Remote sensing cloud detection method considering cloud information representation index and geometric morphology characteristics
By designing a remote sensing cloud detection method for cloud information characterization index and geometric morphological characteristics, the problems of cloud detection complexity and efficiency in the existing technology are solved, and efficient and accurate cloud recognition is achieved, suitable for the quality of high-resolution images Inspection and image retrieval.
Patent Information
- Application Number
- CN202310046948.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-01-31
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2043-01-31
AI Technical Summary
The existing cloud detection method has complex implementation processes and poor detection efficiency and effectiveness, especially in high-resolution remote sensing images, which makes cloud detection difficult to accurately identify, resulting in occlusion of ground object information and affecting the quality and application effect of remote sensing data.
Using a remote sensing cloud detection method that takes into account cloud information characterization index and geometric morphological characteristics, the cloud body is extracted and purified by designing cloud information characterization index, adaptive threshold segmentation, geometric feature filtering and morphological filtering, and finally converted into vector results for cloud volume. statistics.
It realizes efficient and accurate cloud detection, with no sharp corners or serrated edges, no holes inside, simple detection process, strong generalization and practicality, and supports high-resolution image quality inspection and cloud-free image replenishment.
Smart Images

Figure CN117315455B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of remote sensing image cloud object detection, and in particular to a remote sensing cloud detection method taking into account cloud information representation index and geometric morphological features. Background Art
[0002] With the successful launch of the Gaofen special series of satellites, my country has entered the era of remote sensing big data, and the quality and quantity of remote sensing images have increased dramatically. However, the southwestern mountainous areas are located in the subtropical humid climate zone, with high precipitation, high humidity, and sufficient water vapor. Affected by the terrain, a large amount of moisture is gathered above the basin or valley, and the mountains block it, making it difficult to convect with the outside air, resulting in high-resolution remote sensing images in the southwestern mountainous areas often having more clouds. Due to the occlusion of ground object information by clouds, the cloud areas in high-resolution remote sensing images cannot obtain correct information about ground targets, directly or indirectly inhibiting remote sensing image pattern recognition, ground object classification, surface remote sensing parameter inversion, ground object change detection, and time series dynamic change analysis. In practical applications, the collected images first need to be checked for quality, among which cloud coverage is an important quality indicator. Therefore, rapid identification of clouds and cloud volume statistics can support the quality inspection of original images. In addition, in the image production process, images of cloud-covered areas need to be re-collected. By first identifying the clouds, then removing the cloud images, and re-collecting the removed areas, the collected cloud-free images are used to fill in, and finally a usable cloud-free image is formed. Therefore, cloud detection in high-resolution remote sensing images is one of the important steps in remote sensing data processing.
[0003] At present, cloud detection methods are mainly divided into three categories: threshold method, statistical learning method, and deep learning method. The threshold method has been widely used because of its simple principle and fast operation speed. It is the most widely used method among the many cloud detection methods of remote sensing images. However, the determination of the optimal threshold is a major difficulty. In addition, the threshold method is based on pixel segmentation of clouds, ignoring the geometric morphology and other characteristics of cloud clusters. The cloud detection method based on statistical learning is to achieve cloud detection through sample data, using pattern recognition algorithms such as naive Bayes and support vector machines. However, this method requires the design of effective cloud information features first, and then supervised classification. The implementation process is relatively complicated and inefficient. Deep learning is a technology that has emerged in recent years with the development of big data and computer computing power. It can establish cloud recognition models in a black box manner from a large number of sample sets. However, deep learning currently relies heavily on a large number of high-precision samples, and labeling samples requires a huge amount of manpower and time costs. In addition, since deep learning cloud detection is pixel-level semantic segmentation, the results are often of poor geometric shape, with problems such as edge jagged edges, broken spots, and cloud holes. Multiple post-processing is required to meet practical applications. In addition, the generalization and universality of deep learning models are low, and there are still deficiencies in practical applications over a large range, across regions, and in multiple scenarios.
[0004] It can be seen that the existing cloud detection methods have the problems of complex implementation process and poor detection efficiency and effect. Summary of the invention
[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a remote sensing cloud detection method that takes into account cloud information characterization index and geometric morphological characteristics, so as to solve the problems of the cloud detection method in the prior art having a complex implementation process and poor detection efficiency and effect.
[0006] The present invention provides a remote sensing cloud detection method taking into account cloud information representation index and geometric morphological characteristics, comprising the following steps:
[0007] Step 1, designing a cloud information characterization index based on a haze optimization transformation index, texture contrast, and Canny gradient obtained from high-resolution remote sensing images, wherein the haze optimization transformation index, texture contrast, and Canny gradient are all normalized and stretched;
[0008] Step 2, using the Otsu algorithm to perform adaptive threshold segmentation on the cloud information representation index to obtain a binary image including the background and the cloud target;
[0009] Step 3, filtering non-cloud objects and purifying cloud bodies in the binary image using geometric features;
[0010] Step 4, designing a morphological cloud filter operator DMCO, using morphological filtering to filter the binary image again, and further purifying it to obtain a cloud binary image;
[0011] Step 5, performing morphological closing operation on the cloud binary image, filling holes, outputting a binary grid image, and converting the binary grid image into a vector, deleting the background, and obtaining a cloud detection vector result.
[0012] Optionally, the cloud information characterization index is designed based on the haze optimization transformation index, texture contrast and Canny gradient obtained from high-resolution remote sensing images, including:
[0013] The cloud information characterization index design formula is:
[0014] CRI=HOT-Con-G
[0015] Among them, the CRI is the cloud information characterization index, the HOT is the haze optimization transformation index, Con is the texture contrast, and G is the Canny gradient.
[0016] Optionally, the step of using the Otsu algorithm to perform adaptive threshold segmentation on the cloud information representation index comprises the following steps:
[0017] The total inter-class variance g of the cloud and background in the cloud information representation index is calculated as follows:
[0018] g=w0*(u0-u) 2 +w1*(u1-u) 2
[0019] Among them, let the initial segmentation threshold be t, w0 is the proportion of cloud pixels in the image, u0 is the average grayscale, w1 is the proportion of background pixels in the image, u1 is the average grayscale, u is the total average grayscale of the image, u=w0*u0+w1*u1;
[0020] By iteratively calculating the total inter-class variance g, find the segmentation threshold that maximizes the inter-class variance g, set to t max , the maximum segmentation threshold t max The optimal threshold is set, and the binary image is segmented according to the optimal threshold.
[0021] Optionally, the geometric features include:
[0022] The geometric features include at least the area, compactness and aspect ratio of the pattern in the binary image, and the straight line and right angle features of the grayscale image.
[0023] Optionally, the area, compactness and aspect ratio of the image spots, and the straight line and right angle features of the grayscale image include:
[0024] The area of the pattern is set to A, A<100 square meters, the aspect ratio of the pattern is set to B, B>50, the compactness of the pattern is set to C, C<0.0001, and the calculation formula of the aspect ratio B of the pattern is:
[0025]
[0026] Wherein, length and width are the length and width of the minimum circumscribed rectangle of the patch, respectively. The calculation formula of the compactness C of the patch is:
[0027]
[0028] Among them, Area and Perimeter are the area and perimeter of the patch respectively.
[0029] Since artificial objects such as buildings and roads have obvious geometric features such as straight lines and right angles, non-cloud objects such as artificial objects can be filtered out by extracting the straight line and right angle features of the grayscale image. The straight line and right angle feature extraction process is as follows:
[0030] The LSD line detection algorithm first performs Gaussian downsampling on the grayscale image, and then calculates the grayscale gradient and level-line of the local boundary of the image. The image grayscale gradient is calculated on a 2×2 template according to the following formula:
[0031]
[0032] The angle of Level-Line is calculated according to the above formula:
[0033]
[0034] Where i(x,y) is the grayscale value of the pixel x,y on the grayscale image, g x (x,y) is the gradient value in the X direction, g y (x, y) is the gradient value in the Y direction, and G(x, y) is the grayscale gradient amplitude at the pixel point (x, y).
[0035] Then, the pixel gradient values are sorted and the straight line area is grown, and the rectangle fitting the target area is further determined by calculating the false alarm index, so as to obtain the straight lines in the target image, and further extract the right angles through the distance and angle relationship between the straight lines.
[0036] Optionally, the designing of a morphological cloud filter operator DMCO, using morphological filtering to filter the binary image again, and further purifying to obtain a cloud binary image, includes:
[0037] Based on the high-resolution remote sensing image, the inter-band mean is taken to obtain a single-band grayscale image, and the grayscale image is processed using the designed morphological cloud filter operator DMCO operator to extract non-cloud objects in the binary image, and then the spots in the binary image that intersect with the spots of non-cloud objects are deleted.
[0038] Optionally, the designing of a morphological cloud filter operator DMCO, using morphological filtering to filter the binary image again to further purify the binary image of the cloud body, further includes:
[0039] After processing the grayscale image, the morphological cloud filter operator DMCO extracts non-cloud objects with DMCO>Et by setting a threshold Et=0. The calculation formula of DMCO is as follows:
[0040]
[0041] D N and S N Respectively represent the number of directions and sizes of the structural elements, Indicates that the structural element direction is d i and size s j The differential morphological characteristics of
[0042]
[0043] in, is the morphological white hat reconstruction, G is the grayscale image, This is the morphological opening operation.
[0044] Optionally, the morphological closing operation includes:
[0045] The morphological closing operation formula is:
[0046]
[0047] Among them, · is the morphological closing operation, is the morphological dilation, ⊙ is the morphological erosion, B is the circular structure element, B=strel('ELLIPSE',s,s), s is the size.
[0048] The remarkable effects of the present invention are: designing a cloud information characterization index through high-resolution remote sensing images, preliminarily extracting cloud bodies through adaptive threshold segmentation, filtering non-cloud objects and purifying cloud bodies through geometric features, designing morphological cloud filtering operators to further purify cloud bodies, and finally realizing detection by converting into vectors and counting cloud amounts. Compared with the dependence of machine learning and deep learning cloud detection methods on sample data, the present invention does not need to rely on manually labeled samples, but only uses the color brightness and geometric morphological characteristics of clouds. The detection process is simple, and the detection results have significant geometric morphological advantages. The edges of clouds have no sharp corners or jagged edges, and there are no holes in the clouds. It can provide support for production processes such as quality inspection of high-resolution images, screening of cloud-free images, and image acquisition and filling of cloud-covered areas to generate cloud-free images, and has strong generalization and practicality. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 It is a schematic diagram of the overall detection process of the present invention;
[0050] Figure 2 It is a schematic diagram of the detection process of the present invention;
[0051] Figure 3 is a high-resolution remote sensing image used in the examples of the present invention;
[0052] Figure 4 It is the Otsu threshold segmentation result of the high-resolution remote sensing image of the present invention;
[0053] Figure 5 It is the result of extracting non-cloud objects from the high-resolution remote sensing image DMCO of the present invention;
[0054] Figure 6 This is a high-resolution remote sensing image cloud detection result diagram of the present invention. DETAILED DESCRIPTION
[0055] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. The functional units with the same labels in the examples of the present invention have the same and similar structures and functions.
[0056] See also Figure 1 and Figure 2 The present invention provides a remote sensing cloud detection method taking into account cloud information representation index and geometric morphological characteristics, comprising the following steps:
[0057] Step 1, designing a cloud information characterization index based on a haze optimization transformation index, texture contrast, and Canny gradient obtained from high-resolution remote sensing images, wherein the haze optimization transformation index, texture contrast, and Canny gradient are all normalized and stretched;
[0058] Step 2, using the Otsu algorithm to perform adaptive threshold segmentation on the cloud information representation index to obtain a binary image including the background and cloud targets;
[0059] Step 3, filtering non-cloud objects and purifying cloud bodies in the binary image using geometric features;
[0060] Step 4, designing a morphological cloud filter operator DMCO, using morphological filtering to filter the binary image again, and further purifying it to obtain a cloud binary image;
[0061] Step 5, performing morphological closing operation on the cloud binary image, filling holes, outputting a binary grid image, and converting the binary grid image into a vector, deleting the background, and obtaining a cloud detection vector result.
[0062] In this embodiment, remote sensing images with multiple sensors and multiple resolutions are used to accurately identify clouds, and the detection results have significant geometric advantages. The edges of the clouds have no sharp corners or jagged edges, and there are no holes inside the clouds. This can provide support for quality inspection of high-resolution images, screening of cloud-free images, and image acquisition and filling of cloud-covered areas to generate cloud-free images. A cloud information characterization index is designed based on the parameters obtained from the remote sensing images, and adaptive threshold segmentation is used to preliminarily extract cloud bodies, geometric feature filtering of non-cloud objects and purification of cloud bodies, and morphological operations are used to further purify cloud bodies. Finally, detection is achieved by converting them into vectors and counting cloud amounts. Compared with the reliance on sample data by machine learning and deep learning cloud detection methods, the present invention has less manual participation, a high degree of automation, and detection results with significant cloud geometric advantages. Only the color brightness and geometric features of clouds are used, and the detection process is simple, with strong generalization and practicality.
[0063] Optionally, the cloud information characterization index is designed based on the haze optimization transformation index, texture contrast and Canny gradient obtained from high-resolution remote sensing images, including:
[0064] The cloud information characterization index design formula is:
[0065] CRI=HOT-Con-G
[0066] Among them, the CRI is the cloud information characterization index, HOT is the haze optimization transformation index, Con is the texture contrast, and G is the Canny gradient.
[0067] See also Figure 3 In this embodiment, two images with a resolution of 0.5 meters and two images with a resolution of 2 meters are used as experimental data. The images with a resolution of 0.5 meters are from the Gaojing-1 satellite and the GeoEye-1 satellite, respectively. The imaging time is July 20 and June 5, respectively. The coverage area is 166 square kilometers and 167 square kilometers. The cloud layer on the image is clustered, with thick clouds and thin clouds. The images with a resolution of 2 meters are from the Ziyuan-3 satellite and the Gaofen-1C satellite, respectively. The imaging time is September 12 and July 13, respectively. The coverage area is 2550 square kilometers and 4488 square kilometers. The cloud layer on the image is scaly cirrocumulus, mainly thick clouds. The four experimental images are numbered (a), (b), (c), and (d), representing the Gaojing-1 data, GeoEye-1 data, Ziyuan-3 data, and Gaofen-1C data, respectively.
[0068] Furthermore, the haze optimization transformation index HOT is calculated based on the blue band and red band of the high-resolution remote sensing image, and the calculation formula is:
[0069]
[0070] Among them, Bblue For the blue band, B red It is the red band.
[0071] Further, the texture contrast Con is calculated based on high-resolution remote sensing images using a gray level co-occurrence matrix (GLCM), specifically;
[0072] The high-resolution remote sensing image is converted into a single-band grayscale image by taking the inter-band mean value. The grayscale level is compressed to L (L=64). The w*w neighborhood of each pixel is used as a window (w=7) to calculate the GLCM of four directions (0°, 45°, 90° and 135°) and one pixel interval. Four GLCMs are obtained for each neighborhood window. The GLCM is a matrix with L rows and L columns.
[0073] The four contrast values calculated based on the four GLCMs are averaged as the texture contrast feature value of the central pixel of the neighborhood window. The sliding window traverses all pixels to obtain the texture feature image.
[0074]
[0075] Where P(i,j) is the value of the i-th row and j-th column in the GLCM.
[0076] Furthermore, based on high-resolution remote sensing images, the process of calculating the Canny gradient G is as follows;
[0077] Take the inter-band mean of high-resolution remote sensing images, convert them into single-band grayscale images, and use Gaussian filtering to reduce noise;
[0078] The first-order partial derivative operator is used to calculate the partial derivative G of the image grayscale along the horizontal and vertical directions. x , G y , find the magnitude of the gradient G f and orientation θ;
[0079]
[0080]
[0081] Perform non-maximum suppression on the gradient amplitude and find the local gradient maximum, which is the Canny gradient G.
[0082] Optionally, the step of using the Otsu algorithm to perform adaptive threshold segmentation on the cloud information representation index comprises the following steps:
[0083] The total inter-class variance g of the cloud and background in the cloud information representation index is calculated as follows:
[0084] g=w0*(u0-u) 2 +w1*(u1-u)2
[0085] Among them, let the initial segmentation threshold be t, w0 is the proportion of cloud pixels in the image, u0 is the average grayscale, w1 is the proportion of background pixels in the image, u1 is the average grayscale, u is the total average grayscale of the image, u=w0*u0+w1*u1;
[0086] By iteratively calculating the total inter-class variance g, find the segmentation threshold that maximizes the inter-class variance g, set to t max , the maximum segmentation threshold t max The optimal threshold is set, and the binary image is segmented according to the optimal threshold.
[0087] See also Figure 4 In this embodiment, since the cloud pixels are concentrated in the high-value area of the CRI image, the inter-class pixel variance between the background and the cloud is large, so the Otsu algorithm can segment the cloud. Through iterative calculation, find the t that maximizes the inter-class variance g. max , is the optimal threshold, according to which cloud pixels can be segmented. Finally, the optimal segmentation threshold t(a) for the two 0.5-meter images is max =120, t(b) max =62, the optimal segmentation threshold t(c) for two 2-meter images max =75, t(d) max =104.
[0088] Optionally, the geometric features include:
[0089] The geometric features include at least the area, compactness and aspect ratio of the spots in the binary image, and the straight lines and right angles of the grayscale image.
[0090] See also Figure 2 ,In this embodiment, the area, compactness and aspect ratio of the patches in the binary image, the ,straight lines and right angles of the grayscale image and other geometric ,features are used to filter out small areas, broken buildings, narrow roads and other artificial ,ground objects and non-cloud objects.
[0091] Optionally, the area, compactness and aspect ratio of the image spots, and the straight line and right angle features of the grayscale image include:
[0092] The area of the pattern is set to A, A<100 square meters, the aspect ratio of the pattern is set to B, B>50, the compactness of the pattern is set to C, C<0.0001, and the calculation formula of the aspect ratio B of the pattern is:
[0093]
[0094] Wherein, length and width are the length and width of the minimum circumscribed rectangle of the patch, respectively. The calculation formula of the compactness C of the patch is:
[0095]
[0096] Among them, Area and Perimeter are the area and perimeter of the patch respectively.
[0097] Since artificial objects such as buildings and roads have obvious geometric features such as straight lines and right angles, non-cloud objects such as artificial objects can be filtered out by extracting the straight line and right angle geometric features of the grayscale image. The straight line and right angle feature extraction process is as follows:
[0098] The LSD line detection algorithm first performs Gaussian downsampling on the grayscale image to reduce the impact of edge jaggedness on the detection result, and then calculates the grayscale gradient and level-line of the local boundary of the image. The image grayscale gradient is calculated on a 2×2 template according to the following formula:
[0099]
[0100] The angle of Level-Line is calculated according to the above formula:
[0101]
[0102] Where i(x,y) is the grayscale value of the pixel (x,y) on the grayscale image, g x (x,y) is the gradient value in the X direction, g y (x, y) is the gradient value in the Y direction, and G(x, y) is the grayscale gradient amplitude at the pixel point (x, y).
[0103] Then, the pixels are sorted and the straight area is grown according to their gradient values. The larger the gradient value, the greater the probability that the pixel is an edge point. It can be used as a seed point for line detection. Finally, the false alarm index is calculated to further determine the fitted rectangle of the target area, and then the straight lines in the target image are obtained. The right angles are further extracted through the distance and angle relationship between the straight lines.
[0104] Optionally, the designing of a morphological cloud filter operator DMCO, using morphological filtering to filter the binary image again, and further purifying to obtain a cloud binary image, includes:
[0105] Based on the high-resolution remote sensing image, the inter-band mean is taken to obtain a single-band grayscale image, and the grayscale image is processed using the designed morphological cloud filtering operator DMCO to extract non-cloud objects in the grayscale image, and then the spots in the binary image that intersect with the spots of non-cloud objects are deleted.
[0106] See also Figure 2 ,In this embodiment, the designed DMCO operator is used to further purify the cloud bodies in the binary ,image.
[0107] Optionally, the designing of a morphological cloud filter operator DMCO, using morphological filtering to filter the binary image again to further purify the binary image of the cloud body, further includes:
[0108] After processing the grayscale image, the morphological cloud filter operator DMCO extracts non-cloud objects with DMCO>Et by setting a threshold Et=0. The calculation formula of DMCO is as follows:
[0109]
[0110] D N and S N Respectively represent the number of directions and sizes of the structural elements, Indicates that the direction of the structure is d i and size s j The differential morphological characteristics of
[0111]
[0112] in, is the morphological white hat reconstruction, G is the grayscale image, This is the morphological opening operation.
[0113] See also Figure 2 In this embodiment, the morphological filtering method is further used to filter out artificial objects such as large buildings and open-air squares, purify the cloud body, and take the inter-band mean according to the remote sensing image in step 1 to obtain a single-band grayscale image. By setting the threshold Et=0, non-cloud objects with DMCO>Et are extracted.
[0114] See also Figure 5 Specifically, considering the actual size of the cloud cluster in the image, this embodiment sets the number of directions of the structure element D N =12, number of sizes S N =10, differential morphological features The structural element direction d in i The value sequence of is D = {15°, 30°, 45°, 60°, 75°, 90°, 105°, 120°, 135°, 150°, 165°, 180°}, and the size s j The value sequence of is S = {20, 40, 60, 80, 100, 120, 140, 160, 180, 200}, which is the non-cloud objects extracted by DMCO.
[0115] Optionally, the morphological closing operation includes:
[0116] The morphological closing operation formula is:
[0117]
[0118] Among them, · is the morphological closing operation, is the morphological dilation, ⊙ is the morphological erosion, B is the circular structure element, B=strel('ELLIPSE',s,s), s is the size.
[0119] In this embodiment, according to the binary image filtered and purified in step 3 and step 4, a morphological closing operation is performed to fill the holes, and then the binary raster image is converted into a vector, and the background is deleted to obtain the following: Figure 5 The cloud detection vector results shown in the figure are statistically analyzed, and the cloud area and cloud percentage of the 0.5-meter image and 2-meter image are shown in Table 1:
[0120] Table 1. Statistics of experimental image cloud detection results (area unit: square kilometers)
[0121] Image number Data imaging satellite Image resolution Image coverage area Cloud area Cloud cover percentage (a) Gaojing-1 satellite 0.5m 166 27.28 16.43% (b) GeoEye-1 satellite 0.5m 167 8.43 5.05% (c) Ziyuan-3 satellite 2 meters 2550 326.57 12.81% (d) Gaofen-1C 2 meters 4488 1756.07 39.13%
[0122] Furthermore, the morphological closing operation formula is:
[0123]
[0124] Among them, · is the morphological closing operation, is the morphological dilation, ⊙ is the morphological erosion, B is the circular structure element, B = strel('ELLIPSE', s, s), s = 10 meters.
[0125] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0126] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A remote sensing cloud detection method taking into account cloud information representation index and geometric morphological characteristics, characterized in that: The steps include: Step 1, designing a cloud information characterization index based on the haze optimization transformation index, texture contrast and Canny gradient obtained from high-resolution remote sensing images, wherein the haze optimization transformation index, texture contrast and Canny gradient are all normalized and stretched, and the cloud information characterization index design formula is: CRI=HOT-Con-G Wherein, the CRI is the cloud information characterization index, the HOT is the haze optimization transformation index, Con is the texture contrast, and G is the Canny gradient; Step 2, calculate the total inter-class variance g of the cloud and background in the cloud information representation index, and the calculation formula is: g=w0*(u0-u) 2 +w1*(u1-u) 2 Among them, let the initial segmentation threshold be t, w0 is the proportion of cloud pixels in the image, u0 is the average grayscale, w1 is the proportion of background pixels in the image, u1 is the average grayscale, u is the total average grayscale of the image, u=w0*u0+w1*u1; By iteratively calculating the total inter-class variance g, find the segmentation threshold that maximizes the inter-class variance g, set to t max , the maximum segmentation threshold t max Set as an optimal threshold, segment the binary image according to the optimal threshold, and obtain a binary image including the background and the cloud target; Step 3, filtering non-cloud objects and purifying cloud bodies in the binary image using geometric features, wherein the geometric features at least include the area, compactness and aspect ratio of the spots in the binary image, the straight line and right angle features of the grayscale image, and the area, compactness and aspect ratio of the spots, and the straight line and right angle features of the grayscale image, including: The area of the pattern is set to A, A<100 square meters, the aspect ratio of the pattern is set to B, B>50, the compactness of the pattern is set to C, C<0.0001, and the calculation formula of the aspect ratio B of the pattern is: Wherein, length and width are the length and width of the minimum circumscribed rectangle of the patch, respectively. The calculation formula of the compactness C of the patch is: Among them, Area and Perimeter are the area and perimeter of the patch respectively. The straight line and right angle features of the grayscale image are extracted by the LSD line detection algorithm. The grayscale image is first Gaussian downsampled, and then the grayscale gradient and level-line of the local boundary of the image are calculated. The grayscale gradient of the image is calculated on a 2×2 template according to the following formula: The angle of Level-Line is: Where i(x,y) is the grayscale value of the pixel (x,y) on the grayscale image, g x (x,y) is the gradient value in the X direction, g y (x, y) is the gradient value in the Y direction, and G(x, y) is the grayscale gradient amplitude at the pixel point (x, y). Then, the pixel gradient values are sorted and the straight line area is grown, and the fitting rectangle of the target area is further determined by calculating the false alarm index, so as to obtain the straight lines in the target image, and further extract the right angles through the distance and angle relationship between the straight lines. The extracted straight line and right angle features are used to filter the non-cloud object spots such as artificial objects in the binary image, and the cloud body is refined and purified; Step 4, designing a morphological cloud filter operator DMCO, using morphological filtering to filter the binary image after the geometric feature filtering again, and further purifying it to obtain a cloud binary image; Step 5, performing morphological closing operation on the cloud binary image, filling holes, outputting a binary grid image, and converting the binary grid image into a vector, deleting the background, and obtaining a cloud detection vector result.
2. The remote sensing cloud detection method taking into account cloud information representation index and geometric morphological characteristics according to claim 1, characterized in that: The morphological cloud filter operator DMCO is designed to filter the binary image again by morphological filtering to further purify the binary cloud image, including: Based on the high-resolution remote sensing image, the inter-band mean is taken to obtain a single-band grayscale image, and the grayscale image is processed using the designed morphological cloud filtering operator DMCO to extract non-cloud objects in the binary image, and then the spots in the binary image that intersect with the spots of non-cloud objects are deleted.
3. The remote sensing cloud detection method taking into account cloud information representation index and geometric morphological characteristics according to claim 2, characterized in that: The designed morphological cloud filter operator DMCO uses morphological filtering to filter the binary image again to further purify the cloud binary image, and also includes: After processing the grayscale image, the morphological cloud filter operator DMCO extracts non-cloud objects with DMCO>Et by setting a threshold Et=0. The calculation formula of DMCO is as follows: D N and S N Respectively represent the number of directions and sizes of the structural elements, Indicates that the structural element direction is d i and size s j The differential morphological features of in, is the morphological white hat reconstruction, G is the grayscale image, This is the morphological opening operation.
4. The remote sensing cloud detection method taking into account cloud information representation index and geometric morphological characteristics according to claim 1, characterized in that: The morphological closing operation includes: The morphological closing operation formula is: Among them, · is the morphological closing operation, is the morphological dilation, ⊙ is the morphological erosion, B is the circular structure element, B=strel('ELLIPSE',s,s), s is the size.
