Remote sensing image cloud detection method, system and device based on reference base map and storage medium

Abstract

The invention discloses a remote sensing image cloud detection method, system and device based on a reference base map and a storage medium, belongs to the technical field of cloud region detection of remote sensing images, and solves the problem that an existing cloud detection method is insufficient in precision and robustness. Comprising the following steps: S1, constructing a data set and a cloud detection model, and training the cloud detection model based on the data set; S2, constructing a cloud-free base map of a single season and a single geographical range, and registering the cloud-free base map with a to-be-detected photograph to form a time sequence image; if the cloud exists, executing the operation of the step S4; and S4, inputting the time sequence influence with clouds into the cloud detection model trained in the step S1 for detection, and then outputting a cloud mask. According to the invention, cloud range detection is carried out on the to-be-detected image with geographic information.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud region detection in remote sensing images, and in particular to a remote sensing image cloud detection method, system, device and storage medium based on a reference base map. Background Art

[0002] In the process of remote sensing image analysis and application, clouds are a common interference factor, especially in high-resolution and multispectral images. The presence of clouds will block surface information, resulting in a decrease in the availability and accuracy of remote sensing data, seriously affecting the processing effect of downstream tasks such as land cover classification, change detection, disaster monitoring, environmental assessment and agricultural monitoring.

[0003] Existing cloud detection methods usually rely on fixed thresholds, spectral features or statistical characteristics, which are difficult to adapt to the complex and changing surface background and different cloud morphologies, resulting in insufficient accuracy and robustness of cloud detection.

[0004] In summary, the existing cloud detection methods are difficult to adapt to the complex and changeable surface background and different cloud morphologies, resulting in insufficient accuracy and robustness of cloud detection. Summary of the Invention

[0005] The present invention solves the problem of insufficient accuracy and robustness of existing cloud detection methods.

[0006] The present invention provides a remote sensing image cloud detection method based on a reference base map, comprising the following steps:

[0007] Step S1: Build a data set and a cloud detection model, and train the cloud detection model based on the data set:

[0008] Step S2: Construct a cloud-free base map of a single season and a single geographical range, and register it with the photography to be detected to form a time series image:

[0009] Step S3, determining whether there are clouds in the time series image. If there are no clouds, directly output a blank cloud mask. If there are clouds, execute the operation of step S4.

[0010] In step S4, the temporal impact of clouds is input into the cloud detection model trained in step S1 for detection, and a cloud mask is output.

[0011] Furthermore, in one embodiment of the present invention, in step S2, constructing a cloud-free base map of a single geographical range in a single season includes the following steps:

[0012] Step S201, constructing historical data into cloud-free base maps of multiple geographical ranges in four seasons;

[0013] Step S202: determining the geographical range of the photography to be detected, selecting the photography that has the same geographical range as the photography to be detected from the cloud-free base maps of multiple geographical ranges in four seasons, thereby constructing a cloud-free base map of a single geographical range in four seasons;

[0014] Step S203 , determining the season of the photography to be detected, selecting the photography of the same season as the photography to be detected from the cloud-free base maps of the four seasons and the single geographical range, thereby constructing a cloud-free base map of the single geographical range in the single season.

[0015] Furthermore, in one embodiment of the present invention, in step S203, the determining of the season of the photography to be detected is specifically:

[0016] The season of the photography to be detected is determined based on the global average NDVI value of the photography to be detected.

[0017] Furthermore, in one embodiment of the present invention, in step S3, determining whether there are clouds in the time-series images includes the following steps:

[0018] Step S301, calculating a difference map based on the photography to be detected and a cloud-free base map of a single geographical range in a single season;

[0019] Step S302, calculating a normalized difference index based on the difference map;

[0020] Step S303, setting a first threshold based on the normalized difference index calculated in step S302, and calculating the number of pixels in the difference map that exceed the first threshold;

[0021] In step S304 , a second threshold is set, and whether there is cloud in the time series image is determined based on the relationship between the second threshold and the number of pixels.

[0022] Furthermore, in one embodiment of the present invention, in step S301, the difference map is calculated based on the photography to be detected and a cloud-free base map of a single geographical range in a single season, specifically:

[0023] Diff = Image2 - Image1;

[0024] Among them, Diff is the difference map, Image2 is the photography to be detected, and Image1 is the cloud-free base map of a single geographical range in a single season.

[0025] Furthermore, in one embodiment of the present invention, in step S303, the number of pixels in the difference map exceeding the first threshold is calculated as follows:

[0026]

[0027] Among them, M is the number of pixels, T1 is the threshold, NDI is the normalized difference index, NDI i is the normalized difference index of each pixel in the difference map, and N is the total number of pixels in the difference map.

[0028] Furthermore, in one embodiment of the present invention, in step S304, the second threshold is set, and whether the time series image has clouds is determined based on the relationship between the second threshold and the number of pixels, specifically:

[0029] A second threshold is set. When the second threshold is greater than the number of pixels, the time series image is determined to be cloudless. When the second threshold is less than the number of pixels, the time series image is determined to be cloud-covered.

[0030] The remote sensing image cloud detection system based on a reference base map described in the present invention includes the following modules:

[0031] Module S1 builds a dataset and a cloud detection model, and trains the cloud detection model based on the dataset:

[0032] Module S2 constructs a cloud-free basemap for a single season and a single geographic area, and registers it with the photography to be detected to form a time series image:

[0033] Module S3 determines whether there are clouds in the time series image. If there are no clouds, a blank cloud mask is directly output. If there are clouds, the operation of module S4 is executed.

[0034] Module S4 inputs the cloud detection model trained by module S1 into the temporal impact of clouds, and then outputs a cloud mask.

[0035] An electronic device according to the present invention comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0036] Memory for storing computer programs;

[0037] The processor is used to implement the method steps described above when executing the program stored in the memory.

[0038] The present invention provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, the method steps described above are implemented.

[0039] This invention solves the problem of insufficient accuracy and robustness of existing cloud detection methods. Specific beneficial effects include:

[0040] 1. The present invention describes a remote sensing image cloud detection method based on a reference base map. In practical applications, surface conditions will change over time (such as vegetation growth, seasonal changes, land use changes, etc.). Existing cloud detection methods are difficult to cope with these surface changes, resulting in false detections. To solve the above technical problems, the present invention constructs and uses a multi-temporal cloud-free reference base map, which can dynamically adjust the cloud detection model, distinguish between cloud layer changes and surface changes, and improve the adaptability and accuracy of cloud detection.

[0041] 2. The present invention describes a remote sensing image cloud detection method based on a reference basemap. In high-resolution remote sensing images, clouds and bright surface features (such as snow, sand dunes, saline-alkali land, and urban buildings) may exhibit similar spectral characteristics, which can easily lead to existing cloud detection methods misidentifying these features as clouds. This invention utilizes a cloud-free reference basemap and compares the spectral differences and texture features of real-time images with the cloud-free basemap to accurately identify cloud areas, significantly reducing false positives.

[0042] 3. The present invention describes a method for cloud detection in remote sensing images based on a reference basemap. With the explosive growth of remote sensing data, existing cloud detection algorithms face the problem of low computational efficiency when processing large-scale remote sensing image data. This bottleneck has limited the widespread deployment and promotion of cloud detection algorithms in large-scale remote sensing applications. This invention designs an efficient algorithm framework that can improve processing speed while maintaining detection accuracy, making it suitable for large-scale datasets and real-time cloud detection needs.

[0043] The present invention describes a remote sensing image cloud detection method based on a reference basemap, which can effectively identify and segment cloud areas in remote sensing images, thereby providing technical support for image cloud removal or compensation. This method can be widely used in a variety of fields, including environmental monitoring and assessment, agricultural planting and yield forecasting, land use and land cover change detection, disaster monitoring (such as floods and forest fires), ecological protection, resource exploration, meteorological research, ocean monitoring, and urban remote sensing monitoring. By improving the quality and usability of remote sensing imagery, the present invention can significantly enhance the analytical accuracy and decision-making support capabilities of various applications in these fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0045] Figure 1 1 is the AttUNet network structure diagram described in the first embodiment;

[0046] Figure 2 1 is a structural diagram of the attention mechanism described in embodiment 1;

[0047] Figure 3This is the AttUNet network training flow chart described in Implementation Method 1;

[0048] Figure 4 This is the cloud detection flow chart described in Implementation Method 1. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe various embodiments of the present invention in conjunction with the accompanying drawings. The embodiments described with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0050] Implementation method 1: A remote sensing image cloud detection method based on a reference base map described in this implementation method includes the following steps:

[0051] Step S1: Build a data set and a cloud detection model, and train the cloud detection model based on the data set:

[0052] Step S2: Construct a cloud-free base map of a single season and a single geographical range, and register it with the photography to be detected to form a time series image:

[0053] Step S3, determining whether there are clouds in the time series image. If there are no clouds, directly output a blank cloud mask. If there are clouds, execute the operation of step S4.

[0054] In step S4, the temporal impact of clouds is input into the cloud detection model trained in step S1 for detection, and a cloud mask is output.

[0055] In this embodiment, in step S2, the step of constructing a cloud-free base map for a single geographical area in a single season includes the following steps:

[0056] Step S201, constructing historical data into cloud-free base maps of multiple geographical ranges in four seasons;

[0057] Step S202: determining the geographical range of the photography to be detected, selecting the photography that has the same geographical range as the photography to be detected from the cloud-free base maps of multiple geographical ranges in four seasons, thereby constructing a cloud-free base map of a single geographical range in four seasons;

[0058] Step S203 , determining the season of the photography to be detected, selecting the photography of the same season as the photography to be detected from the cloud-free base maps of the four seasons and the single geographical range, thereby constructing a cloud-free base map of the single geographical range in the single season.

[0059] In this embodiment, in step S203, the determination of the season of the photograph to be detected is specifically:

[0060] The season of the photography to be detected is determined based on the global average NDVI value of the photography to be detected.

[0061] In this embodiment, in step S3, determining whether there are clouds in the time-series images includes the following steps:

[0062] Step S301, calculating a difference map based on the photography to be detected and a cloud-free base map of a single geographical range in a single season;

[0063] Step S302, calculating a normalized difference index based on the difference map;

[0064] Step S303, setting a first threshold based on the normalized difference index calculated in step S302, and calculating the number of pixels in the difference map that exceed the first threshold;

[0065] In step S304 , a second threshold is set, and whether there is cloud in the time series image is determined based on the relationship between the second threshold and the number of pixels.

[0066] In this embodiment, in step S301, the difference map is calculated based on the photograph to be detected and the cloud-free base map of a single season and a single geographical range, specifically:

[0067] Diff = Image2 - Image1;

[0068] Among them, Diff is the difference map, Image2 is the photography to be detected, and Image1 is the cloud-free base map of a single geographical range in a single season.

[0069] In this embodiment, in step S303, the number of pixels in the difference map exceeding the first threshold is calculated as follows:

[0070]

[0071] Among them, M is the number of pixels, T1 is the threshold, NDI is the normalized difference index, NDI i is the normalized difference index of each pixel in the difference map, and N is the total number of pixels in the difference map.

[0072] In this embodiment, in step S304, the second threshold is set, and whether the time series image contains clouds is determined based on the relationship between the second threshold and the number of pixels. Specifically,

[0073] A second threshold is set. When the second threshold is greater than the number of pixels, the time series image is determined to be cloudless. When the second threshold is less than the number of pixels, the time series image is determined to be cloud-covered.

[0074] Existing cloud detection methods are difficult to adapt to the complex and changing surface background and different cloud morphologies, resulting in insufficient accuracy and robustness of cloud detection.

[0075] For example, current cloud detection methods using optical remote sensing images are prone to missed and false detections in complex backgrounds, such as snow-covered areas, highly reflective buildings, desert regions, and bodies of water. These scenes share spectral and textural characteristics with clouds, making it difficult for detection models to distinguish them, thus reducing detection accuracy and reliability.

[0076] To solve the above technical problems, this embodiment designs a remote sensing image cloud detection method based on a reference base map, which includes the following steps:

[0077] Step S1, cloud detection model training phase:

[0078] Step S101, data set construction:

[0079] The dataset consists of image data and annotations. The image data consists of a 2N-dimensional image collection of images from different seasons and different land cover types. The 2N-dimensional images are composed of cloud-free and cloud-covered image bands accurately registered in the same area, where 1 to N dimensions are cloud-free image bands and N+1 to 2N dimensions are cloud-covered image bands. The annotations are generated by annotating cloud-covered images. The dataset is annotated using the OSTU threshold method combined with manual review and modification. To improve the generalization ability of the model, the dataset is subjected to data augmentation (data enhancement) such as rotation, translation, flipping, shearing, scaling, random cropping, and color transformation to further increase data diversity and prevent model overfitting.

[0080] Step S102: Building a cloud detection model:

[0081] like Figure 1 As shown in the figure, AttUNnet is selected as the main network framework of the model. AttUNnet is an improved version of UNet, which is introduced in the jump connection part of U-Net. Figure 2 The attention module shown in the figure is used to enhance the attention to the region of interest to improve the performance of segmentation of complex backgrounds.

[0082] The AttUNnet network framework structure is an existing structure and will not be described in detail here.

[0083] Step S103: Training and tuning of cloud detection model:

[0084] Based on the dataset constructed in step S1, such as Figure 3 As shown in the figure, DiceLoss is selected as the loss function, Adam is selected as the optimizer, the initial learning rate of the model is set to 0.0001, and the learning rate is continuously lowered as the epoch increases. The hyperparameters such as Batch Size (batch size) and Epochs (number of training rounds) are set.

[0085] Using Intersection of Union (IoU) as the metric, the model with the highest IoU on the validation set is considered the best model.

[0086] In existing technologies, cloud detection methods using optical remote sensing images mostly focus on the overall detection accuracy of free datasets, but lack the impact of temporal and seasonal factors on cloud detection accuracy. Such limitations may lead to a decrease in the accuracy of cloud detection methods in practical applications.

[0087] To solve the above technical problems, this embodiment designs the following steps:

[0088] Step S2, cloud detection application phase, such as Figure 4 As shown:

[0089] Step S201: Constructing cloud-free basemaps for multiple seasons and multiple geographic areas:

[0090] According to the characteristics of different seasons, the historical archived data are constructed into cloud-free base maps of multiple geographical ranges in the four seasons of spring, summer, autumn and winter.

[0091] Step S202: Determine the geographical scope of the image to be inspected:

[0092] The geographic range of the image to be inspected is determined according to the coordinate range of the image to be inspected. At this time, cloud-free base maps of the four seasons of spring, summer, autumn and winter with the same geographic range as the coordinate range of the image to be inspected are selected.

[0093] Step S203, determining the season of the image to be inspected:

[0094] The corresponding season is determined according to the global NDVI mean value of the image to be inspected, and the base map of the corresponding season that is most similar to the NDVI of the image to be inspected is selected as a reference.

[0095] Step S204: Registration of time-series images:

[0096] The image to be detected is registered with the cloud-free basemap of a selected single season and single geographic range, and the overlapping area is constructed as a 2*N-dimensional tensor, where N is the dimension of the basemap.

[0097] Therefore, this embodiment improves the detection accuracy of cloud detection by subsequent cloud detection models by considering the influence of registered time-series images and seasonal factors on cloud detection accuracy.

[0098] Step S3: Quickly determine whether there is cloud in the time series image. Figure 4 As shown:

[0099] To improve the time efficiency of cloud detection, first quickly determine whether there are clouds in the time-series images. For time-series images without clouds, directly output an empty white cloud mask to improve the utilization efficiency of computing resources. For time-series images with clouds, input them into the cloud detection model for cloud detection.

[0100] The quick determination of whether there are clouds in the time-series images includes the following steps:

[0101] Step S301, calculate the difference map. The calculation formula of the difference map is:

[0102] Diff = Image2 - Image1;

[0103] Among them, Diff is the difference map, Image2 is the to-be-detected photograph, and Image1 is a cloud-free base map of a single season and a single geographical range.

[0104] Step S302, calculate NDI (Normalized Difference Index) based on the difference map. The calculation formula of NDI is:

[0105]

[0106] Among them, Green diff and Red diff are the green channel and the red channel in the difference map respectively.

[0107] Step S303, set the first threshold T1, and calculate the number M of pixels in the difference map that exceed the first threshold T1. The calculation formula of the number M of pixels is:

[0108]

[0109] Among them, M is the number of pixels, T1 is the threshold, NDI is the Normalized Difference Index, NDI i is the Normalized Difference Index of each pixel in the difference map, and N is the total number of pixels in the difference map.

[0110] Step S304, set the second threshold T2. When M < T2, determine that the time-series image is a cloud-free image and output an empty white cloud mask; when M > T2, input the time-series image into AttUNet for cloud detection and output a cloud mask.

[0111] Step S4, detection of cloud range:

[0112] Input a 2*N-dimensional tensor into AttUNet for cloud detection, obtain the cloud detection range result, and output a cloud range vector.

[0113] Therefore, the remote sensing image cloud detection method based on a reference base map described in this embodiment can improve the time efficiency and detection accuracy of optical remote sensing image cloud detection and effectively reduce the false detection rate and missed detection rate.

[0114] In order to better illustrate the remote sensing image cloud detection method based on a reference base map described in this embodiment, the following examples are described in detail:

[0115] Step S1, cloud detection model training phase:

[0116] Step S101, data set construction:

[0117] First, we collected 0.5m resolution, four-band data from the wide-band Jilin-1 satellite series. This data included both cloud-free and cloud-covered images of the same region. For the data portion of the dataset, we registered the cloud-covered and cloud-free images, combining their four bands into an eight-band dataset. The annotation portion of the dataset was performed on the cloud-covered images. We used the OTSU method for preliminary cloud detection and morphological processing, then outputted cloud vectors. We used QGIS software to edit any inaccurate cloud ranges from the OTSU cloud detection method to ensure accurate cloud vectors, which were then converted into binary cloud masks. This resulted in a dataset with a one-to-one correspondence between data and annotations. The dataset should cover diverse land cover types, including forestland, cultivated land, grassland, built-up areas, bare soil, water bodies, and deserts, as well as data from different seasons.

[0118] Step S102: Building a cloud detection model:

[0119] AttUNnet is selected as the main network framework structure of the model. AttUNnet is an improved version of UNet. It introduces an attention module in the jump connection part of U-Net to enhance the focus on the region of interest and improve the performance of segmenting complex backgrounds.

[0120] Step S103: Training and tuning of cloud detection model:

[0121] The dataset is input into the AttUnet network, the number of iterations is set to 400,000, the Adam optimizer is used, the initial learning rate is set to 1e-4, and the learning rate is set to decay by 1 / 2 every 100,000 iterations. The Dice-loss is used as the loss function to guide the training, and the IoU between the inferred cloud mask and the true value is calculated for each epoch. The model with the largest IoU on the validation set is saved as the best model.

[0122] Step S2, cloud detection application stage:

[0123] Step S201: Constructing cloud-free basemaps for multiple seasons and multiple geographic areas:

[0124] A multi-season, large-area cloud-free basemap is constructed through steps such as regional block adjustment, registration fusion, orthophotography, mosaicking, and color uniformity.

[0125] Step S202: Determine the geographical scope of the image to be inspected:

[0126] After orthophotoing the image to be detected, the base map of the corresponding range is determined according to the coordinates of the four corner points, and expanded by 100 meters on all sides. This is to avoid the image missing the corresponding base map range after registration, thereby constructing a cloud-free base map of a single geographical range in four seasons.

[0127] Step S203, determining the season of the image to be inspected:

[0128] The cloud-free basemaps for a single geographic range in the four seasons determined in the previous step contain basemaps for four different seasons. You need to determine the specific season of the image to be inspected and select the basemap for the corresponding season.

[0129] The specific method is: calculate the NDVI mean of the image to be tested, compare it with the NDVI mean of the four base maps, select the season with the most similar NDVI value, and construct a cloud-free base map of a single season and a single geographical range.

[0130] Step S204: Registration of time-series images:

[0131] After determining the season of the image to be inspected, the cloud-free base map corresponding to a single season and a single geographical range is registered with the image to be inspected using SIFT-based feature matching.

[0132] The specific process is as follows: the SIFT feature points of the image to be inspected and the cloud-free base map of a single season and a single geographical range are detected respectively. For each feature point, a 128-dimensional feature descriptor is generated, and the K-nearest neighbor matching method is used to determine the most similar feature point pairs in the two images. The RANSAC algorithm is used to remove incorrect matching point pairs, and based on the correct matching point pairs after screening, the geometric transformation matrix between the two images is estimated. The transformation matrix is ​​applied to the image to be inspected, so that the image to be inspected and the cloud-free base map of a single season and a single geographical range can be geometrically aligned to form a time series image.

[0133] Step S3: Quickly determine whether there are clouds in the time series images.

[0134] Step S4, cloud range detection:

[0135] The two images are combined into an 8-band multi-temporal image and input into AttUNet for the final cloud range detection, and the cloud mask range is output.

[0136] In summary, this embodiment provides an efficient, accurate, and robust cloud detection technology that aims to use historical cloud-free imagery as a reference basemap to improve the accuracy and robustness of cloud detection, thereby solving a series of key problems facing the existing cloud detection field.

[0137] Implementation method 2: A remote sensing image cloud detection system based on a reference base map described in this implementation method includes the following modules:

[0138] Module S1 builds a dataset and a cloud detection model, and trains the cloud detection model based on the dataset:

[0139] Module S2 constructs a cloud-free basemap for a single season and a single geographic area, and registers it with the photography to be detected to form a time series image:

[0140] Module S3 determines whether there are clouds in the time series image. If there are no clouds, a blank cloud mask is directly output. If there are clouds, the operation of module S4 is executed.

[0141] Module S4 inputs the cloud detection model trained by module S1 into the temporal impact of clouds, and then outputs a cloud mask.

[0142] Embodiment 3: An electronic device described in this embodiment includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0143] Memory for storing computer programs;

[0144] The processor is configured to implement the method steps described in the first embodiment when executing the program stored in the memory.

[0145] Implementation method 4: This implementation method describes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in implementation method 1 are implemented.

[0146] The above is a detailed introduction to the remote sensing image cloud detection method, system, device and storage medium based on a reference base map proposed in the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A remote sensing image cloud detection method based on a reference base map, characterized in that: The following steps are involved: Step S1, constructing a data set and a cloud detection model, and training the cloud detection model based on the data set: Step S2, construct a cloud-free base map of a single season and a single geographical range, and align it with the photography to be detected to form a time series image: Step S3, determining whether there are clouds in the time series image, if there are no clouds, directly outputting a blank cloud mask, if there are clouds, executing the operation of step S4; In step S4, the cloud-related temporal impact is input into the cloud detection model trained in step S1 for detection, and a cloud mask is output.

2. The remote sensing image cloud detection method based on a reference base map according to claim 1, characterized in that: In the step S2, the construction of a cloud-free base map of a single season and a single geographical range includes the following steps: Step S201, constructing historical data into cloud-free base maps of multiple geographical ranges in four seasons; Step S202, determining the geographical range of the photography to be detected, selecting the photography with the same geographical range as the photography to be detected from the cloud-free base maps of multiple geographical ranges in four seasons, thereby constructing a cloud-free base map of a single geographical range in four seasons; Step S203, determining the season of the photography to be detected, selecting the photography of the same season as the photography to be detected from the cloud-free base maps of a single geographical range in four seasons, thereby constructing a cloud-free base map of a single geographical range in a single season.

3. The remote sensing image cloud detection method based on a reference base map according to claim 2, characterized in that: In the step S203, the season of the photography to be detected is determined as follows: The season of the photography to be detected is determined based on the global NDVI mean of the photography to be detected.

4. The remote sensing image cloud detection method based on a reference base map according to claim 1, characterized in that: In the step S3, the step of determining whether there is cloud in the time series image comprises the following steps: Step S301, calculating a difference map based on the photography to be detected and a cloud-free base map of a single geographical range in a single season; Step S302, calculating a normalized difference index based on the difference map; Step S303, setting a first threshold based on the normalized difference index calculated in step S302, and calculating the number of pixels in the difference map that exceed the first threshold; Step S304, setting a second threshold, and determining whether there are clouds in the time series image according to the relationship between the second threshold and the number of pixels.

5. The remote sensing image cloud detection method based on a reference base map according to claim 4, characterized in that: In the step S301, the difference map is calculated based on the photography to be detected and the cloud-free base map of a single season and a single geographical range, specifically: Diff = Image2 - Image1; Among them, Diff is the difference map, Image2 is the photography to be detected, and Image1 is the cloud-free base map of a single geographical range in a single season.

6. The remote sensing image cloud detection method based on a reference base map according to claim 4, characterized in that: In step S303, the number of pixels in the difference map exceeding the first threshold is calculated as follows: Where M is the number of pixels, T1 is the threshold, NDI is the normalized difference index, and NDI is the i is the normalized difference index of each pixel in the difference map, and N is the total number of pixels in the difference map.

7. The remote sensing image cloud detection method based on a reference base map according to claim 4, characterized in that: In step S304, the second threshold is set, and whether the time series image has clouds is determined according to the relationship between the second threshold and the number of pixels, specifically: A second threshold is set. When the second threshold is greater than the number of pixels, the time series image is determined to be cloudless. When the second threshold is less than the number of pixels, the time series image is determined to be clouded.

8. A remote sensing image cloud detection system based on a reference base map, characterized in that: Includes the following modules: Module S1, builds a data set and a cloud detection model, and trains the cloud detection model based on the data set: Module S2 constructs a cloud-free basemap of a single season and a single geographic range, and aligns it with the photography to be detected to form a time series image: Module S3 determines whether there are clouds in the time series image. If there are no clouds, a blank cloud mask is directly output. If there are clouds, the operation of module S4 is executed. Module S4 inputs the cloud detection model trained by module S1 with the temporal impact of clouds for detection and outputs a cloud mask.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing the method steps described in any one of claims 1 to 7 when executing a program stored in a memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 7 are implemented.

Images