Five-spectrum satellite remote sensing image cloud shadow identification method based on cloud cover constraint

Abstract

The invention relates to a five-spectrum satellite remote sensing image cloud shadow identification method based on cloud cover constraint. The method comprises the following steps: converting spectral radiance into surface reflectance; a relatively accurate water body position is obtained; calculating gradient and slope direction information of the image area, and performing terrain correction on the near-infrared spectrum; performing threshold iteration in a cloud-free and water-free area in the image by taking the cloud cover as a main constraint condition, and searching a potential shadow area; carrying out cloud shadow position prediction on clouds in the image one by one, and matching cloud shadows with potential shadows; judging and supplementing the residual unmatched potential shadows one by one; carrying out statistics on the cloud shadow matching condition of the whole image; and guiding filtering is carried out on the obtained cloud shadow mask, refining and supplementing of the cloud shadow edge are realized, and a final cloud shadow result is obtained. The five-spectrum satellite cloud shadow recognition method helps to solve the problems that a current five-spectrum satellite cloud shadow recognition method is not high in accuracy and is difficult to put into practical application.

Description

Technical Field

[0001] The invention relates to the technical field of optical remote sensing, and in particular to a cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints. Background Art

[0002] The International Satellite Cloud Climatology Project estimates that the average annual global cloud cover is as high as 66%. Due to the limitations of the optical remote sensor structure, the presence of a large number of clouds and cloud shadows will pollute satellite observation images, resulting in the obscuration of important land objects, seriously affecting the availability of remote sensing data. Therefore, the detection of clouds and cloud shadows in optical remote sensing images is crucial. The estimation of the amount of clouds and cloud shadows can be used as a basis for evaluating image quality and image availability, and is an essential step in the satellite image processing process.

[0003] Clouds account for a large proportion of optical remote sensing images and have a great impact, so the current research on cloud recognition algorithms is relatively mature. As an additional function of cloud recognition algorithms, cloud shadow recognition has also gradually attracted attention. Compared with cloud recognition algorithms, cloud shadow recognition algorithms are more difficult to develop: on the one hand, the accuracy of cloud shadow recognition algorithms is heavily dependent on cloud recognition algorithms, and therefore is also restricted by the accuracy of cloud recognition algorithms; on the other hand, cloud shadows are very easily confused with low-reflective objects, and there is currently no very effective means to completely distinguish between the two.

[0004] The most common satellite spectral band setting in the remote sensing field is a five-band setting, that is, a full color spectrum band plus four standard multi-spectral bands of red, green, blue, and near-infrared. This design is not only technically mature and can reduce costs, but also easy to interpret visually and sufficient to meet most remote sensing observation needs. However, compared with optical remote sensing satellites with more observation bands, the spectral information provided by the images taken by the five-band satellite is not detailed and comprehensive enough, especially in the infrared range that is more sensitive to shadows, there is only one available band, which brings greater challenges to cloud shadow identification.

[0005] Currently, there are few cloud shadow recognition algorithms for five-spectral satellite images, and almost all of them are derived from classic multi-spectral (greater than five-spectral) algorithms. Cloud shadow recognition mainly depends on the radiation information contained in each spectral image. Therefore, the limitation on the number of available spectral bands will inevitably lead to a decrease in algorithm accuracy and fail to meet application requirements. Summary of the invention

[0006] The present invention aims to solve the technical problem of insufficient cloud shadow recognition accuracy caused by the limitation of the number of satellite spectral bands in the prior art, and provide a cloud shadow recognition method for five-spectral-band satellite remote sensing images based on cloud amount constraints.

[0007] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0008] A cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints includes the following steps:

[0009] Step 1: Convert the DN value of the satellite image into spectral radiance, and then convert the spectral radiance into apparent reflectance; then, based on the 6S model, remove the influence of the atmosphere on radiation transmission, obtain the true reflectance of the surface, and execute step 2;

[0010] Step 2: Use reflectivity information to identify water bodies in satellite images, and obtain the water body locations in the satellite image area based on the GSWO water body frequency dataset. Combine the results of the two to obtain a relatively accurate water body location, and then execute step 3.

[0011] Step 3: read the DEM data of the satellite image area, calculate the slope and aspect information of the satellite image area, and perform terrain correction on the near-infrared spectrum based on the SCS+C terrain correction model, and then execute step 4;

[0012] Step 4: Comprehensively consider the full spectrum reflectance, near-infrared spectrum reflectance, cloud mask, and water mask. For satellite images that meet the cloud cover requirements, use cloud cover as a constraint to perform threshold iteration in the cloudless and waterless areas of the satellite image to find potential shadow areas. The iteration method is:

[0013]

[0014] Among them, P shadow is the theoretical cloud shadow volume; n p is the total number of pixels that meet the threshold condition after the pth iteration; n0 is the total number of pixels in the cloud-free and water-free area before the iteration; T PAN,p is the threshold value of the full chromatogram segment at the pth iteration; T NIR,p is the near-infrared spectral threshold at the pth iteration; ρ PAN,q is the reflectance of the qth pixel in the full color spectrum, ρ NIR,q is the reflectance of the qth pixel in the near-infrared spectrum; n p,q It is the mark of the qth pixel satisfying the threshold condition after the pth iteration; the iteration termination condition is:

[0015] P cloud / 2<P shadow ×c<P cloud

[0016] Among them, P cloud is the cloud cover; c is the scale factor;

[0017] After each iteration, the rationality of the threshold is judged based on the reflectivity mean and standard deviation of the potential shadow area and the overall reflectivity mean information of the clean atmosphere area. Only when the iteration termination condition and the threshold rationality condition are met, it is determined that a suitable potential shadow threshold has been found; otherwise, the threshold will be updated, and the p-th threshold will be obtained by using the maximum inter-class variance method for the reflectivity of all pixels that meet the threshold requirements after the p-1th iteration, and the iteration will continue; each time, only the threshold of one spectral segment is updated, with the near-infrared spectral segment being given priority, and only when the updated near-infrared spectral segment threshold cannot meet the threshold rationality condition, the full chromatographic segment threshold is updated; if the iteration termination condition and the threshold rationality condition cannot be met through iteration, the threshold is set to a specific empirical value; after confirming the potential shadow area, step five is executed;

[0018] Step 5: Use the zenith angle and azimuth of the satellite sensor, the zenith angle and azimuth of the sun, and the relative height and position information of the cloud to predict the cloud shadow position of each cloud in the satellite image.

[0019]

[0020] Among them, (x img ,y img ) is the coordinate of the cloud pixel in the image, (x nadir ,y nadir ) is the projection coordinate of the lowest point of the cloud pixel on the ground, (x shadow ,y shadow ) is the coordinate of the corresponding cloud shadow pixel, h c is the distance between the cloud and the ground, θ υ and are the zenith angle and azimuth angle of the satellite sensor, θ s and are the solar zenith angle and azimuth, respectively, and γ is the angle between the true north direction and the y-axis direction of the image; for each cloud, the degree of coincidence between the predicted cloud shadow and the potential shadow area determined in step 4 is calculated; since the cloud height is unknown, it is necessary to set a certain step size to iterate the cloud height until the predicted cloud shadow can match the potential shadow area or stops the iteration after exceeding the cloud height range;

[0021] The evaluation indicators for evaluating the degree of fit between each cloud shadow and the potential shadow area are: cloud shadow matching degree M, shadow matching degree M s and shadow validity M e :

[0022]

[0023] Among them, n m For all matching pixels of cloud shadows and potential shadows; n o The pixels in the cloud shadow that are beyond the image boundary; nu is the non-cloud pixel in the cloud shadow that has no matching potential shadow; n s All pixels contained in the potential shadow block that matches the cloud shadow. For a potential shadow block, as long as it contains pixels that match the cloud shadow, the shadow block is considered to be matched; n sm n s The shadow in the middle corresponds to the number of shadow pixels outside the image. The cloud pixel position is inferred from the current shadow position and the geometric relationship between the cloud and the cloud shadow; n so n s The number of shadow pixels that can match cloud pixels in ;

[0024] For each cloud shadow, after each cloud height iteration, it is necessary to determine whether it meets the cloud shadow matching condition. When M is greater than the set threshold, it is preliminarily determined that it has matched the potential shadow, and then the shadow matching degree and shadow validity conditions are determined; when M s and M e are all greater than the set threshold, all matching potential shadows are considered as cloud shadows, and the iteration ends; s and M e If one of them is smaller than the threshold, only the matched pixels are saved and the iteration ends; s and M e If all the clouds are smaller than the set threshold, the current result will be abandoned and the iteration will continue; until all the clouds have completed the above iterative matching, the step 6 will be executed;

[0025] Step 6: Identify and supplement the remaining unmatched potential shadows one by one;

[0026] The cloud coordinates are calculated by using the average distance between the matched cloud and cloud shadow and the current shadow position. When the cloud coordinates are outside the satellite image and the current shadow reflectivity is close to the average reflectivity of the matched cloud shadows, the shadow is determined to be a cloud shadow. When the overall terrain slope of the satellite image is small and the terrain shadow can be ignored, the cloud coordinate position is calculated by using the cloud shadow. If there is a cloud at the coordinate position and the current shadow reflectivity is close to the average reflectivity of the matched cloud shadows, the shadow is determined to be a cloud shadow. The current shadow area is calculated. If the current shadow area is not less than the set value and the reflectivity is close to the average reflectivity of the matched cloud shadows, the shadow is determined to be a cloud shadow. Otherwise, the current shadow is determined to be a non-cloud shadow.

[0027] After all clouds and shadows have been identified, the cloud-shadow matching of the satellite image as a whole is counted;

[0028] Define the comprehensive cloud shadow matching degree as M image :

[0029]

[0030] Among them, n cloudis the total number of pixels in the cloud, n shadow is the total number of matched cloud shadow pixels; when M image When it is less than 10%, it is considered that the cloud shadow threshold or cloud mask based on cloud amount constraint is wrong, and the cloud shadow mask is re-determined; if the cloud amount is less than 50%, the shadow threshold will be calculated again without using cloud amount constraint, and the corresponding cloud shadow mask will be obtained; otherwise, the cloud shadow mask will be determined based on the set empirical threshold;

[0031] Finally, the surface reflectance in the near-infrared spectrum is used as the guide image to perform guided filtering on the obtained cloud shadow mask to refine and supplement the cloud shadow edge and obtain the final cloud shadow result.

[0032] In the above technical solution, in step 4, for satellite images whose cloud cover does not meet the requirements, iteration is performed only based on the threshold rationality condition.

[0033] In the above technical solution, in step 4, the cloud cover requirement satisfied by the satellite image is: the cloud cover is less than 50%.

[0034] In the above technical solution, in step 4, the value of c is in the range of 0.5-2.

[0035] In the above technical solution, in step 5, the threshold of M is 25%, M s The threshold value is 20%, M e The threshold is 70%.

[0036] In the above technical solution, in step six, the cloud coordinates are outside the satellite image and the current shadow reflectivity is close to the average of the matched cloud shadow reflectivity, which means that the difference is no more than 0.05.

[0037] In the above technical solution, in step six, the overall terrain slope of the satellite image is small and the terrain shadow can be ignored, which means that the mean and standard deviation of the image terrain slope are not greater than 5°.

[0038] In the above technical solution, in step six, if there is a cloud at the coordinate position and the current shadow reflectivity is close to the average of the matched cloud shadow reflectivity, the difference is no more than 0.05.

[0039] In the above technical solution, in step six, the current shadow area is not less than the set value and the reflectivity is close to the average reflectivity of the matched cloud shadows, which means that the current shadow area is not less than 0.25% of the satellite image area and the difference between the reflectivity and the average reflectivity of the cloud shadows is not greater than 0.02.

[0040] The present invention has the following beneficial effects:

[0041] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention helps to solve the problem that the current five-spectral satellite cloud shadow recognition method has low accuracy and is difficult to put into practical application.

[0042] The cloud shadow recognition method of the five-spectral satellite remote sensing image based on cloud amount constraint of the present invention makes full use of the full color spectrum segment image. Compared with most methods based only on multi-spectral spectrum segment, the full color spectrum segment can provide higher resolution and richer details of ground features, which helps to further improve the recognition effect of potential shadow areas.

[0043] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention adds cloud amount information to constrain the threshold iteration of potential shadow areas. The use of cloud amount information can provide a cloud shadow recognition criterion independent of radiation information without increasing the time cost, thereby achieving improved cloud shadow recognition accuracy when radiation information is limited.

[0044] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention is based on a variety of cloud shadow matching evaluation indicators and proposes a flexible cloud shadow matching retention strategy, which largely avoids the problems of misidentification and missed identification caused by a single strategy and effectively improves the recognition accuracy of cloud shadow areas.

[0045] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention adds a secondary matching step for unmatched cloud shadows, which can alleviate the interference of projection deformation problems caused by terrain, three-dimensional cloud structure, etc. and cloud mask errors caused by inaccurate cloud recognition on the algorithm effect to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0047] Figure 1 The present invention is an overall flow chart of the cloud shadow recognition method of five-spectral satellite remote sensing images based on cloud amount constraints.

[0048] Figure 2 The present invention is a flow chart of the water body identification steps of the cloud shadow identification method of five-spectral satellite remote sensing images based on cloud amount constraints.

[0049] Figure 3 This is a cloud shadow recognition effect diagram of the five-spectral satellite remote sensing image cloud shadow recognition method based on cloud amount constraint of the present invention applied to a typical water-containing scene, wherein: Figure 3 (a) is a true color image of a water scene. Figure 3 (b) is the water body identification result. Figure 3 (c) is the reflectance result in the near infrared spectrum. Figure 3(d) is the near-infrared spectral reflectance result after terrain correction. Figure 3 (e) is the potential shadow mask recognition result. Figure 3 (f) is the final cloud and cloud shadow mask results.

[0050] Figure 4 This is a cloud shadow recognition effect diagram of the five-spectral satellite remote sensing image cloud shadow recognition method based on cloud amount constraint of the present invention applied to a typical mountain scene, wherein: Figure 4 (a) is a true color image of a mountain scene. Figure 4 (b) is the water body identification result. Figure 4 (c) is the reflectance result in the near infrared spectrum. Figure 4 (d) is the near-infrared spectral reflectance result after terrain correction. Figure 4 (e) is the potential shadow mask recognition result. Figure 4 (f) is the final cloud and cloud shadow mask results. DETAILED DESCRIPTION

[0051] In order to better illustrate the cloud shadow recognition method of the five-spectrum satellite remote sensing image based on cloud amount constraint of the present invention, cloud shadow recognition was performed on the remote sensing images collected by the "Jilin-1" wide-band 01B satellite (Jilin-1 KF01B) using the present invention. The specific implementation process will be described below with reference to the accompanying drawings.

[0052] The cloud shadow recognition method for five-spectrum satellite remote sensing images based on cloud amount constraints of the present invention uses the official cloud mask product of the wide-band 01B satellite of "Jilin-1" as the true value of the cloud position. If a data set with higher cloud mask accuracy is used for testing, the recognition accuracy of the present invention is expected to be further improved.

[0053] The following is combined with Figure 1-4 The present invention is described in detail.

[0054] Step 1: Convert the DN value of the satellite image into spectral radiance, and then convert the spectral radiance into apparent reflectance; then, based on the 6S model, remove the influence of the atmosphere on radiation transmission, obtain the true reflectance of the surface, and execute step 2;

[0055] Specifically, the DN value of the satellite image is converted into spectral radiance, and then the spectral radiance is converted into apparent reflectance.

[0056] L i =DN·gain i +offset i

[0057]

[0058] Among them, L irepresents the spectral radiance of spectral band i; gain i and offset i is the absolute radiation correction coefficient of spectral band i, DN is the grayscale value of the image product, ρ i is the apparent reflectivity of spectral band i; d is the distance between the sun and the earth; E 0,i is the extra-atmospheric solar spectral irradiance of spectral band i; θ is the solar altitude angle.

[0059] Subsequently, in order to eliminate the influence of the atmosphere on the radiation transfer process, further atmospheric correction is performed based on the 6S model. By inputting the configuration information of the wide-band 01B satellite and the relevant parameters in the image metafile, the 6S model can simulate the influence of the atmosphere on the radiation transfer and remove these influences from the image to obtain the true surface reflectance of each spectral band.

[0060] Step 2: Use reflectivity information to identify water bodies in satellite images, and obtain the location of water bodies in the satellite image area based on the public GSWO water body frequency dataset. Combine the results of the two to obtain a relatively accurate location of water bodies, and then execute step 3;

[0061] Specifically, two different water body identification results are obtained based on the radiation information and GSWO dataset respectively. The water body identification process is as follows: Figure 2 Due to the limitation of the resolution of the GSWO dataset, when the water body is small, identifying the water body based on radiation information often has a better edge recognition effect, such as Figure 3 (b) is the water body case; when the two recognition results are less consistent, the result based on the GSWO dataset has higher confidence.

[0062] Step 3: read the DEM data of the satellite image area, calculate the slope and aspect information of the satellite image area, and perform terrain correction on the near-infrared spectrum where the darkening effect is most obvious based on the semi-empirical SCS+C terrain correction model, and then execute step 4;

[0063] Specifically, according to the geographic information of the satellite image, the DEM data of the satellite image area is read and the slope and aspect information is calculated. Based on the semi-empirical SCS+C model, the terrain correction is performed on the near-infrared spectrum where the darkening effect is most obvious. In this embodiment, the number of sampling points is set to 40,000 according to the size of the wide-band 01B star image, that is, the number of pixels in the cloudless and water-free area is required to be no less than 40,000. Figure 4 (d) shows the terrain correction effect of a mountain scene, compared with the image before correction. Figure 4 (c) By comparison, it can be seen that terrain correction can effectively weaken the influence of terrain shadows. When the number of sampled pixels in the target area is insufficient, the calculated fitting parameters are often not accurate enough. In order to prevent abnormalities in the correction process, terrain correction will not be performed.

[0064] Step 4: Comprehensively consider the full spectrum reflectance, near-infrared spectrum reflectance, cloud mask, and water mask. For images that meet the cloud cover requirements, use the cloud cover as a constraint to perform threshold iteration in the cloudless and waterless areas of the satellite image to find potential shadow areas.

[0065] Specifically, the full spectrum reflectance, near-infrared spectrum reflectance, cloud mask, and water mask are integrated. For images with cloud cover less than 50%, the cloud cover is used as a constraint to perform threshold iteration in the cloudless and waterless areas in the satellite image to find potential shadow areas. The iteration method is:

[0066]

[0067] Among them, P shadow is the theoretical cloud shadow amount. After steps 2 and 3 are executed, the low reflectivity areas caused by water bodies and terrain shadows are largely eliminated. It can be assumed that this variable is approximately equal to the potential shadow amount in the actual iteration, so it is also used to represent the potential shadow amount. p is the total number of pixels that meet the threshold condition after the pth iteration; n0 is the total number of pixels in the cloud-free and water-free area before the iteration; T PAN,p is the threshold value of the full chromatogram segment at the pth iteration; T NIR,p is the near-infrared spectral threshold at the pth iteration; ρ PAN,q is the reflectance of the qth pixel in the full color spectrum, ρ NIR,q is the reflectance of the qth pixel in the near-infrared spectrum; n p,q It is the mark of the qth pixel satisfying the threshold condition after the pth iteration. The iteration termination condition is:

[0068] P cloud / 2<P shadow ×c<P cloud

[0069] Among them, P cloudis the cloud cover; c is the proportional factor, and the value of this constant decreases as the cloud cover increases. In this embodiment, the value of c is in the range of 0.5-2. After each iteration, the threshold rationality judgment is performed, that is, the mean reflectivity of the potential shadow area should be within a reasonable range and less than the overall reflectivity mean of the clean atmosphere area, and the standard deviation of the reflectivity of the cloud shadow area should not be too large. Only when the iteration termination condition and the threshold rationality condition are met, it is determined that a suitable potential shadow threshold has been found; otherwise, the threshold will be updated, and the p-th threshold will be obtained by using the maximum inter-class variance method for the reflectivity of all pixels that meet the threshold requirements after the p-1th iteration, and the iteration will continue. Only the threshold of one spectral segment is updated each time, with the near-infrared spectral segment being given priority. Only when the updated near-infrared spectral segment threshold cannot meet the threshold rationality condition, the full chromatographic segment threshold is updated. If the iteration termination condition and the threshold rationality condition cannot be met through iteration, the threshold is set to a specific empirical value.

[0070] For images with cloud cover between 50% and 90%, the above iterations are performed only based on the threshold rationality condition; when the cloud cover is greater than 90%, the image is almost entirely cloud, and cloud shadow recognition will no longer be performed. The potential shadow area recognition results for water scenes and mountain scenes are shown in Figure 2. Figure 3 (e) and Figure 4 (e) as shown.

[0071] Step 5: Use the zenith angle and azimuth of the satellite sensor, the zenith angle and azimuth of the sun, the relative height and position of the clouds and other information to predict the cloud shadow position of each cloud in the satellite image.

[0072]

[0073] Among them, (x img ,y img ) is the coordinate of the cloud pixel in the image, (x nadir ,y nadir ) is the projection coordinate of the lowest point of the cloud pixel on the ground, (x shadow ,y shadow ) is the coordinate of the corresponding cloud shadow pixel, h c is the distance between the cloud and the ground, θ υ and are the zenith angle and azimuth angle of the satellite sensor, θ s and are the solar zenith angle and azimuth, respectively, and γ is the angle between the true north direction and the image y axis (i.e., image north). For each cloud, the predicted cloud shadow is calculated to be consistent with the potential shadow area determined in step 4. Since the cloud height is unknown, it is necessary to set a certain step size to iterate the cloud height until the predicted cloud shadow can match the potential shadow or stops the iteration after it exceeds the cloud height range.

[0074] In order to evaluate the degree of fit between each cloud shadow and the potential shadow area, three different evaluation indicators are designed, namely, cloud shadow matching degree M, shadow matching degree M s and shadow validity M e :

[0075]

[0076] Among them, n m For all matching pixels of cloud shadows and potential shadows; n o The pixels in the cloud shadow that are beyond the image boundary; n u is the non-cloud pixel in the cloud shadow that has no matching potential shadow; n s All pixels contained in the potential shadow block that matches the cloud shadow. For a potential shadow block, as long as it contains pixels that match the cloud shadow, the shadow block is considered to be matched; n sm n s The shadow in the middle corresponds to the number of shadow pixels outside the image. The cloud pixel position is inferred from the current shadow position and the geometric relationship between the cloud and the cloud shadow; n so n s The number of shadow pixels that can match cloud pixels in .

[0077] For each cloud shadow, after each cloud height iteration, it is necessary to determine whether it meets the cloud shadow matching condition. When M is greater than the set threshold, it is preliminarily determined that it may have matched the potential shadow, and then the shadow matching degree and shadow validity conditions are determined. s and M e All of them are greater than the set threshold, indicating that the matching effect is very good. All potential shadows that are matched are regarded as cloud shadows as a whole, and the iteration ends. s and M e If one of them is smaller than the threshold, it means that the potential shadow area is misidentified or multiple shadows are connected. Only the matched pixels are saved and the iteration ends. s and M e If both are less than the set threshold, it means that there is a problem with the match, and the current result will be abandoned and the iteration will continue. s The threshold value is 20%, M e The threshold is 70%. When all clouds have completed the above iterative matching, execute step 6.

[0078] Step six, identify and supplement the remaining unmatched potential shadows one by one; after all clouds and shadows are identified, the cloud shadow matching of the satellite image as a whole is statistically analyzed; finally, the surface reflectivity of the near-infrared spectrum is used as a guide image to perform guided filtering on the obtained cloud shadow mask to achieve refinement and supplementation of the cloud shadow edge and obtain the final cloud shadow result.

[0079] Specifically, first, the unmatched cloud shadows are re-identified one by one to further improve the cloud shadow recognition rate.

[0080] For wide-width 01B images, the average distance between the matched cloud and cloud shadow and the position of the current shadow are used to back-calculate possible cloud coordinates. When the cloud coordinates are outside the satellite image and the difference between the current shadow reflectivity and the mean cloud shadow reflectivity is no more than 0.05, the shadow is determined to be a cloud shadow. When the mean and standard deviation of the overall terrain slope of the satellite image are no more than 5° and the terrain shadow can be ignored, the cloud coordinate position is back-calculated using the cloud shadow. If there is a cloud at the coordinate position and the difference between the current shadow reflectivity and the mean cloud shadow reflectivity is no more than 0.05, the shadow is determined to be a cloud shadow. The current shadow area is calculated. If the current shadow area is no less than 0.25% of the image area and the difference between the reflectivity and the mean cloud shadow reflectivity is no more than 0.02, the shadow is determined to be a cloud shadow; otherwise, the current shadow is determined to be a non-cloud shadow.

[0081] After all clouds and shadows are identified, the cloud-shadow matching of the entire satellite image is counted.

[0082] Define the comprehensive cloud shadow matching degree as M image :

[0083]

[0084] Among them, n cloud is the total number of pixels in the cloud, n shadow is the total number of cloud shadow pixels matched. image If the cloud amount is less than 10%, the cloud shadow threshold or cloud mask based on cloud amount constraint is considered to be incorrect, and the cloud shadow mask is reconfirmed. If the cloud amount is less than 50%, the shadow threshold will be calculated again without using cloud amount constraint, and the corresponding cloud shadow mask will be obtained; otherwise, the cloud shadow mask will be determined based on the set empirical threshold.

[0085] Finally, the surface reflectance in the near infrared spectrum is used as the guide image to perform guided filtering on the obtained cloud shadow mask to refine and supplement the cloud shadow edge and obtain the final cloud shadow result. The cloud and cloud shadow mask results for water scenes and mountain scenes are shown in Figure 2. Figure 3 (f) and Figure 4 (f) as shown.

[0086] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention provides a cloud shadow recognition method with high accuracy, high recognition efficiency, stability and reliability for five-spectral satellite images.

[0087] The present invention is tested on a cloud shadow dataset consisting of wide-band 01B images of the five-spectral satellite "Jilin-1" as an experimental object. The dataset consists of 40 cloud images with artificially marked cloud shadows, and the dataset covers a variety of surface types such as cities, mountains, deserts, farmlands, and woodlands, with cloud cover ranging from 2% to 82%. The test results show that the cloud shadow recognition method of the five-spectral satellite remote sensing image based on cloud cover constraint of the present invention has an overall recognition accuracy of 92.6%, and the average recognition time of each image is about 12s, which can meet the universality and timeliness requirements of batch data processing; the algorithm has a certain feedback capability, and can provide feedback for situations where the cloud mask or cloud shadow mask is obviously wrong, and has good practicality.

[0088] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention helps to solve the problem that the current five-spectral satellite cloud shadow recognition method has low accuracy and is difficult to put into practical application.

[0089] The cloud shadow recognition method of the five-spectral satellite remote sensing image based on cloud amount constraint of the present invention makes full use of the full color spectrum segment image. Compared with most methods based only on multi-spectral spectrum segment, the full color spectrum segment can provide higher resolution and richer details of ground features, which helps to further improve the recognition effect of potential shadow areas.

[0090] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention adds cloud amount information to constrain the threshold iteration of potential shadow areas. The use of cloud amount information can provide a cloud shadow recognition criterion independent of radiation information without increasing the time cost, thereby achieving improved cloud shadow recognition accuracy when radiation information is limited.

[0091] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention is based on a variety of cloud shadow matching evaluation indicators and proposes a flexible cloud shadow matching retention strategy, which largely avoids the problems of misidentification and missed identification caused by a single strategy and effectively improves the recognition accuracy of cloud shadow areas.

[0092] The cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints of the present invention adds a secondary matching step for unmatched cloud shadows, which can alleviate the interference of projection deformation problems caused by terrain, three-dimensional cloud structure, etc. and cloud mask errors caused by inaccurate cloud recognition on the algorithm effect to a certain extent.

[0093] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A cloud shadow recognition method for five-spectral satellite remote sensing images based on cloud amount constraints, characterized in that: The following steps are involved: Step 1: Convert the DN value of the satellite image into spectral radiance, and then convert the spectral radiance into apparent reflectance; Then, based on the 6S model, the influence of the atmosphere on radiation transmission is removed to obtain the true reflectivity of the surface, and step 2 is performed; Step 2: Use reflectivity information to identify water bodies in satellite images, and obtain the water body locations in the satellite image area based on the GSWO water body frequency dataset. Combine the results of the two to obtain a relatively accurate water body location, and then execute step 3. Step 3: read the DEM data of the satellite image area, calculate the slope and aspect information of the satellite image area, and perform terrain correction on the near-infrared spectrum based on the SCS+C terrain correction model, and then execute step 4; Step 4: Comprehensively consider the full spectrum reflectance, near-infrared spectrum reflectance, cloud mask, and water mask. For satellite images that meet the cloud cover requirements, use cloud cover as a constraint to perform threshold iteration in the cloudless and waterless areas of the satellite image to find potential shadow areas. The iteration method is: Among them, P shadow is the theoretical cloud shadow volume; n p is the total number of pixels that meet the threshold condition after the pth iteration; n0 is the total number of pixels in the cloud-free and water-free area before the iteration; T PAN,p is the threshold value of the full chromatogram segment at the pth iteration; T NIR,p is the near-infrared spectral threshold at the pth iteration; ρ PAN,q is the reflectance of the qth pixel in the full color spectrum, ρ NIR,q is the reflectance of the qth pixel in the near-infrared spectrum; n p,q It is the mark of the qth pixel satisfying the threshold condition after the pth iteration; the iteration termination condition is: P cloud / 2<P shadow ×c<P cloud Among them, P cloud is the cloud cover; c is the scale factor; After each iteration, the rationality of the threshold is judged based on the reflectivity mean and standard deviation of the potential shadow area and the overall reflectivity mean information of the clean atmosphere area. Only when the iteration termination condition and the threshold rationality condition are met, it is determined that a suitable potential shadow threshold has been found; otherwise, the threshold will be updated, and the p-th threshold will be obtained by using the maximum inter-class variance method for the reflectivity of all pixels that meet the threshold requirements after the p-1th iteration, and the iteration will continue; each time, only the threshold of one spectral segment is updated, with the near-infrared spectral segment being given priority, and only when the updated near-infrared spectral segment threshold cannot meet the threshold rationality condition, the full chromatographic segment threshold is updated; if the iteration termination condition and the threshold rationality condition cannot be met through iteration, the threshold is set to a specific empirical value; after confirming the potential shadow area, step five is executed; Step 5: Use the zenith angle and azimuth of the satellite sensor, the zenith angle and azimuth of the sun, and the relative height and position information of the clouds to predict the cloud shadow positions of the clouds in the satellite image one by one; Among them, (x img ,y img ) is the coordinate of the cloud pixel in the image, (x nadir ,y nadir ) is the projection coordinate of the lowest point of the cloud pixel on the ground, (x shadow ,y shadow ) is the coordinate of the corresponding cloud shadow pixel, h c is the distance between the cloud and the ground, θ υ and are the zenith angle and azimuth angle of the satellite sensor, θ s and are the solar zenith angle and azimuth, respectively, and γ is the angle between the true north direction and the y-axis direction of the image; for each cloud, the degree of coincidence between the predicted cloud shadow and the potential shadow area determined in step 4 is calculated; since the cloud height is unknown, it is necessary to set a certain step size to iterate the cloud height until the predicted cloud shadow can match the potential shadow area or stops the iteration after exceeding the cloud height range; The evaluation indicators for evaluating the degree of fit between each cloud shadow and the potential shadow area are: cloud shadow matching degree M, shadow matching degree M s and shadow validity M e : Among them, n m For all matching pixels of cloud shadows and potential shadows; n o The pixels in the cloud shadow that are beyond the image boundary; n u is the non-cloud pixel in the cloud shadow that has no matching potential shadow; n s All pixels contained in the potential shadow block that matches the cloud shadow. For a potential shadow block, as long as it contains pixels that match the cloud shadow, the shadow block is considered to be matched; n sm n s The shadow in the middle corresponds to the number of shadow pixels outside the image. The cloud pixel position is inferred from the current shadow position and the geometric relationship between the cloud and the cloud shadow; n so n s The number of shadow pixels that can match cloud pixels in ; For each cloud shadow, after each cloud height iteration, it is necessary to determine whether it meets the cloud shadow matching condition. When M is greater than the set threshold, it is preliminarily determined that it has matched the potential shadow, and then the shadow matching degree and shadow validity conditions are determined; when M s and M e are all greater than the set threshold, all matching potential shadows are considered as cloud shadows, and the iteration ends; s and M e If one of them is smaller than the threshold, only the matched pixels are saved and the iteration ends; s and M e If all the clouds are smaller than the set threshold, the current result will be abandoned and the iteration will continue; until all the clouds have completed the above iterative matching, the step 6 will be executed; Step 6: Identify and supplement the remaining unmatched potential shadows one by one; The cloud coordinates are calculated by using the average distance between the matched cloud and cloud shadow and the current shadow position. When the cloud coordinates are outside the satellite image and the current shadow reflectivity is close to the average reflectivity of the matched cloud shadows, the shadow is determined to be a cloud shadow. When the overall terrain slope of the satellite image is small and the terrain shadow can be ignored, the cloud coordinate position is calculated by using the cloud shadow. If there is a cloud at the coordinate position and the current shadow reflectivity is close to the average reflectivity of the matched cloud shadows, the shadow is determined to be a cloud shadow. The current shadow area is calculated. If the current shadow area is not less than the set value and the reflectivity is close to the average reflectivity of the matched cloud shadows, the shadow is determined to be a cloud shadow. Otherwise, the current shadow is determined to be a non-cloud shadow. After all clouds and shadows have been identified, the cloud-shadow matching of the satellite image as a whole is counted; Define the comprehensive cloud shadow matching degree as M image : Among them, n cloud is the total number of pixels in the cloud, n shadow is the total number of matched cloud shadow pixels; when M image When it is less than 10%, it is considered that the cloud shadow threshold or cloud mask based on cloud amount constraint is wrong, and the cloud shadow mask is re-determined; if the cloud amount is less than 50%, the shadow threshold will be calculated again without using cloud amount constraint, and the corresponding cloud shadow mask will be obtained; otherwise, the cloud shadow mask will be determined based on the set empirical threshold; Finally, the surface reflectance in the near-infrared spectrum is used as the guide image to perform guided filtering on the obtained cloud shadow mask to refine and supplement the cloud shadow edge and obtain the final cloud shadow result.

2. The cloud shadow recognition method based on five-spectral satellite remote sensing images with cloud amount constraints according to claim 1 is characterized in that: In step 4, for satellite images whose cloud cover does not meet the requirements, iteration is performed only based on the threshold rationality condition.

3. The cloud shadow recognition method based on five-spectral satellite remote sensing images with cloud amount constraints according to claim 1 is characterized in that: In step 4, the cloud cover requirement that the satellite image meets is: cloud cover less than 50%.

4. The cloud shadow recognition method based on five-spectral satellite remote sensing images with cloud amount constraints according to claim 1 is characterized in that: In step 4, the value of c is in the range of 0.5-2.

5. The cloud shadow recognition method based on five-spectral satellite remote sensing images with cloud amount constraints according to claim 1 is characterized in that: In step 5, the threshold of M is 25%, M s The threshold value is 20%, M e The threshold is 70%.

6. The cloud shadow recognition method based on five-spectral satellite remote sensing images with cloud amount constraints according to claim 1 is characterized in that: In step 6, the cloud coordinates are outside the satellite image and the current shadow reflectivity is close to the average of the matched cloud shadow reflectivity, which means that the difference is no more than 0.

05.

7. The cloud shadow recognition method based on five-spectral satellite remote sensing images with cloud amount constraints according to claim 1 is characterized in that: In step six, the overall terrain slope of the satellite image is small and the terrain shadow can be ignored, which means that the mean and standard deviation of the image terrain slope are not greater than 5°.

8. The cloud shadow recognition method based on five-spectral satellite remote sensing images with cloud amount constraints according to claim 1 is characterized in that: In step 6, if there is cloud at the coordinate position and the current shadow reflectivity is close to the average of the matched cloud shadow reflectivity, the difference is no more than 0.

05.

9. The cloud shadow recognition method based on five-spectral satellite remote sensing images with cloud amount constraints according to claim 1 is characterized in that: In step six, the current shadow area is not less than the set value and the reflectivity is close to the average reflectivity of the matched cloud shadows, which means that the current shadow area is not less than 0.25% of the satellite image area and the difference between the reflectivity and the average reflectivity of the cloud shadows is not greater than 0.02.

Images