JP2001143054A - Satellite image processing method - Google Patents

Abstract

(57) [Summary] [Problem] To remove cloud components from a cloud cover area in a satellite image,
Provided is a method for restoring the ground surface information of the overcloud area. A cloud-covered cloud area in a satellite image is extracted in a cloud-covered cloud area extraction step. Then, in the position correcting step 14, the position of the reference satellite image having no overcloud area in the same area as the image is corrected so as to overlap the satellite image. Thereafter, a relational expression of the DN value (digital number) of the pixel excluding the cloudy area of both images is created in a relational equation creating step 17, and a cloudy cloud in the reference satellite image is created in a cloudy area area restoration step 18. Restores the image excluding the effect of clouds from the DN value of the partial area.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the processing of a satellite image taken by an earth observation satellite equipped with an optical sensor, and more particularly to an area covered by clouds in a satellite image (hereinafter referred to as a cloud cover area). The present invention relates to a method for restoring the ground surface information at the time of satellite image capturing by removing clouds from the above.

[0002]

2. Description of the Related Art A database of map data representing geographical information such as residential land, farmland, roads, rivers, etc., is provided to consumers by displaying or printing necessary map data on a display unit as appropriate. Service systems are known. In such a service system, aerial photographs, satellite images, or data measured on the ground (hereinafter referred to as ground measurement data) are periodically acquired to generate map data, and the map data is generated as time passes. If the content of has changed,
The operator searches the database for the corresponding map data and corrects it. In recent years, computerized pattern recognition of aerial photographs, etc., and automatic generation of individual map data and detection of changes in the contents of multiple map data have been automated, saving labor in database maintenance work and improving the accuracy of map data correction. Attempts have also been made.

[0003] In a satellite image taken by an earth observation satellite equipped with an optical sensor at the time of cloudy sky, sunlight is blocked by clouds, and it is not possible to know the ground surface information of the overcloud area. Methods for removing noise such as clouds and shadows contained in such a satellite image are described in, for example, Japanese Patent Application Laid-Open No. H10-269347. The constituent pixels of the geographic image generated based on the weather information and the like are binarized as necessary, and the color pixels in which the number of connected pixels in the designated direction is equal to or less than a predetermined connection upper limit value are detected, and the detected color pixels are detected. Is regarded as a noise pixel and is removed from the corresponding area, or as described in JP-A-10-269354, the constituent pixels of the geographic image generated based on the satellite image data are binarized as necessary. Then, height information of a natural object or a building in the geographic image is acquired from the satellite image data, and the irradiation direction of the sunlight to the object, the irradiation angle, the width and length of the shadow of the object, and the Calculate the affected area. Then, the shadow area on the geographic image is specified by subtracting the calculated influence area from the product of the width of the shadow and its length, and the pixels of the specified shadow area are deleted. In addition, in order to prevent erroneous detection of pixel distribution, cloud information at the time of photographing included in satellite image data is obtained to predict the geographic image density, and when the predicted value is equal to or less than a predetermined reference value, the density of the image is corrected. There are methods.

[0004]

However, these methods cannot restore the ground surface information even if the cloud can be removed from the overcloud area. In order to solve such a problem, an object of the present invention is to provide a method of removing clouds in a cloud-covered area from a satellite image and restoring ground surface information in the cloud-covered area.

[0005]

According to a method of the present invention for solving the above-mentioned problems, a satellite image including a cloud-covered area is provided with a threshold value for judging a cloud in a constituent pixel to define the cloud-covered area. Extract. Then, the position of the reference satellite image that does not include the overcloud portion area obtained by photographing the same area as the image is corrected so as to overlap the satellite image. Thereafter, a relational expression of the DN value (digital number) of the pixel excluding the cloud cover region of both images is created. The DN value is obtained by discretizing the radiance measured by the earth observation satellite from 0 to 255. By using this relational expression, the DN value of the cloud cover area is calculated, and the image excluding the influence of the cloud is restored.

[0006]

Embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a satellite image processing method according to the present invention. This image processing method is realized, for example, by the information processing apparatus 100 including a general-purpose workstation or a personal computer reading the computer program 410 and executing the computer program 410. The computer program 410 is usually incorporated in the internal storage device or the external storage device of the information processing device 100, and is distributed integrally with the information processing device 100. However, the computer program 410 can be separated from the information processing device 100. Recording medium of the present invention, for example, a CD-ROM
(Compact disk type ROM), FD (flexible disk), or the like, or distributed through an information communication network or the like.
0 may be installed at any time and provided for execution at any time.

[0008] The information processing apparatus 100 includes a CD-ROM or a DVD-ROM.
A satellite image data input method 300 for reading the satellite image data 420 by the information communication network 50 from a recording medium such as a (digital video disk type ROM) or an image archive 30 storing the satellite image data 420 captured by the earth observation satellite 10. A database 400 composed of a large-capacity hard disk or the like is connected, a CRT 500 is displayed as display means for confirming the processing result, and a pointing device and a keyboard 600 are connected as necessary condition input means.

In the database 400, in addition to the storage area for the computer program 410, a storage area for satellite image data 420 and a storage area for ground measurement data 430 are formed.

[0010] The ground measurement data 430 is a raw image obtained by photographing the ground cover of a residential land, farmland, road, river, or the like with a spectrophotometer or a CCD camera capable of measuring a wavelength range from the visible region to the infrared region on the ground. Data or image processing method 200
It is data processed so that it can be used in.

FIG. 2 shows a flowchart in the image processing method 200. The method comprises three steps: a cloud cover area extraction step 201, a position correction step 240, and a cloud cover area restoration step 202. First, the cloud cover area extraction step 201 will be described.

(Cloud cover area extraction step) Data input / output step
At 210, a database 400 in which satellite image data 420 read from the satellite image data input method 300 is stored in advance.
, And converts the satellite image data 420 into a data format that can be processed by the image processing method 200. Further, between the database 400, the satellite image data 420,
The input and output of the processed image data and the ground measurement data 430 are performed.

In the cloud cover area pixel extraction step 220, a threshold value is set for the DN value of each band for the satellite image 421 including the cloud cover area shown in FIG. To extract the cloud cover area.

A specific threshold value setting method is as follows. First, three bands are selected from the visible band and the infrared band in the satellite image 421, and a color obtained by adding red, green, and blue shading to each band is selected. Create a composite image. The operator cuts out a part of the cloud cover area by viewing the color composite image. Next, the DN value of each band of the cut region is stored in a file, the lower limit value of the DN value of each band is read from the file, and the value is set as a threshold value.

Alternatively, the operator can extract an overcloud area by comparing the image obtained by mechanically classifying the satellite image by clustering and maximum likelihood classification described later with the color composite image. In addition, it is also possible to cut out several cloud cover areas from the satellite image 421 based on the color composite image and use them as teacher data to supervise and classify the satellite image 421 to extract the cloud cover area.

In the mask processing step 230, a mask of the cloud-covered area extracted in the cloud-covered area pixel extraction step 220 is created for the satellite image 421, and mask processing for excluding the mask from the subsequent steps is performed. As a result, of the area covered by the satellite image 421, only the pixels other than the masked cloud cover area are subjected to the subsequent steps.

(Position Correction Step) Next, the position correction step 240 will be described. In the position correction step 240, the reference satellite image 422 that does not include the overcloud portion area, which is obtained by photographing the same area as the satellite image 421 illustrated in FIG. Here, the satellite images 421 and 422 do not necessarily need to be captured by the same earth observation satellite 10. When satellite images captured by different earth observation satellites 10 are used, the resolution may be adjusted by resampling each satellite image. Satellite images with different observation times for the same area 42
In steps 1 and 422, the operator specifies an image portion having a characteristic shape on the CRT 500 screen, obtains a shift amount from the position information, and corrects the position shift between the images. Or JP 11
As described in -085970, a calculation position is set at a plurality of specific positions on a satellite image, a correlation coefficient between the two images is calculated,
Based on the calculation result, it is determined whether or not each of the calculated positions is appropriate as a measurement area of the displacement amount.The calculated position determined to be inappropriate is shifted by a predetermined amount, reset, and the above process is repeated. When the number of the determined measurement regions exceeds a predetermined number, a correction amount can be obtained based on the positional deviation amount calculated in each measurement region, and the positional deviation between satellite images can be corrected based on the correction amount. Thus, the satellite image 42
By performing the position correction of 1,422, it becomes possible to compare pixels at the same point.

(Uncovered Cloud Area Restoration Process) Land Cover Classification Process
Reference numeral 250 denotes a satellite image 421 obtained by masking the overcloud area in the masking step 230, and a reference satellite image 4 whose position is corrected in the position correcting step 230 so as to overlap the satellite image 421.
Classify 22 as land cover. As for the satellite image 422, if there is already an image classified as ground cover in the database 400, that image is used.

FIG. 4 shows a flowchart of the land cover classification step 250. In the clustering step 251, the satellite image 42
Cluster 1,422. Clustering is
(1) The initial value of the category center value is arbitrarily given by the number of categories to be classified, (2) each pixel is attributed to the category of the closest central value, and (3) a new value is calculated from the average value of the pixels belonging to each category. (4) Repeat the above (2) and (3) until the central value converges,
The steps are as follows. This clustering algorithm is called an iterative optimization clustering method and is described in detail in a known document (for example, Remote Sensing Digitalimage An
alysis: an introduction / John A. Richards.−2nd Editi
on, Springer-Verlag, 1900).
The initial data is set in advance for the parameters for clustering, the number of clusters, the sampling interval of pixels, and the like, but the operator can also set them each time.

In the foremost classification step 252, the satellite images 421 and 422 are foremost classified based on the cluster. The worst-case classification is a statistical method that is most often used as a classification method using statistics of a population such as teacher data and clusters. In this method, a likelihood indicating the likelihood that each pixel data belongs to each classification class is obtained, and the pixel data is classified into the classification class having the highest likelihood. The likelihood L _k (x) is defined as the probability that when pixel data x is observed, it is obtained from the classification class k. Equation 1 shows the calculation formula of the likelihood L _k (x).

[0021]

(Equation 1)

Here, x: pixel data (n-dimensional column vector) L _k (x): likelihood that pixel data x belongs to classification class k P (x | k): condition that x is observed from classification class k Probability P (k): Prior probability of classification class k In order to classify by maximum likelihood method, it is necessary to know a probability density function that can calculate P (x | k) for any x, as can be seen from Equation 1. There is. Generally, the probability density function is identified by a mean or a variance / covariance estimated from teacher data or clusters, assuming a multidimensional normal distribution. At this time, the likelihood L _k (x) of the pixel data x to the classification class k is expressed as in Expression 2. Note that P (k) in Equation 1 is the prior probability of the classification class k, which is assumed to be the same for all classes. Since the denominator is a common value for all classes, it can be ignored when comparing between classes. Are omitted here.

[0023]

(Equation 2)

Where n: dimension of feature space x: pixel data (n-dimensional column vector) L _k (x): likelihood that pixel data x belongs to classification class k u _k : average vector of classification class k (n dimensional column vector) sigma _k: covariance Detshiguma _k classification class k: the variance-covariance matrix equation ground data verification step 253 the classification class k, classification of the satellite images 421 and 422 classified top尢法If there is map information such as topographic maps and land use maps for the target area of the class, land cover classification is performed using those map information. When there is no map information, the classification class of the satellite images 421 and 422 is classified into land cover using ground measurement data 430 measured by a spectrophotometer, a CCD camera, or the like. By the above processing, the satellite image 42
It is ready to determine a pixel area to be referred to (hereinafter, referred to as a reference area) to create the relational expression of 1,422.

In the relational expression creating method selecting step 260, the reference area for creating the relational expression between the satellite images 421 and 422 is changed according to the location of the overcloud portion area in the satellite image 421. Specifically, cases are divided into the following three cases.

Case 1: In the image a, the overcloud area covers an entire category.

Case 2: In the image a, the cloud cover area exists in a certain category.

Case 3: In the image a, the overcloud area extends over a plurality of categories.

Specifically, the satellite image 422 is masked in the same manner as the satellite image 421 is masked in the mask processing step 230. Next, the number of pixels of each category of the satellite image 422 before and after the mask processing is calculated. If the number of pixels of at least one category becomes 0 after the mask processing, and the entire area is covered with the cloud cover area, it is determined that the case is 1. If the number of pixels of only one category changes after the mask processing and the overcloud area exists in a certain category, it is determined as Case 2, and the number of pixels of multiple categories changes and the number of overcloud areas becomes multiple. Case 3
Is determined.

FIG. 5 shows a specific example of Case 1. Satellite image
At 423, the cloud cover area covers the entire category, in this case, the paddy field. For Case 1, satellite image 423,
In each of the reference satellite images 424, the entire pixel region excluding the overcloud portion region is set as the reference region.

FIG. 6 shows a specific example of Case 2. Satellite image
In 425, the overcloud area exists only in the paddy field. As described above, when the cloud cover area exists only in one category, both the satellite image 425 and the reference satellite image 426 use the pixel area of one category as the reference area in FIG.

FIG. 7 shows a specific example of Case 3. Satellite image
In 427, the overcloud area spans multiple categories, such as paddy fields, forests, urban areas, and fields. In this case, the satellite image 42
7. In each of the reference satellite images 428, a pixel region of each category is used as a reference region, and a relational expression is created for each category.

In the relational expression creating step 270, a relational expression is created for the reference region of the satellite image determined in the relational expression creating method selecting step 260.

First, as shown in Expression 3, the regression coefficients a and b are obtained by the least square method, and the relational expression 4 is created.

[0035]

(Equation 3)

Here, Z: the sum of the squares of the residuals at the relational expression creation target pixel y _{i, j} : DN value of the relational expression creation target pixel of the satellite image 421 x _{i, j} : Relational expression creation target pixel of the satellite image 422 DN values of a and b: regression coefficients

[0037]

(Equation 4)

Y _{i, j} : D of the pixel in the overcloud area of the satellite image 421
N value x _{i, j} : DN value of the pixel in the overcloud portion of the satellite image 422 a, b: Regression coefficient that minimizes Z For example, in case 1, the reference region is the entire region other than the overcloud portion region , Calculation becomes huge. Therefore, Equation 4 may be obtained not for all pixels in the reference area but for pixels sampled at certain fixed intervals. Also, when Equation 4 is obtained for each category as in Cases 2 and 3, the reference area may be pixels sampled at certain intervals instead of all pixels.

When a satellite image photographed by the same earth observation satellite 10 is used as a reference satellite image, comparison is made using data in the same wavelength band, so a simple regression equation of Equation 4 is created. However, when a satellite image taken by a different earth observation satellite 10 is used as a reference satellite image, or a multispectral reference image is used to restore the overcloud area of a panchromatic image taken in a single band In such cases, data in the same wavelength band cannot be directly compared.
Four simple regression equations may be created, or multiple regression equations may be created instead of Equation 4 using a plurality of wavelength band data as reference images.

When Equation 4 is obtained for each category as in Cases 2 and 3, all other categories are masked.

In the overcloud portion area pixel restoration step 280, the DN value of the reference image 422 corresponding to the overcloud portion area of the satellite image 421 is substituted into Equation 4 created in the relational expression creation step 270, and the satellite image 4
The information of the ground surface of the 21 cloud cover area is restored.

[0042]

According to the present invention, it is possible to restore the information on the ground surface under the overclouded area from which the influence of the cloud has been removed from the image including the overclouded area.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing one embodiment of a satellite image processing method of the present invention.

FIG. 2 is a flowchart illustrating an image processing method.

FIG. 3 is a diagram showing a satellite image including a cloud-covered area and a reference satellite image not including the cloud-covered area in the same area.

FIG. 4 is a flowchart showing a land cover classification step.

FIG. 5 is an explanatory diagram for setting a reference region for creating a relational expression between satellite images covering the entire category in which a cloud cover region covers a certain category;

FIG. 6 is an explanatory diagram for setting a reference region for creating a relational expression between satellite images existing in a category where a cloud cover region exists;

FIG. 7 is an explanatory diagram for setting a reference area for creating a relational expression between satellite images in which a cloud cover area spans a plurality of categories;

[Explanation of symbols]

Reference numeral 10: Earth observation satellite, 30: Image archive, 50: Information and communication network, 100: Information processing device, 200: Image processing method, 201: Overcloud area extraction step, 240: Position correction step, 20
2 ... Cloud cover area restoration process, 300: Satellite image data input method, 400: Database, 410: Computer program, 420: Satellite image data, 430: Ground measurement data, 500
… CRT, 600… KB.

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Kazuo Tsutsui 5-2-1, Omika-cho, Hitachi City, Ibaraki Prefecture F-term within the Hitachi, Ltd. Omika Works 5B050 BA02 BA17 EA19 FA02 FA09 5B057 AA14 BA02 BA24 CA01 CA12 CB01 CB12 CD01 CE08 CE09 DA08 DB02 DC22

Claims (4)

[Claims] 1. An image processing method for removing a cloud component from a cloud-covered area in a satellite image, comprising: a cloud extraction step of extracting only pixel data of the cloud-covered area from the satellite image. A position correction step of correcting the position so that the reference satellite image taken of the same area is overlapped with the satellite image,
Using the reference satellite image corrected in the position correction step and the satellite image, a cloud component is removed from the overcloud area extracted in the cloud extraction step, and ground surface information at the time of capturing the satellite image is restored. An image processing method for a satellite image, comprising: 2. The image processing method according to claim 1, wherein the overcloud area restoration step creates a relational expression between the satellite image and the pixel data excluding the overcloud area of the reference satellite image. An image processing method for a satellite image, comprising performing a process of restoring information at the time of capturing a satellite image in a cloud-covered area using a formula. 3. The image processing method according to claim 2, wherein the step of restoring the cloud-covered area includes a reference for creating the relational expression between the satellite images according to the location of the cloud-covered area in the satellite image. An image processing method on a satellite image, which performs a process of changing a pixel area. 4. The image processing method according to claim 3, wherein the step of restoring the cloud cover area includes a step of classifying a land cover of the satellite image, and classifying a reference pixel area for creating the relational expression between the satellite images. An image processing method on a satellite image, which is created for each class.