A fast repair method for chromatic aberration bands after mosaicking remote sensing images based on cloud platform

By utilizing the random forest algorithm and image fusion technology on the Google Earth Engine platform, we can quickly repair color aberrations in remote sensing images, solving the problems of time-consuming image mosaicking and severe color aberrations in traditional methods, and improving image quality and quantitative inversion accuracy.

CN115115542BActive Publication Date: 2024-11-05CHINA UNIV OF MINING & TECH (BEIJING)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210718755.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-06-23
Publication Date
2024-11-05
Estimated Expiration
2042-06-23

AI Technical Summary

Technical Problem

Existing remote sensing image mosaicking techniques based on local ENVI software are time-consuming, and color difference restoration of single-scene images leads to image distortion in the study area. Traditional cloud platform methods cannot effectively eliminate color difference problems in mosaicked images.

Method used

On the Google Earth Engine cloud platform, Landsat images are filtered and preprocessed to segment chromatic aberration regions. Then, random forest algorithms and image fusion techniques are used to perform image matching and inpainting based on histograms and probability density functions, achieving rapid and effective elimination of chromatic aberration in the images.

Benefits of technology

It enables rapid restoration of chromatic aberration in images covering a large study area on a cloud platform, reducing image restoration time, avoiding over-restoration, and improving image quality and quantitative inversion accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115115542B_ABST
    Figure CN115115542B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for quickly repairing color difference bands after mosaicking remote sensing images based on a cloud platform, comprising the following steps: 1) obtaining a Landsat image mosaicked by multiple scene images on the Google Earth Engine (GEE) cloud platform, segmenting the image regions with different color difference, and taking the portion with an area less than 50% after segmentation as the image to be repaired, and the rest as the reference image; 2) respectively obtaining the grayscale histogram cumulative distribution function and probability density function of the reference image and the image to be repaired, and determining the grayscale threshold range; 3) using the random forest method on the GEE cloud platform to perform histogram matching on the image to be repaired, thereby realizing the uniform color processing of the remote sensing image color difference. Compared with the prior art, the present invention has the advantages of being applicable to a large-scale research area, fast speed, and good repair effect.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of optical remote sensing image processing, and more specifically, to a method for quickly repairing chromatic aberration bands after mosaicking remote sensing images based on a cloud platform. Background Art

[0002] Google Earth Engine (GEE) cloud platform brings together a large number of satellite remote sensing images. Online remote sensing image processing has become the mainstream. The image preprocessing process based on the cloud platform is different from the traditional method. The traditional method downloads a single image of the study area, performs uniform color processing to eliminate chromatic aberration, and then performs mosaic processing. The cloud-based image first obtains the mosaic of the image and then processes it, such as the maximum synthesis of annual and growing season images. Since the shooting time and climatic conditions of each image have different radiation sensitivities to the sensor, strip chromatic aberration is often generated after mosaicking during the remote sensing satellite's earth observation process, which has a serious impact on the quantitative inversion accuracy and visual effects of the remote sensing image.

[0003] The mosaicking method in traditional software ENVI and ArcGIS uses the grayscale value of one of the images in the overlapping area for matching. This can neither coordinate the overall color difference of all images in the study area, nor effectively eliminate the color difference problem of the mosaicked images. Summary of the invention

[0004] It is necessary to solve the technical problem that in the existing mosaicking technology of remote sensing images based on local ENVI software, the mosaicking process is not only time-consuming but also the image of the entire study area is distorted due to the chromatic aberration restoration based on a single-view image.

[0005] In view of the problems existing in the prior art, the purpose of the present invention is to solve the defects of the existing technology based on the mosaic process of single-scene remote sensing images, and to provide a method for quickly repairing chromatic aberration bands after mosaicking of remote sensing images based on a cloud platform.

[0006] Technical Solution

[0007] To solve the above problems, the present invention adopts the following technical solutions.

[0008] A method for quickly repairing chromatic aberration bands after mosaicking remote sensing images based on a cloud platform, comprising the following steps:

[0009] S1. Obtain multiple Landsat images that need to be mosaicked in the Google Earth Engine cloud platform;

[0010] S2, load Landsat images, and perform pre-processing operations such as image screening and cloud removal according to the image acquisition time and the geographical location of the target study area;

[0011] S3. Determine whether there is a phenomenon of obvious color difference stripes in the target study area after image mosaicking, segment the area with inconsistent color difference, and use the part with an area less than 50% as the image to be restored, and the rest as the reference image;

[0012] S4, respectively obtaining the grayscale histograms of each band of the reference image and the image to be restored, obtaining the cumulative distribution function and probability density function of each band, and determining the number of DN values ​​and the threshold range of each part of the image;

[0013] S5, statistically analyzing the probability density functions of the DN values ​​of the reference image and the image to be restored according to the histogram, statistically analyzing the cumulative distribution function according to the linear curve, and checking the difference between the distribution of the reference image and the image to be restored according to the statistical results;

[0014] S6, using the statistical DN value and probability density of the reference image as training samples for random forest classification, and using the training results to map the grayscale value of the reference image to the image to be repaired;

[0015] S7. Re-use the image fusion technology to fuse the reference image and the image to be restored.

[0016] Preferably, the images in steps S2 and S3 are Landsat images synthesized at the maximum values ​​of the monthly and annual scales, and the areas producing stripe chromatic aberration in the mosaicked images are segmented.

[0017] Preferably, the method for determining the grayscale value of the image band in step S4 is to statistically calculate the DN value and probability distribution of each band of the reference image and the image to be restored.

[0018] Preferably, the random forest algorithm in step S6 is as follows:

[0019] (1) The DN values ​​of each band of the reference image are used as the training subset, and the random forest function is used to obtain the DN value classifier of the reference image;

[0020] (2) The probability of DN values ​​of each band of the image to be restored is used as a training subset, and the random forest function is used to obtain a probability classifier for the image to be restored;

[0021] (3) Using the random forest probability classifier of the image to be restored, the probability of the DN value of the image to be restored is calculated;

[0022] (4) The result obtained in (3) is classified using the DN value classifier of the reference image, that is, the DN value of the reference image is matched with the DN value of the restoration image, thereby mapping the probability distribution of the DN value of the reference image to the image to be restored.

[0023] Preferably, in step S5, the probability density functions of the DN values ​​of the reference image and the image to be restored are statistically analyzed according to a histogram, and the cumulative distribution functions are statistically analyzed according to a linear curve, and the difference between the distribution of the reference image and the image to be restored is checked based on the statistical results.

[0024] Preferably, in step S5, the probability density functions and corresponding cumulative distribution functions of the reference image and the image to be restored are Pr(gr), Pt(gt), Hr(gr), Ht(gt), respectively. First, to make the grayscale value of the image to be restored equal to the grayscale value of the reference image, the following conditions must be met:

[0025] H r (g r )=H t (g t )

[0026]

[0027]

[0028] Let u = H r (g r ),but The DN value and probability distribution of the image in the prescribed form, that is, the reference image, can be calculated through the above calculations to make the radiosity of the reference image and the image to be restored consistent, thereby eliminating the image chromatic aberration and obtaining a high-quality satellite image with consistent chromatic aberration.

[0029] Beneficial Effects

[0030] Compared with the prior art, the advantages of the present invention are:

[0031] 1. Large-scale research area and fast speed: Using the cloud computing platform, the random forest algorithm can be quickly used to restore images in different color difference areas;

[0032] Second, there is no need to repair every scene image: based on traditional restoration methods, it is usually necessary to repair a large number of single scene images one by one and then mosaic them. The mosaicked images are based on the area that needs to be repaired, which reduces the image restoration time;

[0033] 3. Good restoration effect: The image area to be restored can be reasonably determined according to the area of ​​the image restoration in the study area, avoiding excessive restoration of the original image and reducing image distortion. The restoration effect is better than the mosaic image result based on the ENVI software platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a flow chart of the method of the present invention;

[0035] Figure 2It is the transit trajectory map of the single-view Landsat satellite image of the study area;

[0036] Figure 3 This is the image synthesized by using the annual maximum value after Landsat mosaic. The circled part is the image area to be restored, and the rest is the reference image area;

[0037] Figure 4 The probability density function and cumulative distribution function statistical result diagram of the DN values ​​of the reference image and the image to be restored;

[0038] Figure 5 This is the image restoration result map;

[0039] Figure (4a) is the probability density distribution diagram of the DN values ​​of the reference image and the image to be restored, Figure (4b) is the cumulative distribution diagram of the DN values ​​of the reference image and the image to be restored, Figure (5a) is the reference result processed by ENVI software, and Figure (5b) is the restoration result using this method. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention; it is obvious that what is described is only a part of the embodiments of the present invention, rather than all the embodiments, and all other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making creative work are within the scope of protection of the present invention.

[0041] refer to Figure 1-5 A method for quickly repairing chromatic aberration bands after mosaicking remote sensing images based on a cloud platform is shown, comprising the following steps:

[0042] S1. Obtain multiple Landsat images that need to be mosaicked in the Google Earth Engine cloud platform;

[0043] S2, load Landsat images, and perform pre-processing operations such as image screening and cloud removal according to the image acquisition time and the geographical location of the target study area;

[0044] S3. Considering the different sensitivity of satellite sensors to radiation caused by different lighting and climate conditions, the images are pre-processed by screening the study area and time and removing clouds. Then the pre-processed Landsat images are mosaicked according to the maximum value synthesis. It is determined whether there is obvious strip color difference in the mosaicked images. The areas with inconsistent color difference are segmented. The areas with an area less than 50% are used as the image area to be repaired, and the remaining parts are used as the reference image area.

[0045] S4. Obtain the grayscale histograms of the reference image and the image to be restored respectively, determine the cumulative distribution function and probability density function of each band, and determine the number and threshold range of the DN value of each part of the image; the method for determining the grayscale value of the image band is to count the different DN values ​​and probability distribution of each band of the reference image and the image to be restored, and divide the grayscale value of each band of the reference image into N columns (2 to the power of n). The degree of division can be determined according to the size of the study area, and the size of the N columns determines the division accuracy of the DN value; respectively calculate the number of pixels of the grayscale value contained in the N columns and divide it by the total number of pixels contained in the reference image area, so as to obtain N i The proportion of column DN values;

[0046] The probability density function is as follows:

[0047]

[0048] Among them, count is the total number of pixels in the study area, count N is the number of pixels counted in each part of the DN value range, N is the number of DN value range divisions, P N is the probability of the DN value of each column.

[0049] The cumulative distribution function is as follows:

[0050]

[0051] Wherein, P is the cumulative distribution function. According to the cumulative distribution functions of the reference image and the image to be restored, the cumulative distribution of the image to be restored is checked respectively, so as to determine the DN value difference of the color difference between the two.

[0052] S5. The probability density function of the DN value of the reference image and the image to be restored is statistically analyzed according to the histogram, and the cumulative distribution function is statistically analyzed according to the linear curve. The difference between the reference image and the image to be restored is checked according to the statistical results. The probability density function and the corresponding cumulative distribution function of the reference image and the image to be restored are Pr(gr), Pt(gt), Hr(gr), Ht(gt) respectively. First, to make the gray value of the image to be restored equal to the gray value of the reference image, the following conditions must be met:

[0053] H r (g r )=H t (g t )

[0054]

[0055]

[0056] Let u = H r (gr ),but The DN value and probability distribution of the image in the form specified by the image, i.e. the reference image, can be made consistent in radiometric value with the image to be restored through the above calculation, thereby eliminating the image chromatic aberration and obtaining a high-quality satellite image with consistent chromatic aberration.

[0057] S6. The statistical DN value and probability density of the reference image are used as training samples for random forest classification, and the grayscale value of the reference image is mapped to the image to be repaired using the training results. The algorithm of random forest is as follows:

[0058] (1) The DN values ​​of each band of the reference image are used as the training subset, and the random forest function is used to obtain the DN value classifier of the reference image;

[0059] (2) The probability of DN values ​​of each band of the image to be restored is used as a training subset, and the random forest function is used to obtain a probability classifier for the image to be restored;

[0060] (3) Using the random forest probability classifier of the image to be restored, the probability of the DN value of the image to be restored is calculated;

[0061] (4) using the DN value classifier of the reference image to classify the result obtained in (3), that is, matching the DN value of the reference image with the DN value of the restoration image, thereby mapping the probability distribution of the DN value of the reference image to the image to be restored;

[0062] Random Forest Algorithm Classification

[0063] The probability of the DN value of the statistical reference image is used as a training subset for random forest classification. At the same time, the DN values ​​of the image to be repaired with different probabilities are used as training subsets for random forest classification. The model selected by the classifier is standard linear regression. The number of trees and single-family tree nodes in the random forest are set. The contribution of the DN value distribution characteristics between each decision tree is determined according to the Gini index. Finally, the classification form used is determined. The formula of the Gini coefficient is:

[0064]

[0065] In the formula, p represents the feature, and there are K possible values ​​of feature p. Pk represents the proportion of the kth feature in the total number of samples. Let the training set data be D, |D| be the total number of samples, and there are K classes Ck, k = 1, 2, ..., |Ck| represents the number of samples belonging to class Ck;

[0066] S7. Re-use the image fusion technology to fuse the reference image and the image to be restored.

[0067] The above is only a preferred specific implementation of the present invention; however, the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solution and its improved conception within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A rapid repair method for chromatic aberration bands after mosaicking remote sensing images based on a cloud platform, characterized in that: The following steps are involved: S1. Obtain multiple Landsat images that need to be mosaicked in the Google Earth Engine cloud platform; S2, load Landsat images, and perform image screening and cloud removal preprocessing operations according to the image acquisition time and the geographical location of the target study area; S3. Determine whether there is a phenomenon of obvious color difference stripes in the target study area after image mosaicking, segment the area with inconsistent color difference, and use the part with an area less than 50% as the image to be restored, and the rest as the reference image; S4, respectively obtaining the grayscale histograms of each band of the reference image and the image to be restored, obtaining the cumulative distribution function and probability density function of each band, and determining the number of DN values ​​and the threshold range of each part of the image; S5. The probability density functions of the DN values ​​of the reference image and the image to be restored are statistically analyzed according to the histogram, and the cumulative distribution functions are statistically analyzed according to the linear curve. The difference between the reference image and the image to be restored is checked according to the statistical results. Specifically, the probability density functions of the reference image and the image to be restored are Pr(gr) and Pt(gt), respectively, and the cumulative distribution functions of the reference image and the image to be restored are Hr(gr) and Ht(gt), respectively. First, the grayscale value of the image to be restored is equal to the grayscale value of the reference image. The mapping relationship is as follows: H r (g r )=H t (g t ) Through the above calculations, the radiosity of the reference image and the image to be restored can be made consistent, thereby eliminating the image chromatic aberration and obtaining a high-quality satellite image with consistent chromatic aberration. S6. The statistical DN value and probability density of the reference image are used as training samples for random forest classification, and the grayscale value of the reference image is mapped to the image to be repaired using the training results. The algorithm of random forest is as follows: The DN values ​​of each band of the reference image are used as training subsets, and the DN value classifier of the reference image is obtained by using the random forest function. The probability of DN value of each band of the image to be restored is used as a training subset, and the random forest function is used to obtain the probability classifier of the image to be restored. The probability of the DN value of the image to be restored is calculated using the random forest probability classifier of the image to be restored; The results obtained in the previous step are classified using the DN value classifier of the reference image, that is, the DN value of the reference image is matched with the DN value of the restoration image, thereby mapping the probability distribution of the DN value of the reference image to the image to be restored; S7. Re-use the image fusion technology to fuse the reference image and the image to be restored.

2. According to claim 1, a rapid repair method for chromatic aberration bands after mosaicking of remote sensing images based on a cloud platform is characterized by: The images in steps S2 and S3 are Landsat images synthesized at the maximum values ​​of the monthly and annual scales, and the areas where strip color aberrations occur in the mosaicked images are segmented.

3. According to claim 1, a rapid repair method for chromatic aberration bands after mosaicking of remote sensing images based on a cloud platform is characterized by: The method for determining the grayscale value of the image band in step S4 is to statistically calculate the DN value and probability distribution of each band of the reference image and the image to be restored.