A thick cloud-thin cloud detection method and system based on multi-scale depth model

Abstract

The present invention discloses a method and system for integrated thick-and-thin cloud detection based on spectral features and a multi-scale depth model. The method belongs to the field of remote sensing image processing technology and includes the following steps: 1. Training multiple scene-level models of different scales based on a cloud detection dataset to generate multiple scene-level cloud probability maps and scene-level binary cloud masks of different scales; 2. Generating pixel-level cloud probability maps for thick and thin clouds based on the multiple scene-level cloud probability maps of different scales and the dark pixel features of the image; and 3. Combining the scene-level binary cloud mask, fusing the pixel-level cloud probability maps for thick and thin clouds using dark pixel gradients. Then, combining adaptive thresholding and distance weighting to generate a pixel-level binary cloud mask, the method achieves integrated thick-and-thin cloud detection. The method avoids reliance on manually set segmentation thresholds, achieving the fusion of thick and thin cloud probability maps and fully automated and accurate detection of thick and thin clouds, as well as cloud edges.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing image processing, and in particular relates to a thick cloud-thin cloud detection method and system based on a multi-scale depth model. Background Art

[0002] Optical remote sensing satellites often encounter cloud and fog interference when acquiring images, resulting in varying degrees of information loss. Areas covered by thick cloud can completely lose information, while areas covered by thin cloud can cause radiometric distortion, impacting image usability. Therefore, in optical remote sensing image processing, generating accurate cloud masks through cloud detection provides a reliable foundation for subsequent image interpretation and analysis, as well as product production, enhancing the application value of remote sensing imagery.

[0003] Currently, there are two main types of cloud detection methods: multi-temporal and single-temporal. Multi-temporal methods detect clouds by analyzing pixel mutations in time series images. However, their reliance on multiple images of the same area captured at different time periods significantly limits their practical application. Single-temporal methods primarily utilize cloud features within a single image to detect clouds and can be further categorized as physical rule-based and deep learning-based methods. Physical rule-based methods utilize the physical properties of clouds (such as high reflectivity and low temperatures) to design rules and segmentation thresholds for cloud detection. However, in practical applications, these methods typically require manual selection of rules and thresholds, making it difficult to obtain optimal parameters and limiting detection accuracy. In contrast, deep learning-based methods utilize large-scale datasets to train image classification models, automatically learning cloud features and achieving high-precision cloud detection, offering greater adaptability and accuracy.

[0004] The accuracy of existing deep learning-based cloud detection methods relies on large-scale, high-quality training samples. However, existing training samples often lack thin cloud samples. As a result, most current cloud detection methods can only detect thick clouds, but have difficulty accurately detecting thin clouds, which greatly limits the practical application of such methods.

[0005] Therefore, it is necessary to design a thick cloud-thin cloud detection method and system based on a multi-scale depth model to address the above problems. Summary of the Invention

[0006] The purpose of the present invention is to address the problem that existing cloud detection methods are often difficult to accurately detect thin clouds due to factors such as the complexity of thin cloud features and the quality of training samples. This invention provides a thick cloud-thin cloud detection method based on a multi-scale depth model. The thick cloud-thin cloud integrated detection technology that combines spectral features with a multi-scale depth model is based on a multi-scale scene-level depth model and utilizes the differentiated optical characteristics of thick and thin clouds to generate a pixel-level cloud probability map. The adaptive optimization strategy based on adaptive thresholds and distance weighting realizes automatic and accurate detection of thick clouds, thin clouds and cloud edges.

[0007] According to one aspect of the present specification, a thick cloud-thin cloud detection method based on a multi-scale depth model is provided, comprising:

[0008] Step 1: Based on the cloud detection dataset, multiple scene-level models of different scales are trained to generate scene-level cloud probability maps and scene-level binary cloud masks of different scales.

[0009] Step 2: Based on multiple scene-level cloud probability maps of different scales and the dark pixel features of the image to be detected, pixel-level cloud probability maps for thick clouds and thin clouds are generated respectively;

[0010] Step 3: Combined with the scene-level binary cloud mask, the dark pixel gradient is used to fuse the pixel-level cloud probability maps for thick clouds and thin clouds. Then, the adaptive threshold and distance weighting are combined to generate the pixel-level binary cloud mask to achieve integrated thick and thin cloud detection.

[0011] Furthermore, the step 1 generates a plurality of scene-level cloud probability maps and scene-level binary cloud masks of different scales, including:

[0012] Using the trained scene-level model, the images in the cloud detection dataset are divided into blocks to determine whether each image block contains clouds;

[0013] If the image block contains clouds, all pixels contained in the image block are marked as clouds to obtain a scene-level cloud probability map and a scene-level binary cloud mask.

[0014] Furthermore, generating pixel-level cloud probability maps of thick clouds and thin clouds in step 2 includes:

[0015] The dark pixel characteristics are calculated using the reflectance of the blue and green bands in the cloud detection image;

[0016] Multiply the scene-level cloud probability maps of multiple different scales with the dark pixel features pixel by pixel to obtain the pixel-level cloud probability map of thick clouds;

[0017] Singular value decomposition is used to decompose the dark pixel features to obtain a low-rank approximation of the dark pixel features;

[0018] The pixel-level cloud probability map of thin clouds is obtained by multiplying multiple scene-level cloud probability maps of different scales with the low-rank approximation of dark pixel features pixel by pixel.

[0019] Furthermore, in step 3, the scene-level binary cloud mask is combined with the dark pixel gradient to fuse the pixel-level cloud probability maps for thick clouds and thin clouds, including:

[0020] The Sobel operator is used to calculate the gradient amplitude of dark pixel features and segment the binary edge mask of the image to be detected.

[0021] The intersection of the generated scene-level binary cloud masks of different scales is taken as the cloud coverage area. At the same time, the proportion of binary edge masks in the cloud coverage area is clarified. The pixel-level cloud probability maps for thick clouds and thin clouds are fused to obtain a fused pixel-level cloud probability map.

[0022] Furthermore, the step 3 of generating a pixel-level binary cloud mask includes:

[0023] The fused pixel-level cloud probability map is segmented by adaptive thresholding to generate an initialized binary cloud mask;

[0024] Distance-weighted adaptive optimization is used to enhance the fused pixel-level cloud probability map;

[0025] Combined with the initialized binary cloud mask, the enhanced fused pixel-level cloud probability map is re-segmented by adaptive thresholding to generate the final pixel-level binary cloud mask.

[0026] Furthermore, the extraction of the adaptive threshold includes:

[0027] By counting the differences in coverage areas of scene-level binary cloud masks at different scales, the adaptive threshold of the image to be detected is automatically extracted.

[0028] Furthermore, the expression of distance-weighted adaptive optimization is adopted, specifically:

[0029]

[0030] Where, is the enhanced pixel-level cloud probability map, is the fused pixel-level cloud probability map, is the adaptive distance constant, is the distance from the pixel at row i and column j in the image to its nearest cloud block, is the probability of thick clouds, is the probability of thin clouds.

[0031] According to one aspect of the present specification, a thick cloud-thin cloud detection system based on a multi-scale depth model is provided, comprising:

[0032] The multi-scale scene-level model module is used to train scene-level models of multiple scales based on the cloud detection dataset, generating scene-level cloud probability maps and scene-level binary cloud masks of multiple scales;

[0033] The cloud probability map generation module is used to generate pixel-level cloud probability maps for thick clouds and thin clouds based on scene-level cloud probability maps of multiple scales and dark pixel features of the image to be detected;

[0034] The thick-thin cloud integrated detection module is used to combine the scene-level binary cloud mask and use the dark pixel gradient to fuse the pixel-level cloud probability maps for thick and thin clouds. It then combines the adaptive threshold and distance weighting to generate the pixel-level binary cloud mask to achieve integrated thick-thin cloud detection.

[0035] According to one aspect of the present specification, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the thick cloud-thin cloud detection method based on the multi-scale depth model are implemented.

[0036] According to one aspect of the present specification, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the thick cloud-thin cloud detection method based on the multi-scale depth model are implemented.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. This method extracts the optical features of thick and thin clouds separately, and fuses them with a multi-scale scene-level cloud probability map using a differentiated weighting strategy to generate a pixel-level cloud probability map. Compared with single-feature cloud detection methods, this method more accurately reflects the distribution characteristics of thick and thin clouds in images, significantly reducing labor costs while achieving high-precision detection of thick and thin clouds.

[0039] 2. Based on the difference in the characteristic gradient amplitude of dark pixels in thick and thin cloud areas, this paper combines multi-scale scene-level binary cloud masks to achieve the fusion of thick cloud probability maps and thin cloud probability maps.

[0040] 3. By analyzing the differences in coverage areas of multi-scale scene-level cloud masks, the present invention avoids relying on manually set segmentation thresholds, adaptively extracts the segmentation thresholds of images, and combines a distance-weighted adaptive optimization strategy to generate binary cloud masks, thus achieving fully automated and accurate detection of thick clouds, thin clouds, and cloud edges. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0043] Figure 2 This is a structural diagram of the scene-level deep learning cloud detection model of an embodiment of the present invention. DETAILED DESCRIPTION

[0044] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other.

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] like Figure 1 As shown, an embodiment of the present invention provides a thick cloud-thin cloud detection method based on a multi-scale deep model, including: step 1, training multiple scene-level models of different scales based on a cloud detection dataset and generating multiple scene-level cloud probability maps and scene-level binary cloud masks of different scales; step 2, generating a pixel-level cloud probability map, using the scene-level cloud probability maps of different scales and the dark pixel features of the image to generate pixel-level cloud probability maps for thick clouds and thin clouds respectively; step 3, generating a binary cloud mask, using the dark pixel gradient to fuse the pixel-level cloud probability maps of thick clouds and thin clouds, and combining adaptive threshold and distance weighted optimization to generate a pixel-level binary cloud mask.

[0047] Specifically, the embodiment of the present invention also provides a method for implementing the scene-level cloud probability map and scene-level binary cloud mask of multiple different scales in step 1. First, the existing multi-scale scene-level cloud detection dataset is used to train the deep learning model of different scales. Figure 2 As shown in the figure, the scene-level deep learning cloud detection model used is mainly composed of three parts: the feature extractor, the model backbone, and the classifier. The feature extractor consists of convolutional layers and max pooling layers, which perform preliminary feature extraction of the input image. The model backbone consists of four groups of residual blocks, which extract multi-level features of the image. The classifier consists of average pooling layers and fully connected layers, which map the features extracted by the model to cloud probabilities.

[0048] Specifically, the embodiment of the present invention divides scene-level cloud detection samples of different scales into training sets and test sets in a ratio of 7:3. During the multi-scale model training process, the reflectance of the visible light band of the image in the training set is used as the model input, and the corresponding manually annotated labels are used as the output, where the labels only indicate whether the image has clouds. The model of each scale is trained only using samples of its corresponding scale, and the learning rate of all scale models is uniformly set to 10 -4 Each model is iteratively trained 100 times, and the model with the highest accuracy on the corresponding scale test set is selected as the final scene-level cloud detection model for that scale.

[0049] Specifically, during the model application phase, the image to be detected is segmented into blocks of different scales according to the scale requirements of the multi-scale deep model. These blocks are then fed into the deep model at the corresponding scale to generate a cloud probability for each block, with the probability value for each block being the same. The cloud probabilities for each scale block are then recombined to generate a multi-scale scene-level cloud probability map. Areas with a cloud probability greater than 50% are considered cloud-covered, thus generating a multi-scale scene-level binary cloud mask.

[0050] Specifically, the embodiment of the present invention further provides an implementation method for generating a pixel-level cloud probability map in step 2, and uses the reflectance of the blue band and the green band of the image to be detected to calculate the dark pixel features of the image. The dark pixel features can be calculated by formula (1):

[0051] (1)

[0052] in, is the dark pixel feature of the pixel at the i-th row and j-th column of the image, is the reflectivity of the blue band of the image, It is the reflectance of the green band of the image. The larger the dark pixel value of the image pixel is, the more likely the pixel is a cloud.

[0053] Specifically, in order to obtain the thick cloud pixel-level cloud probability map of the image, the multi-scale scene-level cloud probability map generated in step 1 is multiplied pixel by pixel with the dark pixel features of the image. The final thick cloud pixel-level cloud probability map of the image can be expressed as:

[0054] (2)

[0055] in, is the cloud probability of thick clouds, for The scene-level cloud probability of different scales is arranged from large to small. is the adaptive weight constant of cloud probability maps of different scales, where the value of the weight constant is .

[0056] Specifically, since thin clouds in images usually have the characteristics of large-scale distribution, in order to reduce the interference of local details on the overall feature extraction of thin clouds, the singular value decomposition (SVD) is used to decompose the dark pixel features of the image, which is represented as two orthogonal matrices and a diagonal matrix. By retaining the first 20 largest singular values, a low-rank approximation of the dark pixel features is obtained, which can be expressed as:

[0057] (3)

[0058] in, is a low-rank approximation of the dark pixel features of the image, are the first 20 columns of the left singular vector matrix, are the first 20 singular values, are the first 20 rows of the right singular vector matrix.

[0059] Specifically, this embodiment also multiplies the multi-scale scene-level cloud probability map with the low-rank approximation of the dark pixel features of the image pixel by pixel. The final pixel-level cloud probability map of the thin cloud in the image can be expressed as:

[0060] (4)

[0061] in, The cloud probability for thin clouds.

[0062] Specifically, the embodiment of the present invention further provides an implementation method for generating a binary cloud mask in step 3, using the Sobel operator to calculate the gradient amplitude of the dark pixel features of the image and segmenting the binary edge mask of the image. The binary edge mask can be expressed as:

[0063] (5)

[0064] (6)

[0065] (7)

[0066] Where, is a binary edge mask, is the gradient amplitude of the dark pixel feature of the image. The edge area of ​​thick clouds usually has a higher gradient amplitude, while the thin cloud area has a lower gradient amplitude. are the convolution kernels of the Sobel operator in the horizontal and vertical directions, is the adaptive edge segmentation constant, which is calculated by dividing the gradient amplitude of the dark pixel feature of the image by Sort them in descending order and take the gradient amplitude at the first 15% of the positions after sorting as the segmentation constant, which can be expressed as:

[0067] (8)

[0068] in, is the gradient amplitude of the dark pixel features after sorting from large to small, is the total number of pixels with dark pixel features.

[0069] Specifically, the embodiment of the present invention takes the intersection of the binary cloud masks of different scales generated in step 1 as the clear cloud coverage area , and statistically define the binary edge mask in the cloud coverage area The proportion of thick cloud probability map and thin cloud probability map is realized by fusion. The pixel-level cloud probability map of image fusion can be expressed as:

[0070] (9)

[0071] (10)

[0072] (11)

[0073] in, is the fused pixel-level cloud probability map, is the cloud edge ratio, is the binary scene-level cloud mask, is the fusion constant.

[0074] Specifically, in order to avoid the use of artificial thresholds and improve the versatility of the present invention, the embodiment of the present invention implements an adaptive threshold for automatically extracting images by statistically analyzing the differences in coverage areas of binary scene-level cloud masks at different scales. The adaptive thresholds for different images can be expressed as:

[0075] (12)

[0076] in, is the adaptive threshold of the image, is the probability of thick clouds, which is obtained by counting the probability mean of the area covered by the largest scale scene-level cloud mask. is the probability of thin clouds, which is obtained by statistically averaging the probability of the difference between the maximum-scale scene-level cloud mask and the minimum-scale scene-level cloud mask covering the area.

[0077] Specifically, based on the adaptive threshold Pixel-level cloud probability map after fusion of our segmented images , generates an initialized binary cloud mask, which can be expressed as:

[0078] (13)

[0079] in, To initialize the binary cloud mask, since it uses an adaptive threshold which is not an artificial optimal threshold, it is difficult to achieve accurate coverage of thin clouds. Therefore, a distance-weighted adaptive optimization method is used to optimize the fused pixel-level cloud probability map. To enhance, it can be expressed as:

[0080] (14)

[0081] in, is the enhanced pixel-level cloud probability map, is the adaptive distance constant, which can be obtained by Calculated, is the distance from the pixel at row i and column j to its nearest cloud block. Finally, the adaptive threshold Re-enhanced pixel-level cloud probability map After segmentation, the final binary cloud mask can be generated.

[0082] Specifically, as described above, embodiments of the present invention provide a thick-thin cloud detection method based on a multi-scale deep model. First, a multi-scale scene-level cloud detection model is trained using an existing multi-scale scene-level cloud detection sample set to generate multi-scale scene-level cloud probabilities and a binary cloud mask. Next, pixel-level cloud probability maps for thick and thin clouds are generated by combining the dark pixel features of thick and thin clouds with the multi-scale scene-level cloud probabilities. Finally, the cloud probabilities for thick and thin clouds are fused based on dark pixel gradients, and a binary pixel-level cloud mask is generated using an adaptive threshold and distance-weighted optimization approach. Existing cloud detection methods rely on manual thresholds or high-quality sample sets to achieve high-precision pixel-level cloud detection. However, setting manual thresholds and labeling high-quality pixel-level samples are both labor-intensive. This method extracts the spectral features of thick and thin clouds from images and combines them with a multi-scale scene-level cloud detection model to obtain pixel-level cloud detection results. This significantly reduces labor costs while achieving high-precision detection of thick and thin clouds.

[0083] Specifically, the embodiments of the present invention take into account that existing cloud detection methods rely on manual thresholds or high-quality sample sets to achieve high-precision pixel-level cloud detection. However, setting manual thresholds and labeling high-quality pixel-level samples require a lot of manpower. By combining the spectral characteristics of thick and thin clouds with a multi-scale depth model, the present invention generates cloud probability maps for thick and thin clouds, respectively, and fuses the cloud probability maps using the gradient amplitude difference of the spectral characteristics of thick and thin clouds. In addition, based on the differences in the coverage area of ​​scene-level cloud masks at different scales, an adaptive threshold for the image is automatically extracted, and a high-precision binary pixel-level cloud mask is obtained using a distance-weighted adaptive optimization method.

[0084] The implementation of each embodiment of the present invention is based on programmed processing performed by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of each embodiment of the present invention are packaged into various modules. Based on this reality, and in addition to the aforementioned embodiments, an embodiment of the present invention provides a thick-thin cloud detection system based on a multi-scale depth model. This system is used to implement the thick-thin cloud detection method based on a multi-scale depth model described in the aforementioned method embodiment.

[0085] The system includes: a multi-scale scene-level model module, which is used to train multiple scene-level models of different scales based on a cloud detection dataset to generate scene-level cloud probability maps and scene-level binary cloud masks of multiple scales; a cloud probability map generation module, which is used to generate pixel-level cloud probability maps for thick clouds and thin clouds respectively based on multiple scene-level cloud probability maps of different scales and the dark pixel features of the image to be detected; a thick cloud-thin cloud integrated detection module, which is used to combine the scene-level binary cloud mask and fuse the pixel-level cloud probability maps for thick clouds and thin clouds using the dark pixel gradient, and then generate the pixel-level binary cloud mask in combination with the adaptive threshold and distance weighting to achieve thick cloud-thin cloud integrated detection.

[0086] The thick-thin cloud detection system based on a multi-scale depth model provided by the embodiment of the present invention solves the problem that existing cloud detection methods are limited by factors such as the complexity of thin cloud features and the quality of training samples. The thick-thin cloud integrated detection technology adopts several modules and combines spectral features with a multi-scale depth model. Based on the multi-scale scene-level depth model, it uses the differentiated optical characteristics of thick and thin clouds to generate a pixel-level cloud probability map, and realizes automatic and accurate detection of thick clouds, thin clouds and cloud edges based on an adaptive optimization strategy based on adaptive thresholds and distance weighting.

[0087] Based on the same inventive concept as the aforementioned embodiment, an embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a thick cloud-thin cloud detection method based on a multi-scale depth model as proposed in the aforementioned embodiment.

[0088] An embodiment of the present invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, this program overcomes the difficulty of accurately detecting thin clouds in existing cloud detection methods, which is often limited by factors such as the complexity of thin cloud features and the quality of training samples. This program accurately reflects the distribution characteristics of thick and thin clouds in images, significantly reducing labor costs while achieving high-precision detection of thick and thin clouds.

[0089] The storage medium can be any non-volatile storage device such as a hard disk, solid-state drive, flash drive, optical disk, etc., which is used to store computer program code and necessary data files. The stored computer program includes: a multi-scale scene-level model module, a cloud probability map generation module, and a thick cloud-thin cloud integrated detection module.

[0090] Finally, it should be noted that the above specific embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and is susceptible to numerous variations. Any simple modifications, equivalent variations, and modifications to the above specific embodiments based on the technical essence of the present invention shall be deemed to fall within the scope of protection of the present invention.

Claims

1. A thick cloud-thin cloud detection method based on a multi-scale depth model, characterized in that: include: Step 1: Based on the cloud detection dataset, multiple scene-level models of different scales are trained to generate scene-level cloud probability maps and scene-level binary cloud masks of different scales. Step 2: Based on multiple scene-level cloud probability maps of different scales and the dark pixel features of the image to be detected, pixel-level cloud probability maps for thick clouds and thin clouds are generated respectively; Step 3: Combined with the scene-level binary cloud mask, the dark pixel gradient is used to fuse the pixel-level cloud probability maps for thick clouds and thin clouds. Then, the adaptive threshold and distance weighting are combined to generate a pixel-level binary cloud mask to achieve integrated detection of thick and thin clouds. The Sobel operator is used to calculate the gradient amplitude of the dark pixel features to segment the binary edge mask of the image to be detected. The intersection of the generated scene-level binary cloud masks of different scales is used as the cloud coverage area. At the same time, the proportion of the binary edge mask in the cloud coverage area is determined, and the pixel-level cloud probability maps for thick clouds and thin clouds are fused to obtain a fused pixel-level cloud probability map.

2. The thick cloud-thin cloud detection method based on a multi-scale depth model according to claim 1, characterized in that: The step 1 generates multiple scene-level cloud probability maps and scene-level binary cloud masks of different scales, including: Using the trained scene-level model, the images in the cloud detection dataset are divided into blocks to determine whether each image block contains clouds; If the image block contains clouds, all pixels contained in the image block are marked as clouds to obtain a scene-level cloud probability map and a scene-level binary cloud mask.

3. The thick cloud-thin cloud detection method based on a multi-scale depth model according to claim 1, characterized in that: Generating pixel-level cloud probability maps of thick clouds and thin clouds in step 2 includes: The dark pixel characteristics are calculated using the reflectance of the blue and green bands in the cloud detection image; Multiply the scene-level cloud probability maps of multiple different scales with the dark pixel features pixel by pixel to obtain the pixel-level cloud probability map of thick clouds; Singular value decomposition is used to decompose the dark pixel features to obtain a low-rank approximation of the dark pixel features; The pixel-level cloud probability map of thin clouds is obtained by multiplying multiple scene-level cloud probability maps of different scales with the low-rank approximation of dark pixel features pixel by pixel.

4. The thick cloud-thin cloud detection method based on a multi-scale depth model according to claim 1, characterized in that: Generating a pixel-level binary cloud mask in step 3 includes: The fused pixel-level cloud probability map is segmented by adaptive thresholding to generate an initialized binary cloud mask; Distance-weighted adaptive optimization is used to enhance the fused pixel-level cloud probability map; Combined with the initialized binary cloud mask, the enhanced fused pixel-level cloud probability map is re-segmented by adaptive thresholding to generate the final pixel-level binary cloud mask.

5. The thick cloud-thin cloud detection method based on a multi-scale depth model according to claim 4, characterized in that: The extraction of the adaptive threshold comprises: By counting the differences in coverage areas of scene-level binary cloud masks at different scales, the adaptive threshold of the image to be detected is automatically extracted.

6. The thick cloud-thin cloud detection method based on a multi-scale depth model according to claim 4, characterized in that: The expression of distance-weighted adaptive optimization is as follows: ， Where, is the enhanced pixel-level cloud probability map, is the fused pixel-level cloud probability map, is the adaptive distance constant, is the distance from the pixel at row i and column j in the image to its nearest cloud block, is the probability of thick clouds, is the probability of thin clouds.

7. A thick cloud-thin cloud detection system based on a multi-scale depth model, characterized in that: include: The multi-scale scene-level model module is used to train scene-level models of multiple scales based on the cloud detection dataset, generating scene-level cloud probability maps and scene-level binary cloud masks of multiple scales; The cloud probability map generation module is used to generate pixel-level cloud probability maps for thick clouds and thin clouds based on scene-level cloud probability maps of multiple scales and dark pixel features of the image to be detected; The thick-thin cloud integrated detection module is used to combine scene-level binary cloud masks and use dark pixel gradients to fuse pixel-level cloud probability maps for thick and thin clouds. Adaptive thresholding and distance-weighted optimization are then used to generate pixel-level binary cloud masks to achieve integrated thick-thin cloud detection. The Sobel operator is used to calculate the gradient amplitude of dark pixel features and segment the binary edge masks of the image to be detected. The intersection of the generated scene-level binary cloud masks of different scales is used as the cloud coverage area. The proportion of binary edge masks in the cloud coverage area is determined, and the pixel-level cloud probability maps for thick and thin clouds are fused to obtain a fused pixel-level cloud probability map.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the thick cloud-thin cloud detection method based on the multi-scale depth model are implemented in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the thick cloud-thin cloud detection method based on a multi-scale depth model according to any one of claims 1 to 6 are implemented.

Images