | Full-automatic geometric precise correction method for remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115063291A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116107966A (en) * | 2023-04-14 | 2023-05-12 | 中国水利水电科学研究院 | Continuous noctilucent remote sensing data anomaly discrimination and interpolation method |
| CN116958833A (en) * | 2023-09-20 | 2023-10-27 | 青岛浩海网络科技股份有限公司 | GF4 data geometric fine correction method |
-
2022
- 2022-04-29 CN CN202210476021.5A patent/CN115063291A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116107966A (en) * | 2023-04-14 | 2023-05-12 | 中国水利水电科学研究院 | Continuous noctilucent remote sensing data anomaly discrimination and interpolation method |
| CN116958833A (en) * | 2023-09-20 | 2023-10-27 | 青岛浩海网络科技股份有限公司 | GF4 data geometric fine correction method |
| CN116958833B (en) * | 2023-09-20 | 2023-12-08 | 青岛浩海网络科技股份有限公司 | GF4 data geometric fine correction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106600622B (en) | A kind of point cloud data segmentation method based on super voxel | |
| CN115063291A (en) | Full-automatic geometric precise correction method for remote sensing image | |
| KR100488685B1 (en) | Image Processing Method for Automatic Image Registration and Correction | |
| CN109917356B (en) | Airborne laser scanning system error calibration method | |
| US5606627A (en) | Automated analytic stereo comparator | |
| CN113240740B (en) | Attitude measurement method based on phase-guided binocular vision dense marking point matching | |
| CN114936971A (en) | Unmanned aerial vehicle remote sensing multispectral image splicing method and system for water area | |
| CN108759788B (en) | Unmanned aerial vehicle image positioning and attitude determining method and unmanned aerial vehicle | |
| CN110223389B (en) | Scene modeling method, system and device fusing image and laser data | |
| CN101826206B (en) | Camera self-calibration method | |
| KR20190026452A (en) | A method of automatic geometric correction of digital elevation model made from satellite images and provided rpc | |
| Wang et al. | Planar block adjustment and orthorectification of ZY-3 satellite images | |
| CN108564532B (en) | Large-scale ground distance satellite-borne SAR image mosaic method | |
| CN106382917A (en) | Continuous acquisition method for three-dimensional information of indoor environment | |
| CN114755661A (en) | Parameter calibration method and device for mobile laser scanning system | |
| CN109709551A (en) | A kind of regional network planimetric adjustment method of satellite-borne synthetic aperture radar image | |
| CN103115627B (en) | Multi-track combined on-track geometrical checking method of remote sensing satellite linear array sensor | |
| CN107705272A (en) | A kind of high-precision geometric correction method of aerial image | |
| KR102015817B1 (en) | A method of automatic correction of provided rpc of stereo satellite images | |
| Özcihan et al. | A comprehensive analysis of different geometric correction methods for the Pleiades-1A and Spot-6 satellite images | |
| CN115147471A (en) | Laser point cloud automatic registration method based on curvature density characteristics | |
| CN111623957B (en) | Point cloud registration and splicing method for X-ray focusing lens splicing interference detection | |
| CN114494020A (en) | Data splicing method for cable channel point cloud data | |
| Cao et al. | Precise sensor orientation of high-resolution satellite imagery with the strip constraint | |
| Zhang et al. | Registration of CBERS-02B satellite imagery in quick GIS updating |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |