Mass remote sensing image super-resolution reconstruction method and device based on Ray cluster and medium

Abstract

The invention discloses a mass remote sensing image super-resolution reconstruction method and device based on a Ray cluster and a medium, and relates to the technical field of super-resolution reconstruction of super-wide remote sensing images, and the method comprises the steps: carrying out the blocking processing of large-wide remote sensing data; a super-resolution remote sensing RGB image is reconstructed and obtained by using the parallel computing capability of a Ray cluster GPU; identifying a nodata area of the large-breadth remote sensing data, and accurately matching a Mask layer; and taking the Mask image layer as an Alpha channel, and fusing the Mask image layer with the RGB image to obtain an RGBA image. The method supports a super-resolution reconstruction task of a single 50GB remote sensing image, a high-efficiency super-resolution reconstruction task of mass data, artifact elimination after super-resolution, and intelligent automatic scheduling, realizes high-efficiency and high-quality super-resolution reconstruction of super-large scale remote sensing data, breaks through the technical bottleneck of a traditional method, and improves the super-resolution reconstruction efficiency. And powerful support is provided for high-precision application of remote sensing images.

Description

Technical Field

[0001] The present invention relates to the technical field of ultra-wideband remote sensing image super-resolution reconstruction, and in particular to a Ray cluster-based method, device and medium for super-resolution reconstruction of massive remote sensing images. Background Art

[0002] With the rapid development of modern science and technology, remote sensing images have become an important means of earth observation and analysis; their wide application covers many fields, such as environmental monitoring, agricultural management, urban planning, disaster warning, land and resources survey, etc. These applications not only improve the efficiency and accuracy of various industries, but also provide more detailed and accurate data support for decision makers. For example, in environmental monitoring, remote sensing images can track and evaluate the changes in natural resources in real time, helping to formulate scientific management and protection strategies; in the agricultural field, by analyzing remote sensing images, the growth status of crops can be accurately evaluated and agricultural production can be optimized; in urban planning, remote sensing images provide detailed geographic information, which helps to rationally layout urban functional areas and improve the level of urban management.

[0003] However, the application of remote sensing images also faces some technical bottlenecks, especially the problem of image blur. Since the acquisition of remote sensing images is limited by factors such as sensor resolution, imaging conditions and atmospheric influence, the original images often have the defect of insufficient resolution. This blur not only affects the clarity and detail recognition of the image, but also limits the effect of the image in high-precision analysis and application. Improving the resolution of remote sensing images has become a problem that needs to be solved urgently.

[0004] In order to solve this problem, the super-resolution reconstruction (SR) technology of remote sensing images came into being. This technology uses deep learning and image processing algorithms to reconstruct low-resolution images into high-resolution images, thereby improving the details and clarity of the images. The application of this technology has greatly expanded the scope and effect of remote sensing images. For example, through super-resolution reconstruction, low-resolution satellite images can be converted into high-resolution images, making the details clearer and facilitating refined analysis and application. In environmental monitoring, super-resolution images can more accurately monitor and evaluate environmental changes; in urban planning, high-resolution images help to analyze urban structure and layout in more detail and improve the scientific nature of planning.

[0005] However, although super-resolution technology performs well in improving the resolution of remote sensing images, it still has some limitations in practical applications.

[0006] First, common remote sensing image super-resolution technology can usually only process a single small-sized image and cannot effectively process TIFF data with a larger width. For TIFF images with a single data size of more than 50GB, due to the out of memory (OOM) problem during GPU inference, it is difficult for existing technologies to achieve effective super-resolution processing. This problem makes super-resolution technology unable to be applied when processing large-width remote sensing images.

[0007] Secondly, for the application of the Geographic Information System (GIS) industry, the demand for super-resolution reconstruction is more extensive and complex. The GIS industry needs to perform super-resolution processing on data from the entire province, the entire country, and even the entire world. Taking the national data volume as an example, there are more than 700 pieces of ultra-wide TIFF data in key cities, and the total amount of data is tens of TB. When processing such a large amount of data, super-resolution technology is not only time-consuming, but also strongly dependent on the reasoning ability of the GPU. Due to the inefficiency of existing technologies in processing massive data and insufficient GPU resources, the large-scale application of super-resolution technology in the GIS industry is difficult to achieve.

[0008] Therefore, although super-resolution technology has achieved remarkable results in improving the resolution of remote sensing images, it still faces major challenges in processing wide-width images and massive data. These limitations hinder the widespread promotion and implementation of super-resolution technology in practical applications.

[0009] The invention application with application number 202211183041.X discloses a remote sensing monitoring and identification system based on high-resolution reconstruction of remote sensing images. The application scheme realizes the timely acquisition and accurate monitoring, identification and early warning of the spatial distribution and development trend of crop diseases and insect pests in crop fields. However, the scheme also has the problem of being unable to efficiently reconstruct ultra-wide remote sensing images with super-resolution.

[0010] Therefore, in practical applications, a super-resolution reconstruction method for massive remote sensing images based on Ray clusters is needed to break through the bottleneck of existing technologies, achieve efficient super-resolution reconstruction of large-width and massive remote sensing images, and provide more powerful technical support for the application of remote sensing images. Summary of the invention

[0011] In view of the above-mentioned problems, the purpose of the present invention is to provide a method, device and medium for super-resolution reconstruction of massive remote sensing images based on Ray cluster, so as to realize efficient super-resolution reconstruction of massive ultra-wide-width remote sensing images.

[0012] Embodiments of the present invention provide a Ray cluster-based super-resolution reconstruction method, device and medium for massive remote sensing images.

[0013] The first aspect: A method for super-resolution reconstruction of massive remote sensing images based on Ray clusters, including:

[0014] S1, block processing of large-width remote sensing data;

[0015] S2, using the parallel computing power of the Ray cluster GPU to reconstruct and obtain super-resolution remote sensing RGB images;

[0016] S3, identify the nodata area of ​​large-width remote sensing data and accurately match the Mask layer;

[0017] S4. Use the Mask layer as an Alpha channel to merge with the RGB image to obtain an RGBA image.

[0018] Furthermore, the block processing of the wide-width remote sensing data in S1 includes the following steps:

[0019] S11, analyzing the original wide-width remote sensing data to determine the optimal data block size;

[0020] S12. Using the sliding window technology, a series of data blocks are evenly cut out along the length and width directions of the image according to the optimal data block size.

[0021] Furthermore, the reconstructing and acquiring the super-resolution remote sensing RGB image in S2 includes:

[0022] Overlapping areas are set between adjacent data blocks. During image reconstruction, statistical methods are used to smoothly transition these overlapping areas to achieve seamless connection and visual continuity of the reconstructed image.

[0023] Furthermore, the accurate matching of the Mask layer in S3 includes:

[0024] An adaptive upsampling algorithm, an edge-preserving algorithm, and a detail enhancement algorithm are used for the Mask layer to achieve accurate size matching between the Mask layer and the super-resolution RGB image.

[0025] Furthermore, the precise matching includes complete transfer of boundary information and precise alignment at the pixel level.

[0026] Furthermore, the S2 utilizes the parallel computing capability of the Ray cluster GPU to reconstruct and obtain a super-resolution remote sensing RGB image, including the steps of:

[0027] S21. Use the Ray framework to build a distributed GPU cluster executor pool, and finely split the GPU cluster executor pool into multiple sub-GPU executors;

[0028] S22, dividing the task pool of reconstructing super-resolution remote sensing RGB images into multiple task execution actors;

[0029] S23, intelligently scheduling the sub-GPU executor to execute according to each execution task actor.

[0030] Furthermore, the task execution actor adopts a resource limitation strategy, including: setting upper limits on GPU, CPU and memory resource usage for each task execution actor.

[0031] Furthermore, when the super-resolution remote sensing RGB image is reconstructed and acquired, a resource and error management mechanism is used to automatically monitor abnormal conditions during the task processing to maintain data integrity.

[0032] A second aspect: An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method provided in the first aspect when executing the program.

[0033] A third aspect: A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the first aspect.

[0034] Beneficial effects of the present invention:

[0035] 1. Aiming at the challenges of memory constraints, low efficiency and boundary effects caused by nodata characteristics faced by super-resolution reconstruction of ultra-wide TIFF remote sensing images, the present invention proposes a super-resolution remote sensing reconstruction method and system, which has the function of efficient and reliable super-resolution reconstruction of massive and wide-width remote sensing data, and can realize the large-scale implementation and application of super-resolution reconstruction technology in the geographic information system industry. By integrating distributed computing framework, deep learning algorithm and advanced image processing technology, it supports the super-resolution reconstruction task of a single 50GB remote sensing image, supports the efficient super-resolution reconstruction task of massive and wide-width remote sensing data, and also supports the elimination of artifacts after super-resolution, supports intelligent automatic scheduling, and realizes efficient and high-quality super-resolution reconstruction of ultra-large-scale remote sensing data. It not only breaks through the technical bottleneck of traditional methods, but also provides strong support for the high-precision application of remote sensing images.

[0036] 2. The present invention reconstructs a complete high-resolution remote sensing image by accurately filling each super-resolution reconstructed data block back to its original position and fusing the overlapping areas. This process not only solves the processing difficulties caused by the excessive amount of original data, but also ensures high-quality output of image reconstruction, providing strong technical support for remote sensing data analysis and application.

[0037] 3. The present invention supports the elimination of artifacts at the boundaries of irregular remote sensing data after super-resolution reconstruction. It can not only effectively overcome the challenges brought by nodata areas, but also make full use of the advantages of super-resolution technology to provide more accurate and clear image data for remote sensing data analysis and application. This method is not only suitable for specific remote sensing data sets, but also has wide applicability and is also effective for processing other image data with nodata areas.

[0038] 4. The present invention closely integrates the distributed GPU cluster with the task scheduling strategy, which not only realizes the efficient use of GPU resources and the parallel acceleration of tasks, but also supports no less than 24 concurrent processing through intelligent task allocation and dynamic resource management, ensuring the stable operation and efficient processing capability of the system when facing massive data. Compared with the traditional single-machine processing mode, the present invention performs particularly well on large-scale data sets, significantly shortens the processing time, and improves the overall efficiency of data processing.

[0039] 5. The system of the present invention also introduces a resource and error management mechanism to automatically monitor abnormal conditions during the processing process to ensure the stable operation of the system and data integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the principle of the remote sensing image super-resolution reconstruction method of the present invention;

[0041] Figure 2 This is a flow chart of the principle of the intelligent task scheduling strategy of the present invention;

[0042] Figure 3 This is a comparison diagram of the artifact elimination effect of the present invention;

[0043] Figure 4 This is a schematic diagram of the structure of the remote sensing image super-resolution reconstruction system of the present invention;

[0044] Figure 5 It is a schematic diagram of the structure of the electronic device of the present invention. DETAILED DESCRIPTION

[0045] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar symbols throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0046] Large-width remote sensing data generally exceeds 50GB. Due to insufficient memory during GPU inference, existing technologies are difficult to achieve effective super-resolution processing, resulting in the inability to apply super-resolution technology when processing large-width remote sensing images.

[0047] In view of the above problems, the present invention provides a method, device and medium for super-resolution reconstruction of massive remote sensing images based on Ray cluster. Figure 1 A schematic diagram of a process for super-resolution reconstruction of massive remote sensing images based on Ray clusters provided in an embodiment of the present invention, the method comprising:

[0048] S1. Process the large-width remote sensing data in blocks.

[0049] The present invention adopts a set of efficient block processing strategies to analyze the original wide-width remote sensing data, determine the optimal data block size, and subdivide the large data set of wide-width remote sensing data into a series of smaller and easier to manage data blocks. The size of each data block is carefully calculated to ensure that they can adapt to the memory limitations of the current computing environment while being large enough to maintain the efficiency and accuracy of image processing.

[0050] The optimal data block size is usually adjusted dynamically based on the memory specifications and processing capabilities of the target hardware platform.

[0051] Then, using the sliding window technology, a series of data blocks are evenly cut out along the length and width of the image. Each data block contains image information within a certain range and is both independent and interrelated.

[0052] S2. Use the parallel computing power of the Ray cluster GPU to reconstruct and obtain super-resolution remote sensing RGB images.

[0053] In the reconstruction processing of data blocks, faced with the task of super-resolution reconstruction of massive remote sensing data, it is difficult to meet the needs of efficient and stable data processing by relying solely on single-dimensional computational optimization.

[0054] The present invention adopts a highly parallel computing strategy, fully utilizing the parallel computing capability of the GPU to independently perform super-resolution reconstruction on each data block. This approach not only greatly shortens the overall processing time, but also fully utilizes the advantages of modern computing hardware and realizes efficient use of computing resources.

[0055] In view of the above problems, the present invention introduces a solution that combines a distributed GPU cluster based on the Ray framework with a precise task scheduling strategy, forming an efficient and robust distributed task processing and scheduling system, which maximizes resource utilization efficiency, improves processing speed, and ensures stable operation of the system.

[0056] Specifically, Figure 2As shown in the figure, first, a distributed GPU cluster is built using the Ray framework to achieve comprehensive integration and fragmentation of GPU resources. The GPU resources in the cluster are finely divided into multiple sub-GPUs, each of which only occupies part of the computing power of a single GPU graphics card. This measure allows the creation of multiple actors to execute tasks on the same physical GPU.

[0057] The task pool of reconstructing super-resolution remote sensing RGB images is divided into multiple task execution actors, each of which represents a super-resolution model instance.

[0058] Through strict resource limitation strategies, upper limits on GPU, CPU and memory resource usage are set for each task actor. Especially in the super-resolution task execution and data fusion stages, precise control of GPU resource usage effectively avoids OOM (Out Of Memory) problems caused by exponential expansion of resources. At the same time, it ensures the reasonable allocation of memory resources and avoids the risk of memory overflow during data processing, thereby ensuring the stable operation and efficient processing capabilities of the system.

[0059] In combination with the above division, the present invention designs a set of intelligent task scheduling strategies to ensure efficient allocation and execution of tasks. Specifically:

[0060] The core of the system - Task Manager, is responsible for reading the image information to be processed from the preset file list and dynamically tracking the task status, ensuring that each task is taken over by only a single processing unit, avoiding duplicate processing, and achieving orderly and efficient allocation of tasks.

[0061] The processing units (Actors) are optimally configured based on the available GPU resources, focusing on the efficient execution of super-resolution reconstruction tasks of remote sensing data. Through the built-in deep learning model, image data is processed quickly and accurately, significantly improving the image processing speed and quality.

[0062] In addition, the system also introduces resource and error management mechanisms to automatically monitor abnormal situations during the processing process to ensure the stable operation of the system and data integrity.

[0063] like Figure 2 As shown in the figure, when a task is submitted to the task pool, the scheduling center will request the executor pool to check whether there is an idle executor. When there is an idle executor, the scheduling center will assign the task to the executor, and the executor will perform the corresponding super-resolution reconstruction task; if the executor is successfully executed, it will respond to the scheduling center with a successful execution and update the executor status in the executor pool to idle; if the executor fails to execute, it will trigger an error record and respond to the scheduling center with a failure.

[0064] The present invention closely integrates the distributed GPU cluster with the task scheduling strategy, and intelligently schedules the execution of sub-GPU executors according to each task actor, thereby achieving efficient utilization of GPU resources and parallel acceleration of tasks. Through intelligent task allocation and dynamic resource management, the stable operation and efficient processing capability of the system when facing massive data are ensured. Compared with the traditional single-machine processing mode, the performance of this solution on large-scale data sets is particularly outstanding, which significantly shortens the processing time and improves the overall efficiency of data processing.

[0065] Furthermore, in order to ensure the integrity and continuity of the reconstructed image, an overlapping area is set between adjacent data blocks. This design ensures that the texture and details of the image can achieve a smooth transition even at the boundary, avoiding possible breaks or artifacts in the reconstructed image and maintaining the overall coherence and visual consistency of the image.

[0066] In the processing of overlapping area fusion, by presetting appropriate overlapping areas between adjacent data blocks, and using statistical methods (such as average, maximum or minimum fusion) to smoothly transition these overlapping parts in the final image synthesis stage, the seamless connection and visual continuity of the reconstructed image are ensured, fundamentally eliminating the boundary fault phenomenon that may be caused by block processing.

[0067] By accurately filling each super-resolution reconstructed data block back to its original position and fusing the overlapping areas, a complete super-resolution remote sensing RGB image can be reconstructed. This process not only solves the processing difficulties caused by the large amount of original data, but also ensures high-quality output of image reconstruction, providing strong technical support for remote sensing data analysis and application.

[0068] S3. Identify the nodata area of ​​wide remote sensing data and accurately match the Mask layer.

[0069] When processing wide-width remote sensing data, the presence of nodata regions poses additional challenges to super-resolution reconstruction.

[0070] The nodata area, usually composed of pixels with a pixel value of 0 or other predefined values, represents missing information or invalid areas in the data acquisition process; when reconstructing super-resolution remote sensing RGB images, if not controlled, these nodata areas will mistakenly participate in the reconstruction process, resulting in undesirable low pixel values ​​at the image boundaries, forming a visual black spot effect, affecting the overall quality and usability of the image.

[0071] To solve this problem, a binary Mask layer is used to extract the nodata area; the Mask layer is essentially a binary image used to mark the valid area and invalid area (nodata) of the data. In the original wide-width remote sensing data, the valid data area is marked as 1, and the nodata area is marked as 0.

[0072] By using the Mask layer, nodata areas can be accurately identified and isolated to avoid adverse effects during super-resolution reconstruction. However, super-resolution reconstruction usually significantly increases the size of the image. For example, when the resolution is increased to 4 times the original image, the size of the image changes. If the Mask layer of the original size is used directly, it will not be able to perfectly match the super-resolution image.

[0073] To solve this problem, the present invention adopts adaptive upsampling technology, combined with edge preservation and detail enhancement algorithms (such as bilinear interpolation or more advanced interpolation technology), which not only achieves accurate matching of the size of the Mask layer and the super-resolution RGB image, but also ensures the complete transmission of boundary information and accurate alignment at the pixel level. This step is crucial because only a size-matched Mask layer can accurately guide which areas in the super-resolution image need to be retained and which areas should be considered invalid and removed or made transparent.

[0074] S4. Use the Mask layer as an Alpha channel to merge with the RGB image to obtain an RGBA image.

[0075] The adaptively upsampled Mask layer and the super-resolution RGB image are precisely matched in size, which can achieve complete transmission of boundary information and precise alignment at the pixel level. The Mask layer is then integrated with the RGB image as an alpha channel to form an RGBA image.

[0076] The present invention integrates the strategy of RGBA images and cleverly displays the nodata area as completely transparent while retaining the rich color information of the valid data area, thereby avoiding the visual artifacts that may be caused by the nodata area after super-resolution reconstruction, ensuring the natural transition and high authenticity of the image, especially showing excellent performance in boundary clarity and detail retention.

[0077] like Figure 3 As shown, Figure 3 A shows the artifacts caused by irregular boundaries. Figure 3 B shows the abnormal effect caused by low pixel generation. Figure 3 C shows the effect after treatment using the method of the present invention.

[0078] By the method provided by the present invention, in the process of super-resolution remote sensing image reconstruction, the RGB layer and the Nodata layer of the remote sensing image data are respectively extracted, the distributed sliding window module is used to split the data and construct the Mask layer, then the parallel executor performs super-resolution reconstruction on the data block, and the adaptive upsampling algorithm is used to calculate the Alpha layer that matches the data size after super-resolution reconstruction; when the parallel executor completes the super-resolution task, the super-resolution reconstructed data block is restored to the original position; finally, the two parts of data are fused to obtain a super-resolution RGBA image.

[0079] The method provided by the present invention can not only effectively overcome the challenges brought by nodata areas, but also fully utilize the advantages of super-resolution technology to provide more accurate and clear image data for remote sensing data analysis and application; this method is not only suitable for specific remote sensing data sets, but also has wide applicability and is also effective for processing other image data with nodata areas.

[0080] The present invention also provides a massive remote sensing image super-resolution reconstruction system based on Ray cluster, such as Figure 4 As shown, the system includes: business layer, task scheduling layer, Ray distributed GPU cluster layer, algorithm environment layer, software environment layer and hardware environment layer.

[0081] The system runs in a multi-GPU, multi-CPU hardware environment, and is equipped with appropriate bandwidth, memory, and SSD storage environment.

[0082] The business layer accepts the user's task list for processing large-width remote sensing data. The task scheduling layer uses the task management module and the task scheduling module to read the image information to be processed from the preset file list, dynamically track the task status, and schedule tasks to ensure that each task is taken over by only a single processing unit to avoid repeated processing and achieve orderly and efficient task allocation; the Ray distributed GPU cluster layer uses the parallel manager, parallel executor, resource control module and error monitoring module to optimize the configuration of available GPU resources, and use the Ray distributed GPU cluster in parallel to improve the execution efficiency of super-resolution reconstruction tasks.

[0083] The software environment uses the Runtime execution engine, Pytorch framework and Tensorflow framework, calls the super-resolution reconstruction algorithm, distributed blocking module and visual artifact removal module, realizes distributed blocking processing of image data, visual artifact removal, and coordination of distributed task processing and scheduling strategies, processes high-resolution reconstruction tasks quickly and accurately, and automatically monitors abnormal situations during the processing process to ensure the stable operation of the system and data integrity.

[0084] Through the application of the above system and method, the method of the present invention supports super-resolution reconstruction of remote sensing data of a single file of 50GB; supports the elimination of artifacts at the boundaries after super-resolution reconstruction of irregular remote sensing data; supports no less than 24-way concurrent processing (taking the system configuration: 128CPU, 1000G memory, 8*4090 graphics card (24GB) as an example); supports orderly and efficient allocation and scheduling of tasks; and supports resource and error management functions.

[0085] The distributed remote sensing data super-resolution reconstruction method and system based on the Ray cluster of the present invention can realize the efficient and reliable super-resolution reconstruction function of massive and wide-bandwidth remote sensing data, and can realize the large-scale implementation and application of super-resolution reconstruction technology in the geographic information system industry.

[0086] The present invention also provides an electronic device, Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 5 As shown, the electronic device may include: a processor, a communications interface, a memory, and a communications bus, wherein the processor, the communications interface, and the memory communicate with each other via the communications bus. The processor may call the logic instructions in the memory, for example, to execute the following method:

[0087] S1, block processing of large-width remote sensing data;

[0088] S2, using the parallel computing power of the Ray cluster GPU to reconstruct and obtain super-resolution remote sensing RGB images;

[0089] S3, identify the nodata area of ​​large-width remote sensing data and accurately match the Mask layer;

[0090] S4. Use the Mask layer as an Alpha channel to merge with the RGB image to obtain an RGBA image.

[0091] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0092] An embodiment of the present invention further provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method provided in each of the above embodiments is implemented, for example, including:

[0093] S1, block processing of large-width remote sensing data;

[0094] S2, using the parallel computing power of the Ray cluster GPU to reconstruct and obtain super-resolution remote sensing RGB images;

[0095] S3, identify the nodata area of ​​large-width remote sensing data and accurately match the Mask layer;

[0096] S4. Use the Mask layer as an Alpha channel to merge with the RGB image to obtain an RGBA image.

[0097] The system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0098] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for super-resolution reconstruction of massive remote sensing images based on Ray cluster, characterized in that: include: S1, block processing of large-width remote sensing data; S2, using the parallel computing power of the Ray cluster GPU to reconstruct and obtain super-resolution remote sensing RGB images; S3, identify the nodata area of ​​large-width remote sensing data and accurately match the Mask layer; S4. Use the Mask layer as an Alpha channel to merge with the RGB image to obtain an RGBA image.

2. The reconstruction method according to claim 1, characterized in that: The S1 processes the wide-bandwidth remote sensing data in blocks, including the following steps: S11, analyzing the original wide-width remote sensing data to determine the optimal data block size; S12. Using the sliding window technology, a series of data blocks are evenly cut out along the length and width directions of the image according to the optimal data block size.

3. The reconstruction method according to claim 2, characterized in that: The super-resolution remote sensing RGB image is reconstructed and acquired in S2, including: Overlapping areas are set between adjacent data blocks. During image reconstruction, statistical methods are used to smoothly transition these overlapping areas to achieve seamless connection and visual continuity of the reconstructed image.

4. The reconstruction method according to claim 1, characterized in that: The exact match Mask layer in S3 includes: An adaptive upsampling algorithm, an edge-preserving algorithm, and a detail enhancement algorithm are used for the Mask layer to achieve accurate size matching between the Mask layer and the super-resolution RGB image.

5. The reconstruction method according to claim 4, characterized in that: The precise matching includes complete transfer of boundary information and precise alignment at the pixel level.

6. The reconstruction method according to claim 1, characterized in that: In S2, the parallel computing capability of the Ray cluster GPU is utilized to reconstruct and acquire a super-resolution remote sensing RGB image, including the steps of: S21. Use the Ray framework to build a distributed GPU cluster executor pool, and finely split the GPU cluster executor pool into multiple sub-GPU executors; S22, dividing the task pool of reconstructing super-resolution remote sensing RGB images into multiple task execution actors; S23, intelligently scheduling the sub-GPU executor to execute according to each execution task actor.

7. The reconstruction method according to claim 6, characterized in that: The task execution actor adopts a resource limitation strategy, including: setting upper limits on GPU, CPU and memory resource usage for each task execution actor.

8. The reconstruction method according to claim 1, characterized in that: When reconstructing and acquiring the super-resolution remote sensing RGB image in S2, the resource and error management mechanism is used to automatically monitor abnormal conditions during the task processing to maintain data integrity.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the reconstruction method according to any one of claims 1 to 8 are implemented.

10. A non-transitory 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 reconstruction method according to any one of claims 1 to 8 are implemented.

Images