CN114049566B - A gradually refined method and device for detecting clouds and cloud shadows in Landsat images - Google Patents

Abstract

The invention discloses a method and a device for detecting cloud and cloud shadow of a gradually refined terrestrial satellite image, wherein the method comprises the following steps: extracting bright image elements in a to-be-detected terrestrial satellite image based on the physical properties of the cloud to generate an initial cloud image layer, and gradually refining the initial cloud image layer based on the time and spectral characteristics of the cloud to remove non-cloud bright image elements from the initial cloud image layer and retain real cloud image elements to obtain a cloud detection result; extracting dark image elements in a to-be-detected terrestrial satellite image based on the physical properties of cloud shadows, generating an initial cloud shadow image layer, gradually refining the initial cloud shadow image layer based on time randomness, spectral characteristics, shape characteristics, geometric relations and coexistence relations between clouds and cloud shadows, so as to eliminate the dark image elements of non-cloud shadows from the initial cloud shadow image layer, and reserving real cloud shadow image elements to obtain a cloud shadow detection result. The method can realize accurate automatic detection of the cloud and the cloud shadow in the land satellite data.

Description

A gradually refined method and device for detecting clouds and cloud shadows in Landsat images

technical field

The invention relates to the technical field of remote sensing image cloud and cloud shadow detection, in particular to a gradually refined method and device for land satellite image cloud and cloud shadow detection.

Background technique

Since its launch, Landsat has accumulated a large amount of data for more than 40 years and is one of the most valuable Earth observation data in the world. Due to the characteristics of high spatial resolution, high temporal resolution, free data access, and long-term data continuity, Landsat data is widely used in land use classification, parameter extraction, change detection and other fields. However, as an optical data, Landsat data is inevitably polluted by clouds and cloud shadows, obscuring the original ground objects, and greatly reducing the number and availability of images. In addition, the existence of clouds and cloud shadows will affect the original spectral reflectance information of ground objects, and directly affect the accuracy and reliability of subsequent remote sensing applications based on these data. Therefore, before using Landsat data for analysis and application, the cloud noise in the image must be removed, which is a very important basis in the preprocessing of Landsat imagery. The manual interpretation method is an early method that uses human eye recognition and manual delineation of clouds and cloud shadows. However, the method of manually interpreting images is time-consuming and labor-intensive, and is only suitable for limited areas and a small amount of data. large-scale, long-term remote sensing application requirements. Therefore, an automated cloud and cloud shadow detection method is an effective solution to this problem.

In recent years, researchers have developed many automated algorithms for cloud and cloud shadow detection based on Landsat data. Among them, the most widely used is the FMASK algorithm.

The FMASK algorithm is a widely used method for cloud and cloud shadow detection in Landsat images. First, the FMASK algorithm performs cloud detection based on the physical properties of the cloud by means of direct conditional constraints. This directly constrained detection idea may result in missed detection of cloud pixels, especially thin clouds. Secondly, the FMASK method regards the cloud as a complete object, and starts from the cloud object, constructs a 3D solid model by estimating the cloud height, and performs object-based cloud shadow matching and detection. The previous cloud shadow detection algorithms mostly use the geometric matching method of clouds and cloud shadows, and they all start from cloud pixels to match cloud shadows. But in fact, sometimes there are clouds and not necessarily cloud shadows. If cloud shadows are matched from cloud pixels, each cloud will match the previous cloud shadow regardless of whether there are actual cloud shadows. Due to the inaccuracy of cloud height estimation and the reality of "clouds may not necessarily have cloud shadows", the use of object-based cloud and cloud shadow 3D object shape matching may result in over-detection and mis-detection of cloud shadows.

SUMMARY OF THE INVENTION

The present invention provides a gradually refined method and device for detecting clouds and cloud shadows in land satellite images, so as to solve the technical problems that the prior art easily causes missed detection of cloud pixels and overdetection and wrong detection of cloud shadows.

In order to solve the above-mentioned technical problems, the present invention provides the following technical solutions:

In one aspect, the present invention provides a gradually refined Landsat image cloud and cloud shadow detection method, the gradually refined Landsat image cloud and cloud shadow detection method comprising:

Based on the physical properties of the cloud, the bright pixels in the Landsat image to be detected are extracted to generate an initial cloud layer; wherein, the bright pixels are pixels whose surface reflectivity is greater than a first preset value;

Based on the temporal and spectral features of the cloud, the initial cloud layer is gradually refined, so as to remove the non-cloud bright pixels from the initial cloud layer, and retain the real cloud pixels to obtain a cloud detection result;

Based on the physical properties of cloud shadows, dark pixels in the Landsat image to be detected are extracted to generate an initial cloud shadow layer; wherein, the dark pixels are pixels whose surface reflectivity is less than a second preset value;

Based on the temporal randomness, spectral features, shape features, geometric relationships between clouds and cloud shadows, and the coexistence relationship of "there must be clouds if there are cloud shadows", the initial cloud shadow layer is gradually refined, so that the In the initial cloud shadow layer, the dark pixels with non-cloud shadows are removed, and the real cloud shadow pixels are retained to obtain the cloud shadow detection result.

Further, the bright pixel is a pixel that meets the following conditions in the Landsat image to be detected:

B _Blue >0.05 and B _SWIR1 >0.03

Among them, B _Blue represents the reflectivity of the blue light band of the pixel, and B _SWIR1 represents the reflectivity of the short-wave infrared band of the pixel.

Further, based on the time and spectral features of the cloud, the initial cloud layer is gradually refined to remove non-cloud bright pixels from the initial cloud layer and retain the real cloud pixels, including:

The initial cloud layer is filtered for the first time, and the pixels that meet the following conditions are removed:

B _Blue -0.5*B _Red -0.08<0

Among them, B _Blue represents the blue light band reflectance of the pixel, and B _Red represents the red light band reflectance of the pixel;

Perform a second filter on the results of the first filter to remove cells that meet the following conditions:

BT>23 and NBLI<-0.2 and P _cloud >90%

NBLI=B _Red -B _TIR /B _Red +B _TIR

P _cloud =A _bright /N

Among them, BT represents the surface temperature converted from the brightness temperature, in degrees Celsius, NBLI represents the normalized bare ground index, P _cloud represents the frequency of the current bright pixel, B _TIR represents the thermal infrared band brightness temperature, and A _bright is the image The number of times that the element is judged as a bright target in the initial detection of all images, and N is the total number of all images involved in the calculation;

Perform a third filter on the results of the second filter to remove cells that meet the following conditions:

NDSI>0.8

NDSI=B _Green -B _SWIR1 /B _Green +B _SWIR1

Among them, NDSI represents the normalized snow cover index, B _Green represents the green light band reflectance of the pixel, and B _SWIR1 represents the short-wave infrared band reflectance of the pixel;

Perform a fourth filter on the results of the third filter to remove cells that meet the following conditions:

B _SWIR1 /B _NIR >1.4

Among them, B _NIR represents the near-infrared band reflectance of the pixel;

Perform the fifth filter on the results of the fourth filter, and remove the cells that meet the following conditions:

NDVI>0.8

NDVI=B _NIR -B _Red /B _NIR +B _Red

Among them, NDVI represents the normalized vegetation index.

Further, the dark pixel is a pixel that meets the following conditions in the Landsat image to be detected:

B _Green <0.08 and B _NIR <0.25 and B _SWIR1 <0.11

Among them, B _Green represents the green light band reflectance of the pixel, B _NIR represents the near-infrared band reflectance of the pixel, and B _SWIR1 represents the short-wave infrared band reflectance of the pixel.

Further, based on the temporal randomness, spectral features, shape features, geometric relationships between clouds and cloud shadows, and the coexistence relationship of "there must be clouds if there are cloud shadows", the initial cloud shadow layer is gradually refined. to cull non-cloud-shaded dark cells from the initial cloud-shadow layer and preserve true cloud-shadow cells, including:

The initial cloud shadow layer is filtered for the first time, and the pixels that meet the following conditions are removed:

P _shadow >90%

P _shadow =A _dark /N

Among them, P _shadow represents the frequency of the current dark pixel, A _dark represents the number of times the pixel is judged as a dark target in the initial detection of all images, and N is the total number of all images involved in the calculation;

According to the first filtering result of the initial cloud shadow layer, the coexistence relationship between clouds and cloud shadows of "there must be cloud shadows", as well as the geometric relationship between the sun, clouds and cloud shadows, is based on azimuth and height. The search direction is calculated from the angle, and the search radius is determined based on the constant temperature decrement rate and the temperature quantile calculation of the image. Starting from the current dark pixel, the search for cloud pixels is carried out within the search radius to determine whether there are cloud pixels.

If no cloud pixel is found within the search radius, it is judged that the current dark pixel is not a real cloud shadow pixel, and the current dark pixel is eliminated;

If there is a cloud pixel within the search radius, perform morphological judgment on the current dark pixel and the searched cloud pixel to determine whether the shape indices of the two satisfy the similarity constraint;

If the similarity constraint is not met, the current dark pixel is culled.

On the other hand, the present invention also provides a gradually refined Landsat image cloud and cloud shadow detection device, the gradually refined Landsat image cloud and cloud shadow detection device includes:

The cloud detection module is used to extract the bright pixels in the Landsat image to be detected based on the physical properties of the cloud, and generate an initial cloud layer; based on the time and spectral characteristics of the cloud, the initial cloud layer is gradually refined, The cloud detection result is obtained by removing the non-cloud bright pixels from the initial cloud layer and retaining the real cloud pixels; wherein, the bright pixels are pixels whose surface reflectivity is greater than a first preset value;

The cloud shadow detection module is used to extract dark pixels in Landsat images based on the physical properties of cloud shadows and generate an initial cloud shadow layer; based on the temporal randomness, spectral characteristics, shape characteristics, The geometric relationship and the coexistence relationship of "there must be cloud shadows", the initial cloud shadow layer is gradually refined to remove dark pixels without cloud shadows from the initial cloud shadow layer, and retain The real cloud shadow pixel is used to obtain the cloud shadow detection result; wherein, the dark pixel is the pixel whose surface reflectivity is less than the second preset value.

Further, the bright pixel is a pixel that meets the following conditions in the Landsat image to be detected:

B _Blue >0.05 and B _SWIR1 >0.03

Among them, B _Blue represents the reflectivity of the blue light band of the pixel, and B _SWIR1 represents the reflectivity of the short-wave infrared band of the pixel.

Further, the cloud detection module is specifically used for:

The initial cloud layer is filtered for the first time, and the pixels that meet the following conditions are removed:

B _Blue -0.5*B _Red -0.08<0

Among them, B _Blue represents the blue light band reflectance of the pixel, and B _Red represents the red light band reflectance of the pixel;

Perform a second filter on the results of the first filter to remove cells that meet the following conditions:

BT>23 and NBLI<-0.2 and P _cloud >90%

NBLI=B _Red -B _TIR /B _Red +B _TIR

P _cloud =A _bright /N

Among them, the surface temperature converted from the BT brightness temperature, in degrees Celsius, NBLI represents the normalized bare ground index, P _cloud represents the frequency of the current bright pixel, B _TIR represents the thermal infrared band brightness temperature, and A _bright represents the pixel. The number of times that all images are judged as bright targets in the initial detection of all images, N is the total number of all images involved in the calculation;

Perform a third filter on the results of the second filter to remove cells that meet the following conditions:

NDSI>0.8

NDSI=B _Green -B _SWIR1 /B _Green +B _SWIR1

Among them, NDSI represents the normalized snow cover index, B _Green represents the green light band reflectance of the pixel, and B _SWIR1 represents the short-wave infrared band reflectance of the pixel;

Perform a fourth filter on the results of the third filter to remove cells that meet the following conditions:

B _SWIR1 /B _NIR >1.4

Among them, B _NIR represents the near-infrared band reflectance of the pixel;

Perform the fifth filter on the results of the fourth filter, and remove the cells that meet the following conditions:

NDVI>0.8

NDVI=B _NIR -B _Red /B _NIR +B _Red

Among them, NDVI represents the normalized vegetation index.

Further, the dark pixel is a pixel that meets the following conditions in the Landsat image to be detected:

B _Green <0.08 and B _NIR <0.25 and B _SWIR1 <0.11

Among them, B _Green represents the green light band reflectance of the pixel, B _NIR represents the near-infrared band reflectance of the pixel, and B _SWIR1 represents the short-wave infrared band reflectance of the pixel.

Further, the cloud shadow detection module is specifically used for:

The initial cloud shadow layer is filtered for the first time, and the pixels that meet the following conditions are removed:

P _shadow >90%

P _shadow =A _dark /N

Among them, P _shadow represents the frequency of the current dark pixel, A _dark represents the number of times the pixel is judged as a dark target in the initial detection of all images, and N is the total number of all images involved in the calculation;

According to the first filtering result of the initial cloud shadow layer, the coexistence relationship between clouds and cloud shadows of "there must be cloud shadows", as well as the geometric relationship between the sun, clouds and cloud shadows, is based on azimuth and height. The search direction is calculated from the angle, and the search radius is determined based on the constant temperature decrement rate and the temperature quantile calculation of the image. Starting from the current dark pixel, the search for cloud pixels is carried out within the search radius to determine whether there are cloud pixels.

If no cloud pixel is found within the search radius, it is judged that the current dark pixel is not a real cloud shadow pixel, and the current dark pixel is eliminated;

If there is a cloud pixel within the search radius, perform morphological judgment on the current dark pixel and the searched cloud pixel to determine whether the shape indices of the two satisfy the similarity constraint;

If the similarity constraint is not met, the current dark pixel is culled.

In another aspect, the present invention also provides an electronic device, which includes a processor and a memory; wherein, the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the above method.

In yet another aspect, the present invention also provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, and the instruction is loaded and executed by a processor to implement the above method.

The beneficial effects brought by the technical solution provided by the present invention at least include:

1. The present invention realizes the accurate detection of clouds, especially the detection of thin clouds, and improves the detection effect and accuracy;

2. The present invention avoids the error caused by searching for cloud shadows from clouds, and realizes accurate detection of the range and position of cloud shadows;

3. The present invention realizes accurate detection of clouds and cloud shadows, improves the preprocessing flow of Landsat data, and lays a solid foundation for the subsequent application of Landsat data.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

FIG. 1 is a schematic execution flow diagram of a gradually refined Landsat image cloud and cloud shadow detection method provided by an embodiment of the present invention;

2 is a comparison diagram of the detection results of a gradually refined Landsat image cloud and cloud shadow detection method provided by an embodiment of the present invention and a traditional FMASK algorithm;

FIG. 3 is a comparison diagram of another detection result between the gradually refined Landsat image cloud and cloud shadow detection method provided by the embodiment of the present invention and the traditional FMASK algorithm.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

first embodiment

This embodiment provides a gradually refined method for detecting clouds and cloud shadows in Landsat images. The method may be implemented by an electronic device, and the electronic device may be a terminal or a server. The general idea of this method is to use the idea of gradual refinement, first use a relatively loose threshold constraint, find as many suspected clouds and cloud shadow pixels in the image as possible, and generate an initial layer to avoid possible loss in the initial detection. Then, the initial layer is gradually excluded and refined in combination with the randomness and shape similarity of clouds and cloud shadows, and finally the correct cloud and cloud shadow detection results are determined.

Specifically, as shown in FIG. 1 , the method mainly includes two parts: cloud detection and cloud shadow detection. Among them, the main idea of cloud detection is to find as many bright targets as possible in the image by using a relatively loose threshold constraint according to the physical properties of the cloud, and generate an initial cloud layer. Then, based on the temporal and spatial randomness and low temperature of the cloud, the temporal and spectral characteristics of the cloud are used to gradually exclude and refine the initial cloud layer. Gradually remove non-cloud bright pixels from the initial cloud layer, preserving true cloud pixels. . The detection of cloud shadows is similar to cloud detection. Cloud shadow detection mainly includes two steps: initial detection and refinement detection. First, use the dark features of cloud shadows to detect the initial cloud shadow layer. In the initial detection, use loose constraints to find as many dark objects in the image as possible to avoid missing possible cloud shadow pixels in the initial detection. . Secondly, based on the randomness of cloud shadows in time and space, and the coexistence relationship between clouds and cloud shadows that are similar in shape and "there must be cloud shadows", combined with the temporal randomness of cloud shadows, spectral characteristics, and cloud The shape characteristics, geometric relationship and coexistence relationship between cloud shadows and cloud shadows can gradually eliminate dark pixels that are not cloud shadows in the initial layer, refine the initial cloud shadow layer, and avoid the interference of other dark objects as much as possible. Preserves true cloud shadow cells. The details are as follows:

1. Cloud detection

1.1 Initial cloud detection

In the initial cloud detection stage, the surface reflectance data of the Landsat image is used as input data, and the bright pixels in the image are extracted based on the physical properties of the cloud to generate the initial cloud layer. In the initial detection, relatively loose constraints are adopted, and as many bright targets as possible are found first, so as to avoid missing possible cloud pixels in the initial detection. The principle is: compared with most other objects on the surface, clouds show high reflectivity in almost all bands, so clouds exist in the form of bright targets on the image. In particular, in the spectral band representation of ground objects, clouds have a greater degree of distinction from other clear sky pixels in the blue light band and the short-wave infrared band. Therefore, in this embodiment, the reflectivity of the blue light and the short-wave infrared band is used to extract bright objects in the image, and an initial cloud layer is generated. The formula is as follows:

B _Blue >0.05 and B _SWIR1 >0.03

In the formula, B _Blue represents the reflectance in the blue light band of the pixel, and B _SWIR1 represents the reflectance in the short-wave infrared band of the pixel.

It should be noted that the purpose of initial cloud detection is to first detect as many potential cloud pixels as possible to obtain the initial cloud layer as the basis for subsequent refinement.

1.2 Gradual refinement of cloud layers

Based on the results obtained in 1.1, the initial cloud layer is gradually refined by combining the time and spectral features of the cloud, and the bright pixels that are not clouds are eliminated, and the real cloud pixels are retained to obtain the cloud detection result.

The purpose of this step is: the initial cloud layer obtained in the initial cloud detection contains all the bright objects in the image. In addition to clouds, there are many bright objects that are not clouds, such as construction land, bright vegetation, snow, etc. The existence of these non-cloud bright pixels will interfere with the results of cloud detection. Therefore, it is necessary to perform secondary optimization on the initial cloud layer to remove the bright objects that are not clouds, leaving the real cloud pixels.

1.2.1 The initial cloud layer is filtered for the first time in combination with HOT (Haze optimal transformation) constraints. Many land cover types under clear sky conditions have a good linear relationship in the red-blue spectral space, and are The spectral response of cloud-polluted pixels in the red-blue spectral space is quite different from the clear sky line defined by HOT. Therefore, the HOT index can be used to separate the cloud and clear sky pixels, and eliminate the pixels that meet the following conditions:

B _Blue -0.5*B _Red -0.08<0

In the formula, B _Blue represents the reflectance of the blue light band of the pixel, and B _Red represents the reflectance of the red light band of the pixel.

Purpose: Use this constraint to remove clear sky cells that meet the clear sky constraints from the initial cloud layer.

1.2.2 Clouds have the characteristics of "cold" and appear randomly in time and space, while land objects such as construction land are relatively stable in time series. After filtering in 1.2.1, combined with the constraints of spectrum, temperature and time series probability, those non-cloud bright target objects with relatively high temperature, such as construction land, which are relatively stable in time and space, are removed from the initial cloud layer. culling, that is, culling pixels that meet the following conditions:

BT>23 and NBLI<-0.2 and P _cloud >90%

NBLI=B _Red -B _TIR /B _Red +B _TIR (1)

P _cloud =A _bright /N (2)

In the formula, BT represents the surface temperature converted from the brightness temperature, in degrees Celsius, NBLI represents the normalized bare ground index, and the calculation formula is shown in formula (1). P _cloud represents the frequency of bright pixels, and the calculation formula is shown in formula (2). B _TIR represents the brightness temperature in the thermal infrared band, and B _Red represents the reflectance of the pixel in the red band. A _bright represents the number of times the pixel is judged as a bright target in the initial detection of all images, and N is the total number of all images involved in the calculation. It should be noted that the longer the time series of surface reflectance data is, the more images are involved in the calculation, and the more representative the P _cloud value is. The threshold of P _cloud can be adjusted according to the data situation and the characteristics of the detection area. The larger the value of P _cloud , the stricter the constraints, the less the pixels are eliminated from the initial cloud layer, and the more pixels are eliminated.

Purpose: Clouds have the characteristics of "cold" and show random characteristics in time and space, while land objects such as construction land are relatively stable in time series. Using this constraint, keep as thin clouds as possible while removing warm objects.

1.2.3 After filtering in 1.2.2, combine NDSI to remove snow from the initial cloud layer.

NDSI>0.8

NDSI=B _Green -B _SWIR1 /B _Green +B _SWIR1 (3)

In the formula, NDSI represents the normalized snow cover index, which can effectively distinguish clouds from snow. Its calculation formula is shown in formula (3), B _Green represents the reflectivity of the pixel in the green light band, and B _SWIR1 represents the reflectivity of the pixel in the short-wave infrared band.

Purpose: Use this constraint to remove snow pixels in the initial cloud layer.

1.2.4 After filtering in 1.2.3, non-cloud bright objects that may be confused with clouds include rocks, deserts, etc. The reflectivity of these objects in the short-wave infrared band is higher than that in the near-infrared band, while clouds are the opposite. Therefore, this method removes the pixels that meet the following conditions to further refine the initial cloud layer.

B _SWIR1 /B _NIR >1.4

In the formula, B _SWIR1 represents the reflectivity in the short-wave infrared band of the pixel, and B _NIR represents the reflectivity in the near-infrared band of the pixel.

Purpose: Use this constraint to remove non-cloud bright pixels such as rocks and deserts in the initial cloud layer.

1.2.5 Vegetation with good growth condition also shows relatively bright characteristics on the image, which is easy to be confused with clouds. After filtering in 1.2.4, combined with NDVI, this part of the confused pixels in the initial cloud layer is eliminated. Comparing the spectral characteristics of these bright vegetation and clouds, it can be found that the normalized vegetation index (NDVI) has a good effect in identifying vegetation and distinguishing it from clouds. cloud layer.

NDVI>0.8

NDVI=B _NIR -B _Red /B _NIR +B _Red (4)

In the formula, NDVI represents the normalized vegetation index, and its calculation formula is shown in formula (4). B _Red represents the reflectance in the red band of the pixel, and B _NIR represents the reflectance in the near-infrared band of the pixel.

Purpose: Use this constraint to remove non-cloud bright pixels such as bright vegetation in the initial cloud layer.

2. Cloud shadow detection

Similar to cloud detection, cloud shadow detection is generally based on a step-by-step refinement strategy. First, based on the physical properties of cloud shadows, combined with the dark characteristics of cloud shadows, a relatively loose threshold constraint is used to find as many dark pixels in the image as possible. Then, based on the randomness of cloud shadows in time and space, as well as the coexistence relationship between clouds and cloud shadows that are similar in shape and "there must be cloud shadows", the temporal randomness, spectral characteristics of cloud shadows, and cloud and cloud shadows are used. Constraints such as shape features, geometric relationships, and coexistence relationships between cloud shadows gradually refine the potential cloud shadow layer, and remove dark pixels that are not cloud shadows.

2.1 Initial cloud shadow detection

Using the dark characteristics of cloud shadows, the visible light band is mainly used for constraints, combined with relatively loose threshold constraints, all dark pixels in the image are found first, so as to avoid missing possible cloud shadow pixels in the initial detection. Cloud shadows, water bodies, terrain shadows and other obfuscated pixels have similar spectral performances, but there is a certain degree of discrimination in near-infrared and short-wave infrared. Therefore, the constraints of near-infrared and short-wave infrared bands are added to the initial cloud shadow detection, as follows:

B _Green <0.08 and B _NIR <0.25 and B _SWIR1 <0.11

In the formula, B _Green represents the reflectivity of the pixel in the green light band, B _NIR represents the reflectivity of the pixel in the near-infrared band, and B _SWIR1 represents the reflectivity of the pixel in the short-wave infrared band.

Purpose: While generating the initial cloud shadow layer, initially eliminate the confused pixels such as water bodies and terrain shadows.

2.2 Gradual refinement of cloud shadows

On the basis of the initial cloud shadow layer obtained in 2.1, the dark pixels that are not cloud shadows are gradually eliminated.

Purpose: In the initial cloud shadow layer, all dark objects in the image are included, but the dark objects in the image are not only cloud shadows, but also some confused pixels that may interfere with the cloud shadow detection results, such as terrain shadows, water bodies, etc. . These pixels also appear dark in the image, which may be confused with cloud shadows. Therefore, this part of the non-cloud shadow dark pixels needs to be culled.

2.2.1 Compared with terrain shadows and water bodies, cloud shadows exhibit more randomness in time and space. Therefore, combined with the time series stability, the initial cloud shadow layer is constrained, and the dark pixels that exist stably in the image, such as some terrain shadows, water bodies, etc., are removed. The formula is as follows:

P _shadow >90%

P _shadow =A _dark /N (5)

In the formula, P _shadow represents the frequency of occurrence of dark pixels, and the calculation formula is shown in formula (5). A _dark represents the number of times the pixel is judged as a dark target in the initial detection of all images, and N represents the total number of all images participating in the calculation. It should be noted that, the longer the length of the time series is, the more images are involved in the calculation, and the more representative the P _shadow value is. The threshold of P _shadow can be adjusted according to the data situation and the characteristics of the detection area. The larger the value of P _shadow , the stricter the constraints, the less pixels are eliminated from the initial cloud shadow layer, and the more pixels are eliminated.

Purpose: Cloud shadows show randomness in time and space, while terrain shadows and water bodies are relatively stable. Therefore, using this condition, the non-cloud shadow dark pixels that show stable characteristics in time series, such as terrain shadows and water bodies, are eliminated from the initial cloud shadow layer.

2.2.2 If there are clouds, there may not be cloud shadows, but if there are cloud shadows, there must be corresponding clouds. Using the coexistence relationship between clouds and cloud shadows, "there must be cloud shadows", as well as the geometric relationship between the sun, clouds and cloud shadows, the search direction is calculated based on the azimuth and altitude angles, and the temperature quantile is based on the constant temperature lapse rate and the image. The search radius is determined by numerical calculation. Starting from the cloud shadow, the retrieval of cloud pixels is carried out within the search radius to determine whether there are cloud pixels. If yes, proceed to the next judgment (2.2.3). If no cloud pixel is found within the search radius, it is judged that the shadow is not a real cloud shadow pixel, and it is eliminated.

Purpose: Use the coexistence relationship between clouds and cloud shadows to judge cloud shadows, and eliminate shadows that cannot find the corresponding cloud.

2.2.3 Because clouds and cloud shadows not only have a coexistence relationship, but also have similarities in shape. On the basis of passing the judgment conditions in 2.2.2, the morphological judgment of the cloud shadow and the searched cloud is carried out to judge whether the shape indices of the two satisfy the similarity constraint. If the similarity constraint is satisfied, the searched cloud is considered to be the cloud corresponding to the cloud shadow. If the constraints are not met, the shadow cell is culled to obtain the final cloud shadow layer.

Purpose: Take the morphological index as the constraint, make a similarity judgment on the found clouds and cloud shadows, and remove the shadow pixels that do not satisfy the similarity constraints.

The land satellite image cloud and cloud shadow detection method of the present invention has been described in detail above. The progress of the method of the present invention is described below by comparing the detection results of the method of the present invention with the existing detection method (FMASK algorithm). The comparison between the detection results of the gradually refined Landsat image cloud and cloud shadow detection method provided by the present invention and the traditional FMASK algorithm is shown in FIG. 2 and FIG. 3 . The first column in the figure is the detection result of the method of the present invention, the second column is the original image, which is synthesized by 4/3/2, and the third column is the detection result of the existing FMASK algorithm. In the detection result graph, white represents clouds, gray represents cloud shadows, and crosshairs represent the same spatial position. As can be seen from the figure, the detection method provided by the present invention can realize accurate automatic detection of clouds and cloud shadows in Landsat data, and is superior to the FMASK algorithm.

To sum up, the method of this embodiment is different from the direct constraint method of the traditional algorithm. It detects clouds and cloud shadows based on the idea of gradual refinement. In the initial detection, relatively loose constraints and threshold conditions are used to first detect as many as possible. Cloud/cloud shadow pixels are detected into the initial layer to avoid losing possible cloud and cloud shadow pixels in the initial detection; then combine the cloud/cloud shadow spectrum, randomness in time and space, shape similarity, etc. Features that accurately determine true clouds and cloud shadows by culling out not true cloud/cloud shadow cells by progressively excluding and refining constraints. In addition, this embodiment is different from the method of finding cloud shadows from clouds in traditional algorithms, and utilizes the coexistence relationship between clouds and cloud shadows, and the shape similarity between clouds and cloud shadows. A method is proposed to search for clouds from cloud shadows and to use the similarity constraints of morphological indices between clouds and cloud shadows. It avoids the error of "there is not necessarily a shadow if there is a cloud" caused by the way of looking for cloud shadow from the cloud used by the traditional algorithm, and the error caused by directly matching the cloud shadow by estimating the cloud height. Thereby, the effect and accuracy of cloud and cloud shadow detection are improved.

Second Embodiment

This embodiment provides a gradually refined Landsat image cloud and cloud shadow detection device, the gradually refined Landsat image cloud and cloud shadow detection device includes the following modules:

The cloud detection module is used to extract as many bright pixels as possible in the Landsat image to be detected based on the physical properties of the cloud and use a relatively loose threshold to generate an initial cloud layer; based on the time and spectral characteristics of the cloud, the The initial cloud layer is gradually refined to remove non-cloud bright pixels from the initial cloud layer, and retain the real cloud pixels to obtain a cloud detection result; wherein, the bright pixels are those whose surface reflectivity is greater than the first a pixel with a preset value;

The cloud shadow detection module is used to extract as many dark pixels as possible in the Landsat image based on the physical properties of cloud shadows, using a relatively loose threshold to generate an initial cloud shadow layer; based on the random time between clouds and cloud shadows The initial cloud shadow layer is gradually refined to eliminate non-clouds from the initial cloud shadow layer. The dark pixel of the shadow is retained, and the real cloud shadow pixel is retained to obtain the cloud shadow detection result; wherein, the dark pixel is the pixel whose surface reflectivity is less than the second preset value.

The gradually refined Landsat image cloud and cloud shadow detection apparatus of this embodiment corresponds to the gradually refined Landsat image cloud and cloud shadow detection method of the first embodiment; wherein, the present Landsat image cloud and cloud shadow The functions implemented by the functional modules in the detection device correspond one-to-one with the flow steps in the above-mentioned first embodiment of the Landsat image cloud and cloud shadow detection method; therefore, details are not repeated here.

Third Embodiment

This embodiment provides an electronic device, which includes a processor and a memory; wherein, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the method of the first embodiment.

The electronic device may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPU) and one or more memories, wherein the memory stores at least one instruction, so The instructions are loaded by the processor and execute the above method.

Fourth Embodiment

This embodiment provides a computer-readable storage medium, where at least one instruction is stored, and the instruction is loaded and executed by a processor to implement the method of the first embodiment. Wherein, the computer-readable storage medium may be ROM, random access memory, CD-ROM, magnetic tape, floppy disk, optical data storage device, and the like. The instructions stored therein can be loaded by the processor in the terminal and execute the above method.

Furthermore, it should be noted that the present invention may be provided as a method, an apparatus or a computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied therein.

Embodiments of the present invention are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, embedded processor or other programmable data processing terminal to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing terminal produce Means implementing the functions specified in one or more of the flowcharts and/or one or more blocks of the block diagrams.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal equipment to operate in a particular manner, such that the instructions stored in the computer readable memory result in an article of manufacture comprising instruction means, the The instruction means implement the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams. These computer program instructions can also be loaded on a computer or other programmable data processing terminal equipment, so that a series of operational steps are performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby executing on the computer or other programmable terminal equipment The instructions executed on the above provide steps for implementing the functions specified in the flowchart or blocks and/or the block or blocks of the block diagrams.

It should also be noted that in this document, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply those entities or operations There is no such actual relationship or order between them. The terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or terminal that includes a list of elements includes not only those elements, but also not expressly listed Other elements, or elements that are inherent to such a process, method, article or end device. Without further limitation, an element defined by the phrase "comprises a..." does not preclude the presence of additional identical elements in the process, method, article or terminal device comprising said element.

Finally, it should be noted that the above are the preferred embodiments of the present invention. It should be pointed out that although the preferred embodiments of the present invention have been described, for those skilled in the art, once the basic inventive concept of the present invention is known , without departing from the principles of the present invention, several improvements and modifications can also be made, and these improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims (2)

1. A method for detecting cloud and cloud shadow of a terrestrial satellite image which is gradually refined is characterized by comprising the following steps:

extracting bright image elements in the terrestrial satellite image to be detected based on the physical properties of the cloud, and generating an initial cloud layer; the bright pixel is a pixel with the surface reflectivity greater than a first preset value;

based on the time and spectral characteristics of the cloud, gradually thinning the initial cloud image layer to remove non-cloud bright image elements from the initial cloud image layer, and reserving real cloud image elements to obtain a cloud detection result;

extracting dark pixels in the terrestrial satellite images to be detected based on the physical properties of the cloud shadow to generate an initial cloud shadow image layer; the dark pixel is a pixel with the surface reflectivity smaller than a second preset value;

based on time randomness, spectral characteristics, shape characteristics, geometric relationships and coexistence relationships of cloud shadow and certain cloud shadow, the initial cloud shadow image layer is gradually refined to remove dark image elements of non-cloud shadow from the initial cloud shadow image layer, and real cloud shadow image elements are reserved to obtain a cloud shadow detection result;

the bright pixels are pixels which meet the following conditions in the terrestrial satellite image to be detected:

B_Blue>0.05 and B_SWIR1>0.03

wherein, B_BlueRepresenting the blue band reflectivity of the pixel, B_SWIR1The short wave infrared band reflectivity of the pixel is represented;

the cloud-based time and spectrum feature gradually refines the initial cloud image layer to eliminate non-cloud bright image elements from the initial cloud image layer and retain real cloud image elements, and the method comprises the following steps:

and filtering the initial cloud layer for the first time, and eliminating pixels meeting the following conditions:

B_Blue-0.5*B_Red-0.08<0

wherein, B_BlueRepresenting the blue band reflectivity of the pixel, B_RedRepresenting the reflectivity of the red light wave band of the pixel;

and (3) carrying out secondary filtration on the primary filtration result, and eliminating pixels meeting the following conditions:

BT>23 and NBLI<-0.2 and P_cloud>90％

NBLI＝B_Red-B_TIR/B_Red+B_TIR

P_cloud＝A_bright/N

wherein BT represents the earth surface temperature obtained by converting brightness temperature, NBLI represents normalized bare earth index, P represents the normalized bare earth index_cloudIndicating the frequency of occurrence of the current bright pixel, B_TIRRepresenting the thermal infrared band brightness temperature, A_brightThe number of times that the pixel is judged as a bright target in all image initial detection is shown, and N is the total number of all images participating in calculation;

and (3) carrying out third filtering on the second filtering result, and eliminating pixels meeting the following conditions:

NDSI>0.8

NDSI＝B_Green-B_SWIR1/B_Green+B_SWIR1

where NDSI represents the normalized snow index, B_GreenRepresenting the green band reflectivity of the pixel, B_SWIR1The short wave infrared band reflectivity of the pixel is represented;

and filtering the third filtering result for the fourth time, and eliminating pixels meeting the following conditions:

B_SWIR1/B_NIR>1.4

wherein, B_NIRRepresenting the near-infrared band reflectivity of the pixel;

and (4) fifth filtering is carried out on the fourth filtering result, and pixels meeting the following conditions are eliminated:

NDVI>0.8

NDVI＝B_NIR-B_Red/B_NIR+B_Red

wherein NDVI represents a normalized vegetation index;

the dark pixels are pixels meeting the following conditions in the terrestrial satellite image to be detected:

B_Green<0.08 and B_NIR<0.25 and B_SWIR1<0.11

wherein, B_GreenRepresenting the green band reflectivity of the pixel, B_NIRRepresenting the near infrared band reflectivity, B, of the pixel_SWIR1The short wave infrared band reflectivity of the pixel is represented;

based on the time randomness, the spectral characteristics, the shape characteristics, the geometric relations and the coexistence relation of cloud shadow and certain cloud, the initial cloud shadow image layer is gradually refined to eliminate the dark image elements of non-cloud shadow from the initial cloud shadow image layer and keep the real cloud shadow image elements, and the method comprises the following steps:

carrying out primary filtering on the initial cloud shadow image layer, and eliminating pixels meeting the following conditions:

P_shadow>90％

P_shadow＝A_dark/N

wherein, P_shadowIndicating the frequency of occurrence of the current dark pixel, A_darkRepresenting the times of judging the pixel as a dark target in all image initial detection, wherein N is the total number of all images participating in calculation;

according to the first filtering result of the initial cloud shadow image layer, calculating a searching direction based on an azimuth angle and an altitude angle by utilizing the coexistence relation of cloud and cloud shadow, namely that cloud shadow and cloud shadow must exist, and the geometric relation of sun, cloud and cloud shadow, calculating and determining a searching radius based on a constant temperature decrement rate and the temperature quantile of an image, starting from the current dark pixel, retrieving the cloud pixel in the searching radius, and judging whether the cloud pixel exists or not;

if the cloud pixel is not found in the searching radius range, judging that the current dark pixel is not a real cloud shadow pixel, and removing the current dark pixel;

if the cloud pixel exists in the search radius range, performing morphological judgment on the current dark pixel and the searched cloud pixel, and judging whether the shape indexes of the current dark pixel and the searched cloud pixel meet the similarity constraint;

and if the similarity constraint condition is not satisfied, removing the current dark pixel.

2. A terrestrial satellite image cloud and cloud shadow detection device that refines gradually, characterized by, includes:

the cloud detection module is used for extracting bright image elements in the terrestrial satellite images to be detected based on the physical properties of the cloud and generating an initial cloud image layer; based on the time and spectral characteristics of the cloud, gradually thinning the initial cloud image layer to remove non-cloud bright image elements from the initial cloud image layer, and reserving real cloud image elements to obtain a cloud detection result; the bright pixel is a pixel with the surface reflectivity greater than a first preset value;

the cloud shadow detection module is used for extracting dark image elements in the terrestrial satellite images based on the physical properties of cloud shadows and generating an initial cloud shadow image layer; based on time randomness, spectral characteristics, shape characteristics, geometric relationships and coexistence relationships of cloud shadow and certain cloud shadow, the initial cloud shadow image layer is gradually refined to remove dark image elements of non-cloud shadow from the initial cloud shadow image layer, and real cloud shadow image elements are reserved to obtain a cloud shadow detection result; the dark pixel is a pixel with the surface reflectivity smaller than a second preset value;

the bright pixels are pixels which meet the following conditions in the terrestrial satellite image to be detected:

B_Blue>0.05 and B_SWIR1>0.03

wherein, B_BlueRepresenting the blue band reflectivity of the pixel, B_SWIR1The short wave infrared band reflectivity of the pixel is represented;

the cloud detection module is specifically configured to:

and filtering the initial cloud layer for the first time, and eliminating pixels meeting the following conditions:

B_Blue-0.5*B_Red-0.08<0

wherein, B_BlueRepresenting the blue band reflectivity of the pixel, B_RedRepresenting the reflectivity of the red light wave band of the pixel;

and (3) carrying out secondary filtration on the primary filtration result, and eliminating pixels meeting the following conditions:

BT>23 and NBLI<-0.2 and P_cloud>90％

NBLI＝B_Red-B_TIR/B_Red+B_TIR

P_cloud＝A_bright/N

wherein BT represents the earth surface temperature obtained by converting brightness temperature, NBLI represents normalized bare earth index, P represents the normalized bare earth index_cloudIndicating the frequency of occurrence of the current bright pixel, B_TIRRepresenting the thermal infrared band brightness temperature, A_brightThe number of times that the pixel is judged as a bright target in all the initial image detection is shown, and N is the total number of all the images participating in the calculation;

and (3) carrying out third filtering on the second filtering result, and eliminating pixels meeting the following conditions:

NDSI>0.8

NDSI＝B_Green-B_SWIR1/B_Green+B_SWIR1

where NDSI represents the normalized snow index, B_GreenRepresenting the green band reflectivity of the pixel, B_SWIR1The short wave infrared band reflectivity of the pixel is represented;

and filtering the third filtering result for the fourth time, and eliminating pixels meeting the following conditions:

B_SWIR1/B_NIR>1.4

wherein, B_NIRRepresenting the near-infrared band reflectivity of the pixel;

and (4) fifth filtering is carried out on the fourth filtering result, and pixels meeting the following conditions are eliminated:

NDVI>0.8

NDVI＝B_NIR-B_Red/B_NIR+B_Red

wherein NDVI represents a normalized vegetation index;

the dark pixels are pixels meeting the following conditions in the terrestrial satellite image to be detected:

B_Green<0.08 and B_NIR<0.25 and B_SWIR1<0.11

wherein, B_GreenRepresenting the green band reflectivity of the pixel, B_NIRRepresenting the near infrared band reflectivity, B, of the pixel_SWIR1The short wave infrared band reflectivity of the pixel is represented;

the cloud shadow detection module is specifically configured to:

carrying out primary filtering on the initial cloud shadow image layer, and eliminating pixels meeting the following conditions:

P_shadow>90％

P_shadow＝A_dark/N

wherein, P_shadowIndicating the frequency of occurrence of the current dark pixel, A_darkRepresenting the times of judging the pixel as a dark target in all image initial detection, wherein N is the total number of all images participating in calculation;

according to the first filtering result of the initial cloud shadow image layer, calculating a searching direction based on an azimuth angle and an altitude angle by utilizing the coexistence relation of cloud and cloud shadow, namely that cloud shadow and cloud shadow must exist, and the geometric relation of sun, cloud and cloud shadow, calculating and determining a searching radius based on a constant temperature decrement rate and the temperature quantile of an image, starting from the current dark pixel, retrieving the cloud pixel in the searching radius, and judging whether the cloud pixel exists or not;

if the cloud pixel is not found in the search radius range, judging that the current dark pixel is not a real cloud shadow pixel, and removing the current dark pixel;

if the cloud pixel exists in the search radius range, performing morphological judgment on the current dark pixel and the searched cloud pixel, and judging whether the shape indexes of the current dark pixel and the searched cloud pixel meet the similarity constraint;

and if the similarity constraint condition is not satisfied, removing the current dark pixel.

Images