Remote sensing image generation method and system capable of controlling cloud cover, and storage medium

Abstract

The invention discloses a cloud cover-controllable remote sensing image generation method and system and a storage medium, and belongs to the technical field of remote sensing image simulation. The method comprises the following steps: constructing cloudless and clouded remote sensing image training sets, and storing a cloud cover proportion in a clouded image file name; constructing a conditional generative adversarial network which comprises generators G and F and discriminators Dx and Dy; using sine-cosine position coding to convert the cloud amount value into a vector, and fusing the vector with image features; training a generator by adding a cloud cover constrained cyclic consistency loss function; and finally, inputting the proportion of the cloudless image to the target cloud cover, and generating a remote sensing image with the specified cloud cover. The method achieves the precise control of the cloud cover, is high in fidelity of the generated image, and is suitable for various scenes, such as remote sensing image sample enhancement and target detection model training.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image simulation technology, specifically to a method, system, and storage medium for generating remote sensing images with controllable cloud cover. Background Technology

[0002] Remote sensing images, as a special type of image, are often affected by clouds during acquisition. Cloud cover is one of the most significant factors affecting the quality and usability of optical remote sensing images. The severe impact of cloud cover on optical remote sensing images during acquisition leads to decreased image quality, obscuring ground feature information, or spectral distortion, limiting their effective application in surveying, environmental monitoring, and resource investigation. Large areas of land surface are constantly disturbed by clouds, and a significant amount of remote sensing data is unusable due to cloud contamination. Therefore, automatic cloud removal has become a research hotspot. The goal of cloud removal is to obtain clear remote sensing images. However, in target detection within remote sensing images, to improve the model's detection capabilities under cloud conditions, it is necessary to consider the impact of clouds on target detection. This involves adding clouds of varying degrees to the training set to expand the number of cloud-covered samples. Therefore, automatically adding clouds to remote sensing images and controlling the proportion of cloud pixels in sample images is of great significance.

[0003] Current research on cloud processing in remote sensing images mainly focuses on cloud detection, cloud removal, and cloud shadow recognition. For example, Chinese invention patent application CN119048915A (published on November 29, 2024) discloses a method for constructing a thin cloud deterioration dataset based on coupled learning. This method mainly includes three stages: data-driven cloud space simulation, cloud band synthesis guided by scattering laws, and dataset construction based on imaging mechanisms. The technical solution of this patent can provide a massive data foundation for training remote sensing image thin cloud correction (or detection) networks, promote the development of related algorithms, and improve the practical utilization rate of remote sensing data; however, it mainly focuses on thin cloud simulation and does not involve precise control of cloud cover.

[0004] Chinese invention patent application CN118015461A (published on May 10, 2024) provides a method, system, and storage medium for cloud shadow recognition in high-resolution optical remote sensing images. The method includes: acquiring high-resolution optical remote sensing data and calculating on-satellite reflectance; extracting low-value target areas based on near-infrared on-satellite reflectance threshold conditions; iteratively calculating shadow areas from the near-infrared on-satellite reflectance data; calculating the difference image between the iterated image and the initial image; and extracting shadow areas based on the difference image and threshold conditions; calculating a set of potential cloud shadow areas by combining cloud pixel positions, remote sensing image imaging geometric parameters, and cloud height range; calculating the spatial overlap area between each element in the potential cloud shadow set and the shadow area; selecting the element with the largest spatial overlap area as the main cloud shadow; and extracting the final cloud shadow based on the main shadow and its buffer zone, combined with the near-infrared shadow area. This invention improves automation, reduces manual intervention, and provides high accuracy in extraction results, enabling rapid identification of cloud shadows in high-resolution remote sensing images; however, its focus is on shadow recognition rather than image generation with controllable cloud cover.

[0005] In summary, while existing technologies in image generation, particularly image-to-image transformation models (such as Pix2Pix and CycleGAN), perform well in style transfer and image inpainting, research on controlling cloud cover ratios in remote sensing images remains lacking. Due to the complex surface background of remote sensing images, accurately controlling the cloud pixel ratio during generation while maintaining image realism and consistency with ground features is a current technological challenge. Summary of the Invention

[0006] To address the above technical problems, this invention provides a method for generating remote sensing images with controllable cloud cover, comprising the following steps: Step S1: Construct training dataset: Obtain training set of cloudless and cloudy remote sensing images, calculate the cloud cover ratio of each cloudy image, and store the ratio value in the file name of the cloudy image for easy retrieval. Step S2: Construct a conditional generative adversarial network: The conditional generative adversarial network includes a generator G, a generator F, a discriminator Dx, and a discriminator Dy. The discriminator Dx is used to determine whether the input image is cloudless, and the discriminator Dy is used to determine whether the input image is cloudy. The generator G generates a cloudy image based on the input cloudless image, and the generator F generates a cloudless image based on the input cloudy image. Step S3, Cloud cover conditional coding: The cloud cover value Cr is converted into a vector using sine-cosine position coding, and the vector is fused with image features; Step S4: Forward computation; Step S5: Image generation: Input the ratio of cloudless image to target cloud cover to generate a remote sensing image with the specified cloud cover.

[0007] Furthermore, in step S2, generators G and F are trained under the guidance of discriminators Dx and Dy, respectively.

[0008] Furthermore, the discriminators Dx and Dy use a convolutional neural network as the backbone network and employ multiple convolutional layers to process the input image. Each layer uses multiple convolutional kernels to extract semantic features from the image and simultaneously downsamples the image. The function of the discriminators Dx and Dy is to discriminate between the sample image and the image output by the generator.

[0009] Furthermore, the conditional generative adversarial network introduces a function to constrain the generator. This function is a joint function of cycle consistency loss with cloud cover constraints, and its formula is: , Where Cr is the cloud cover value, x is the cloudless image, and y is the cloudy image; The cloudless image x is fused with the cloud cover value Cr and then input into the generator G to obtain a simulated image. The simulated image is then fused with the cloud cover value Cr again and input into the discriminator Dy. During training, the discriminator Dy is used to distinguish between cloudless and cloudy images after cloud cover constraints are added. It is judged as 1 only when there are obvious cloud pixels in the image and the cloud cover value is correct. The discriminator Dy obtained by using the cloud cover value as a condition can be used to guide the generator G, so that the generator can generate images with specific cloud cover values.

[0010] Furthermore, step S3 includes the following steps: Step S301: Calculate different frequencies according to the set range. The formula for the frequency is: , Where i is an integer value in the length setting, ranging from [0, dim], dim is the input encoding length, and max_period is the minimum embedding frequency value; Step S302: Multiply the integer value i by the frequency to calculate the sine and cosine values; Step S303: The cloud cover value Cr is in the range of [0,1]. Set the length L of the location code, multiply the cloud cover value by L to get an integer between [0,L], and then encode it to obtain a vector. Step S304: Superimpose the encoded vector with the input data, and also add the vector as the input to generator G and generator F.

[0011] Furthermore, in step 304, when the encoded vector is superimposed on the input data, the dimension of the encoded data is first transformed to be consistent with the dimension of the feature image through a fully connected transformation according to the number of channels of the feature image, and then superimposed on the feature image for forward operation.

[0012] Furthermore, the forward operation in step S4 includes: Input the cloudless image x and the randomly generated cloud cover value into the generator G to obtain the simulated image G(x); The cloud image y is input into the generator F to obtain the simulated image F(y); Input the simulated image G(x) into F to obtain the restored image F(G(x)); Input the simulated image F(y) and the cloud cover value in the y image into G to obtain the restored image G(F(y)); Input x and F(y) into the discriminator Dx for forward operation; Input y and the cloud cover value into Dy, and also input G(x) and the randomly generated cloud cover value into Dy for forward computation of the discriminator.

[0013] Furthermore, the forward operation method is as follows: a set number of random image samples are read from the cloudless and cloudy images in the training dataset, and the cloud cover value in the image sample file name is read at the same time. Finally, the image data and the corresponding cloud cover value are passed to the conditional generative adversarial network.

[0014] This invention provides a remote sensing image generation system with controllable cloud cover, which employs a memory and a controller. The memory stores a computer-readable program, which, when invoked by the controller, can execute the steps of the remote sensing image generation method with controllable cloud cover described in this invention.

[0015] The present invention provides a medium having a computer-readable program stored in its memory. When the computer-readable program is invoked by a controller, it can execute the steps of the remote sensing image generation method for controllable cloud cover as described in the present invention.

[0016] Compared with existing technologies, the advantages and effects of this application are as follows: 1. This application achieves precise and controllable generation of cloud cover. By introducing cloud cover conditional coding and conditional discriminator, the generator can generate remote sensing images with corresponding cloud coverage based on the input cloud cover ratio, filling the gap in the existing technology for quantitative cloud generation.

[0017] 2. This application achieves high image fidelity. By adopting the cycle consistency loss function and adversarial training mechanism, it ensures that key information such as surface texture and spectral features are preserved while controlling cloud cover, thereby improving the practicality and realism of the generated images.

[0018] 3. This application is more adaptable and applicable to various cloud cover levels (such as 0%, 25%, 50%, 75%, 100%), and can be widely used in scenarios such as remote sensing image data enhancement, target detection model training, and cloud removal algorithm evaluation.

[0019] 4. The training efficiency of this application is greatly improved. By embedding the cloud cover ratio into the file name, it is easy to quickly read and input conditions during the training process, thereby improving the efficiency and reproducibility of model training.

[0020] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0021] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0023] in: Figure 1 This is a flowchart of a remote sensing image generation method for controllable cloud cover according to this application; Figure 2 This is an overall schematic diagram of a remote sensing image generation method for controllable cloud cover according to this application; Figure 3 This is a generator network structure diagram of the remote sensing image generation method with controllable cloud cover proposed in this application; Figure 4 This is a discriminator network structure diagram of a remote sensing image generation method for controllable cloud cover according to this application; Figure 5 This is a schematic diagram of cloud cover sample for a remote sensing image generation method for controllable cloud cover according to this application; Figure 6 This is a schematic diagram of a cloudless sample for a remote sensing image generation method with controllable cloud cover according to this application. Figure 7 This diagram illustrates the generation effect of setting different cloud cover values ​​for a remote sensing image generation method with controllable cloud cover according to this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0025] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0026] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0027] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it have an "or" relationship.

[0028] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0029] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0030] Example 1 This embodiment introduces a method for generating remote sensing images with controllable cloud cover.

[0031] Please refer to Figure 1-2 As shown, Figure 1 This is a flowchart of a remote sensing image generation method for controllable cloud cover according to this application; Figure 2 This is an overall schematic diagram of a remote sensing image generation method for controllable cloud cover according to this application.

[0032] A method for generating remote sensing images with controllable cloud cover includes the following steps: Step S1: Construct training dataset: Obtain training set of cloudless and cloudy remote sensing images, calculate the cloud cover ratio of each cloudy image, and store the ratio value in the file name of the cloudy image for easy retrieval. Step S2: Construct a conditional generative adversarial network: The conditional generative adversarial network includes a generator G, a generator F, a discriminator Dx, and a discriminator Dy. The discriminator Dx is used to determine whether the input image is cloudless, and the discriminator Dy is used to determine whether the input image is cloudy. The generator G generates a cloudy image based on the input cloudless image, and the generator F generates a cloudless image based on the input cloudy image. Step S3, Cloud cover conditional coding: The cloud cover value Cr is converted into a vector using sine-cosine position coding, and the vector is fused with image features; Step S4: Forward computation; Step S5: Image generation: Input the ratio of cloudless image to target cloud cover to generate a remote sensing image with the specified cloud cover.

[0033] The technical advantages of this embodiment are: it provides a complete and controllable cloud cover remote sensing image generation process with a clear structure that is easy to implement and deploy. It achieves style transfer from cloudless to cloud-covered images through a conditional generative adversarial network, demonstrating good scalability.

[0034] Example 2 Based on Example 1, this example discloses a further design of a remote sensing image generation method with controllable cloud cover.

[0035] Furthermore, in step S2, generators G and F are trained under the guidance of discriminators Dx and Dy, respectively.

[0036] Furthermore, the discriminators Dx and Dy use a convolutional neural network as the backbone network and employ multiple convolutional layers to process the input image. Each layer uses multiple convolutional kernels to extract semantic features from the image and simultaneously downsamples the image. The function of the discriminators Dx and Dy is to discriminate between the sample image and the image output by the generator.

[0037] Furthermore, the conditional generative adversarial network introduces a function to constrain the generator. This function is a joint function of cycle consistency loss with cloud cover constraints, and its formula is: , Where Cr is the cloud cover value, x is the cloudless image, and y is the cloudy image; The cloudless image x is fused with the cloud cover value Cr and then input into the generator G to obtain a simulated image. The simulated image is then fused with the cloud cover value Cr again and input into the discriminator Dy. During training, the discriminator Dy is used to distinguish between cloudless and cloudy images after cloud cover constraints are added. It is judged as 1 only when there are obvious cloud pixels in the image and the cloud cover value is correct. The discriminator Dy obtained by using the cloud cover value as a condition can be used to guide the generator G, so that the generator can generate images with specific cloud cover values.

[0038] Furthermore, step S3 includes the following steps: Step S301: Calculate different frequencies according to the set range. The formula for the frequency is: , Where i is an integer value in the length setting, ranging from [0, dim], dim is the input encoding length, and max_period is the minimum embedding frequency value; Step S302: Multiply the integer value i by the frequency to calculate the sine and cosine values; Step S303: The cloud cover value Cr is in the range of [0,1]. Set the length L of the location code, multiply the cloud cover value by L to get an integer between [0,L], and then encode it to obtain a vector. Step S304: Superimpose the encoded vector with the input data, and also add the vector as the input to generator G and generator F.

[0039] Furthermore, in step 304, when the encoded vector is superimposed on the input data, the dimension of the encoded data is first transformed to be consistent with the dimension of the feature image through a fully connected transformation according to the number of channels of the feature image, and then superimposed on the feature image for forward operation.

[0040] Furthermore, the forward operation in step S4 includes: Input the cloudless image x and the randomly generated cloud cover value into the generator G to obtain the simulated image G(x); The cloud image y is input into the generator F to obtain the simulated image F(y); Input the simulated image G(x) into F to obtain the restored image F(G(x)); Input the simulated image F(y) and the cloud cover value in the y image into G to obtain the restored image G(F(y)); Input x and F(y) into the discriminator Dx for forward operation; Input y and the cloud cover value into Dy, and also input G(x) and the randomly generated cloud cover value into Dy for forward computation of the discriminator.

[0041] Furthermore, the forward operation method is as follows: a set number of random image samples are read from the cloudless and cloudy images in the training dataset, and the cloud cover value in the image sample file name is read at the same time. Finally, the image data and the corresponding cloud cover value are passed to the conditional generative adversarial network.

[0042] The technical advantages of this embodiment are as follows: The introduction of sine-cosine positional encoding converts cloud cover values ​​into high-dimensional vectors, enhancing the model's ability to perceive continuous cloud cover conditions. A cyclic consistency loss function with cloud cover constraints is designed, effectively improving the accuracy and consistency of generated images in terms of cloud cover control.

[0043] Example 3 Based on Embodiment 1 or 2, this embodiment discloses a further design of the generator and discriminator for a remote sensing image generation method with controllable cloud cover.

[0044] Please refer to Figure 3 As shown, Figure 3 This diagram illustrates the generator network structure of a remote sensing image generation method for controllable cloud cover according to this application. The generator is designed using a combination of conventional convolutional and residual networks, and the encoded result is superimposed with the results of key steps in the network. The network models of generators G and F in this invention are completely identical. The purpose of training is to change the parameter values ​​of the network model according to different input samples, thus achieving different generation effects. The superposition operation is performed at the network structure level to improve the network structure, and the superimposed vectors are also completely identical.

[0045] Please refer to Figure 4 As shown, Figure 4 This is a diagram of the discriminator network structure of a remote sensing image generation method for controllable cloud cover according to this application. The discriminator's function is to identify the sample image and the image output by the generator. The discriminator of this invention uses a convolutional neural network as the backbone network, employing multiple convolutional layers to process the input image. Each layer uses multiple convolutional kernels to extract semantic features from the image and simultaneously downsamples the image.

[0046] The technical advantages of this embodiment are as follows: The generator employs a combination of convolutional and residual networks, along with positional encoding overlay, enhancing feature fusion capabilities. The discriminator, built upon a convolutional neural network, possesses strong image semantic feature extraction and discrimination capabilities, improving the stability of adversarial training.

[0047] Example 4 Based on Example 3, this example discloses the effect verification of a remote sensing image generation method with controllable cloud cover.

[0048] (1) Generation of sample data Using properly labeled GF1 remote sensing imagery, the images were cropped to 128x128 pixels, and the cloud cover ratio in the cropped images was calculated based on manually labeled cloud mask files. For ease of sample storage, the cloud cover ratio was stored in the filename; the cloud pixel ratio can be directly extracted from the sample filename during subsequent training. Please refer to [reference needed]. Figure 5-6 , Figure 5 This is a schematic diagram of cloud cover sample for a remote sensing image generation method for controllable cloud cover according to this application; Figure 6 This is a schematic diagram of a cloudless sample for a remote sensing image generation method with controllable cloud cover according to this application.

[0049] (2) Model forward computation During the forward operation, a set number of random image samples are read from both cloudless and cloudy images in the dataset, and the cloud cover value in the image sample file name is read simultaneously. Finally, the image data and the corresponding cloud cover value are simultaneously passed to the network architecture of this invention. The forward operation includes: inputting the cloudless image x and the randomly generated cloud cover value into the generator G to obtain the simulated image G(x); inputting the cloudy image y into the generator F to obtain the simulated image F(y); inputting the simulated image G(x) into F to obtain the restored image F(G(x)); inputting the simulated image F(y) and the cloud cover value in the y image into G to obtain the restored image G(F(y)); inputting x and F(y) into the discriminator Dx for forward operation; inputting y and the cloud cover value into Dy, and also inputting G(x) and the randomly generated cloud cover value into Dy for forward calculation of the discriminator.

[0050] (3) Model parameter optimization This invention comprises four models: two discriminators, Dx and Dy, and two generators, G and F. Discriminator Dx determines whether the input image is cloudless, while discriminator Dy determines whether the input image is cloudy. Generator G generates a cloudy image from a cloudless input image, and generator F generates a cloudless image from a cloudy input image. Generators G and F aim to preserve as much texture information as possible, making the output image similar to the input image. Model parameter optimization is achieved through training, with the goal of minimizing the loss function. Both the generators and discriminators in this invention employ the adversarial loss function from adversarial networks during training.

[0051] (4) Generate remote sensing images based on cloud cover values. Four scenarios were set up: 0% cloud cover (no clouds), 25% (low cloud cover), 50% (moderate cloud cover), 75% (medium to high cloud cover), and 100% (full cloud cover). The corresponding remote sensing images were generated using the model. Please refer to [reference needed]. Figure 7 , Figure 7 This diagram illustrates the generation effect of a remote sensing image generation method with controllable cloud cover according to this application, using different cloud cover values. The results show that the model can generate remote sensing images that meet the expected cloud cover requirements and maintain high image quality under different cloud cover levels.

[0052] The technical effects of this embodiment are as follows: Through generation experiments with multiple cloud cover ratios (0%~100%), the generation effect and image quality of the model under different cloud cover conditions were verified. The results show that the images generated by this invention are superior to existing methods in terms of visual quality and cloud cover control accuracy, and have high practical value.

[0053] The above description is merely a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. Various modifications and variations are possible with respect to the present invention. Any changes, modifications, substitutions, integrations, and parameter alterations to these embodiments within the spirit and principles of the present invention fall within the scope of protection of the claims of the present invention.

Claims

1. A method for generating remote sensing images with controllable cloud cover, characterized in that, Includes the following steps: Step S1: Construct training dataset: Obtain training set of cloudless and cloudy remote sensing images, calculate the cloud cover ratio of each cloudy image, and store the ratio value in the file name of the cloudy image for easy retrieval. Step S2: Construct a conditional generative adversarial network: The conditional generative adversarial network includes a generator G, a generator F, a discriminator Dx, and a discriminator Dy. The discriminator Dx is used to determine whether the input image is cloudless, and the discriminator Dy is used to determine whether the input image is cloudy. The generator G generates a cloudy image based on the input cloudless image, and the generator F generates a cloudless image based on the input cloudy image. Step S3, Cloud cover conditional coding: The cloud cover value Cr is converted into a vector using sine-cosine position coding, and the vector is fused with image features; Step S4: Forward computation; Step S5, Image Generation: Input the ratio of cloudless image to target cloud cover to generate a remote sensing image with the specified cloud cover.

2. The remote sensing image generation method for controllable cloud cover according to claim 1, characterized in that, In step S2, generators G and F are trained under the guidance of discriminators Dx and Dy, respectively.

3. A method for generating remote sensing images with controllable cloud cover according to claim 1 or 2, characterized in that, The discriminators Dx and Dy use a convolutional neural network as the backbone network and employ multiple convolutional layers to process the input image. Each layer uses multiple convolutional kernels to extract semantic features from the image and simultaneously downsamples the image. The function of the discriminators Dx and Dy is to distinguish between the sample image and the image output by the generator.

4. The remote sensing image generation method for controllable cloud cover according to claim 3, characterized in that, The conditional generative adversarial network introduces a function to constrain the generator. This function is a joint function of cycle consistency loss with cloud cover constraints, and its formula is as follows: min‖xF(G(x,Cr))‖+‖yG(F(y),Cr)‖, Where Cr is the cloud cover value, x is the cloudless image, and y is the cloudy image; The cloudless image x is fused with the cloud cover value Cr and then input into the generator G to obtain a simulated image. The simulated image is then fused with the cloud cover value Cr again and input into the discriminator Dy. During training, the discriminator Dy is used to distinguish between cloudless and cloudy images after cloud cover constraints are added. It is judged as 1 only when there are obvious cloud pixels in the image and the cloud cover value is correct. The discriminator Dy obtained by using the cloud cover value as a condition can be used to guide the generator G, so that the generator can generate images with specific cloud cover values.

5. The remote sensing image generation method for controllable cloud cover according to claim 1, characterized in that, Step S3 includes the following steps: Step S301: Calculate different frequencies according to the set range. The formula for the frequency is: ， Where i is an integer value in the length setting, ranging from [0, dim], dim is the input encoding length, and max_period is the minimum embedding frequency value; Step S302: Multiply the integer value i by the frequency to calculate the sine and cosine values; Step S303: The cloud cover value Cr is in the range of [0,1]. Set the length L of the location code, multiply the cloud cover value by L to get an integer between [0,L], and then encode it to obtain a vector. Step S304: Superimpose the encoded vector with the input data and use it as the input to generator G and generator F. At the same time, add the vector to the input of discriminator Dx and discriminator Dy.

6. The remote sensing image generation method for controllable cloud cover according to claim 5, characterized in that, In step 304, when the encoded vector is superimposed on the input data, the dimension of the encoded data is first transformed to be consistent with the dimension of the feature image through a fully connected transformation according to the number of channels of the feature image, and then superimposed on the feature image for forward operation.

7. The remote sensing image generation method for controllable cloud cover according to claim 6, characterized in that, The forward operation in step S4 includes: Input the cloudless image x and the randomly generated cloud cover value into the generator G to obtain the simulated image G(x); The cloud image y is input into the generator F to obtain the simulated image F(y); Input the simulated image G(x) into F to obtain the restored image F(G(x)); Input the simulated image F(y) and the cloud cover value in the y image into G to obtain the restored image G(F(y)); Input x and F(y) into the discriminator Dx for forward operation; Input y and the cloud cover value into Dy, and also input G(x) and the randomly generated cloud cover value into Dy for forward computation of the discriminator.

8. The remote sensing image generation method for controllable cloud cover according to claim 7, characterized in that, The forward operation method is as follows: a set number of random image samples are read from the cloudless and cloudy images in the training dataset, and the cloud cover value in the image sample file name is read at the same time. Finally, the image data and the corresponding cloud cover value are passed to the conditional generative adversarial network.

9. A remote sensing image generation system with controllable cloud cover, characterized in that, The method employs a memory and a controller, wherein the memory stores a computer-readable program, which, when invoked by the controller, can execute the steps of a remote sensing image generation method for controllable cloud cover as described in any one of claims 1-8.

10. A medium, characterized in that, Its memory contains a computer-readable program, which, when invoked by the controller, can execute the steps of a remote sensing image generation method for controllable cloud cover as described in any one of claims 1-8.

Images