A method for super-resolution reconstruction of land surface temperature and related equipment

Abstract

This application provides a method and related equipment for super-resolution reconstruction of land surface temperature, belonging to the field of remote sensing image processing and land surface temperature downscaling technology. The method includes: acquiring low-resolution land surface temperature data; inputting the low-resolution land surface temperature data into a backbone network based on a state-space model to obtain an initial high-resolution land surface temperature prediction result; and performing physical consistency correction on the initial high-resolution land surface temperature prediction result to obtain final high-resolution land surface temperature data. The physical consistency correction involves block-averaged downsampling of the initial prediction result, calculating residuals, generating a learnable intra-block weight map, and weighting the residuals to the initial prediction result, ensuring that the sum of weights within each non-overlapping scale block is 1. This application effectively improves the accuracy and physical reliability of land surface temperature super-resolution reconstruction while maintaining a lightweight model, and has good applicability for edge deployment.

Description

Technical Field

[0001] This application relates to the fields of remote sensing image processing and land surface temperature downscaling technology, and in particular to a land surface temperature super-resolution reconstruction method and related equipment. Background Technology

[0002] Against the backdrop of global warming and frequent extreme weather events, developing sustainable urban planning schemes and achieving rapid responses to extreme events are crucial. Among these, obtaining high-resolution land surface temperature data is a prerequisite for analyzing the urban heat island effect and supporting environmental decision-making. However, existing acquisition methods have inherent limitations: physical weather stations are sparsely deployed, making it difficult to achieve comprehensive regional coverage; while thermal infrared remote sensing data acquired by meteorological satellites typically suffers from low spatial resolution and long revisit periods, failing to meet the monitoring requirements for high precision and timeliness.

[0003] Downscaling techniques aim to convert low-resolution images into high-resolution images to reconstruct and enhance spatial detail. In the fields of high-resolution climate data generation and land surface temperature downscaling, existing technologies mainly cover statistical regression models, machine learning-based models, fusion models, and physical simulation models.

[0004] In recent years, with the evolution of deep learning technology, it has achieved remarkable results in remote sensing data downscaling tasks. Vaughan et al. proposed the convolutional conditional neural network method, using it as a general deep learning framework for multi-site statistical downscaling to generate continuous stochastic predictions of climate variables at arbitrary geographic locations. This type of method demonstrates that convolutional neural networks can be used to characterize the complex nonlinear mapping relationships of climate variables in the spatial dimension. Yu et al. proposed the DisTrad super-resolution convolutional network, combining traditional statistical downscaling ideas with deep learning methods, effectively improving the spatiotemporal resolution of remote sensing images and providing technical support for the refined dynamic analysis of surface and water surface temperatures.

[0005] Although some progress has been made in using deep learning methods for downscaling meteorological remote sensing data, existing techniques still face the following limitations and challenges: First, in pursuing prediction accuracy, current methods typically rely on deeper network architectures or larger convolutional kernels, leading to long model training times, high computational resource and memory consumption, which hinders rapid model iteration and edge deployment. Second, in land surface temperature downscaling tasks, existing models often focus only on optimizing statistical evaluation indicators, neglecting the physical plausibility of the results. This easily leads to physical unreliability issues such as regional average temperature drift and imbalanced temperature distribution within pixel blocks. Therefore, how to construct a land surface temperature downscaling method that balances prediction accuracy, computational efficiency, and physical reliability still requires further research. Summary of the Invention

[0006] The main objective of this application is to propose a method, system, electronic device, storage medium, and program product for super-resolution reconstruction of land surface temperature based on state space and scale consistency. This aims to solve the technical problems of high computational resource consumption and poor physical reliability in existing land surface temperature downscaling techniques. While maintaining a lightweight structure, a scale consistency correction mechanism is used to achieve a better balance between pixel-level accuracy and regional aggregation consistency in the super-resolution results, and to improve the applicability of the model in edge deployment scenarios.

[0007] To achieve the above objectives, one aspect of this application proposes a super-resolution reconstruction method for land surface temperature, the method comprising: Acquire low-resolution surface temperature data; The low-resolution land surface temperature data is input into a pre-constructed super-resolution reconstruction model to obtain an initial high-resolution land surface temperature prediction result; wherein, the super-resolution reconstruction model includes a backbone network based on a state-space model; The initial high-resolution land surface temperature prediction results are subjected to scale consistency correction to obtain the final high-resolution land surface temperature data; the scale consistency correction includes: The initial high-resolution surface temperature prediction results are downsampled by block averaging to obtain aggregated low-resolution data. Calculate the low-resolution residual based on the low-resolution surface temperature data and the reconstituted low-resolution data; The low-resolution residuals are extended to a high-resolution space to obtain a high-resolution residual map; A high-resolution intra-block weighted map is generated based on the initial high-resolution surface temperature prediction results, and the high-resolution residual map is weighted and distributed to the initial high-resolution surface temperature prediction results using the high-resolution intra-block weighted map to obtain the final high-resolution surface temperature data; wherein, the sum of the weights of the high-resolution intra-block weighted map in each non-overlapping scale block is 1.

[0008] In some embodiments, the state-space model-based backbone network is a lightweight backbone network constructed based on a selective state-space model, which includes a shallow convolutional feature extraction module, a multi-level residual state-space module, and a reconstruction upsampling module connected in sequence; the multi-level residual state-space module is used to enhance the ability to model long-range spatial dependencies.

[0009] In some embodiments, generating a high-resolution intra-block weighted map based on the initial high-resolution surface temperature prediction results includes: An initial weight map is generated using the weight prediction head based on the initial high-resolution surface temperature prediction results, the deep features output by the backbone network, and the high-resolution guidance information. The initial weight map is subjected to intra-block normalization so that the sum of the weights in each non-overlapping scale block is 1, thereby obtaining the high-resolution intra-block weight map.

[0010] In some embodiments, the high-resolution guidance information includes one or more of the following: normalized vegetation index, normalized water index, and normalized difference water-vegetation index.

[0011] In some embodiments, the super-resolution reconstruction model is trained using a joint loss function, which includes: High-resolution reconstruction loss used to constrain the error between the final high-resolution surface temperature data and the true high-resolution surface temperature data; Scale physical consistency loss used to constrain the consistency between the final high-resolution land surface temperature data after back-aggregation and the low-resolution land surface temperature data; And a gradient constraint loss used to constrain the consistency of the final high-resolution land surface temperature data with the real high-resolution land surface temperature data in terms of spatial gradient.

[0012] In some embodiments, the final high-resolution surface temperature data is calculated using the following formula:

[0013] in, This indicates the final high-resolution output. This represents the initial high-resolution prediction result. Indicates the scale factor. Represents a high-resolution intra-block weight map. This represents a high-resolution residual map obtained by expanding the low-resolution residual. This indicates element-wise multiplication.

[0014] To achieve the above objectives, another aspect of this application proposes a surface temperature super-resolution reconstruction system, the system comprising: The data acquisition module is used to acquire low-resolution surface temperature data; The super-resolution reconstruction module is used to input the low-resolution surface temperature data into a pre-constructed super-resolution reconstruction model to obtain an initial high-resolution surface temperature prediction result; wherein, the super-resolution reconstruction model includes a backbone network based on a state-space model; The scale consistency correction module is used to perform scale consistency correction on the initial high-resolution land surface temperature prediction results to obtain the final high-resolution land surface temperature data. The scale consistency correction module includes: The scaling unit is used to perform block-averaged downsampling on the initial high-resolution surface temperature prediction results to obtain aggregated low-resolution data. The residual calculation unit is used to calculate the low-resolution residual based on the low-resolution surface temperature data and the aggregated low-resolution data, and to extend the low-resolution residual to the high-resolution space to obtain a high-resolution residual map. The weight generation unit is used to generate a high-resolution intra-block weight map based on the initial high-resolution surface temperature prediction result, wherein the sum of the weights of the high-resolution intra-block weight map in each non-overlapping scale block is 1. The correction unit is used to apply the weighted distribution of the high-resolution residual map to the initial high-resolution surface temperature prediction result using the high-resolution intra-block weight map, so as to obtain the final high-resolution surface temperature data.

[0015] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0016] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0017] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0018] Compared with the prior art, this application has the following advantages and beneficial effects: 1) By adopting a lightweight backbone structure based on a state-space model (such as the improved Mamba), long-range dependency modeling is achieved with a lower parameter scale, which effectively reduces training resource consumption and improves the model's deployment applicability on edge devices.

[0019] 2) By adopting a learnable weight allocation method within the block and combining it with an auxiliary exponential channel of high-resolution guiding data for adaptive adjustment, the temperature detail distribution within the block can be restored more reasonably while maintaining conservation constraints, avoiding the loss of detail caused by simple uniform allocation.

[0020] 3) By adopting a joint training method that combines reconstruction loss, scale physical consistency loss and gradient loss, the model achieves a good balance between pixel error, structural details and physical credibility, effectively solving the physical unreliability problems such as regional average temperature drift and unbalanced temperature distribution within pixel blocks that exist in the existing technology. Attached Figure Description

[0021] Figure 1 This is a flowchart of the surface temperature super-resolution reconstruction method provided in the embodiments of this application.

[0022] Figure 2 This is an overall framework diagram of the super-resolution reconstruction method in the embodiments of this application.

[0023] Figure 3 This is a schematic diagram comparing the experimental results of the method in the embodiments of this application with those of the baseline model and the ablation model.

[0024] Figure 4 This is a schematic diagram of the three-dimensional rendering effect of a selected area in the embodiments of this application; the transparency ratio of terrain and temperature can be adjusted, and it can be rotated or dragged for viewing.

[0025] Figure 5 This is a schematic diagram of the visualization effect of another selected area in the embodiments of this application.

[0026] Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0029] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0030] 1) Mamba is a novel deep learning architecture based on a selective state-space model. Unlike traditional Transformer architectures and traditional convolutional neural networks that heavily rely on network depth and large kernels, Mamba introduces an input-dependent dynamic selection mechanism and hardware-aware algorithms. This allows it to maintain the global receptive field while achieving a linear (O(N)) increase in computational complexity with the length of the input sequence. This architecture significantly reduces the model's parameter size and runtime memory usage while avoiding complex attention mechanisms, and effectively improves the training and inference throughput for long sequences of data.

[0031] To address the technical problems existing in the prior art, this application provides a method, system, electronic device, storage medium, and program product for super-resolution reconstruction of land surface temperature based on state space and scale consistency. This solution, while maintaining a lightweight structure, achieves a good balance between pixel-level accuracy and regional aggregation consistency in the super-resolution results through block-level conservation correction and scale consistency training mechanisms, thereby improving the model's applicability in edge deployment scenarios. Furthermore, based on the super-resolution reconstruction method, this application proposes a 3D visualization device for land surface temperature super-resolution results. This device standardizes and organizes the super-resolution data, performs terrain-based 3D mapping, and provides interactive display, improving the model's comparison efficiency and practical usability.

[0032] The land surface temperature super-resolution reconstruction method provided in this application relates to the fields of remote sensing image processing and land surface temperature downscaling technology. The land surface temperature super-resolution reconstruction method provided in this application can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the land surface temperature super-resolution reconstruction method, but is not limited to the above forms.

[0033] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0034] like Figure 1 As shown, this embodiment provides a method for super-resolution reconstruction of land surface temperature, including the following steps: Step S1: Obtain low-resolution surface temperature data.

[0035] Step S2: Input the low-resolution surface temperature data into the pre-constructed super-resolution reconstruction model to obtain the initial high-resolution surface temperature prediction result. The super-resolution reconstruction model includes a backbone network based on a state-space model.

[0036] Step S3: Perform scale consistency correction on the initial high-resolution land surface temperature prediction results to obtain the final high-resolution land surface temperature data. The scale consistency correction includes the following sub-steps: S31: Perform block-averaged downsampling on the initial high-resolution surface temperature prediction results to obtain aggregated low-resolution data; S32: Calculate the low-resolution residual based on the low-resolution surface temperature data and the reconstituted low-resolution data; S33: Extend the low-resolution residual to a high-resolution space to obtain a high-resolution residual map; S34: Generate a high-resolution intra-block weighted map based on the initial high-resolution surface temperature prediction result, and use the high-resolution intra-block weighted map to weight and distribute the high-resolution residual map to the initial high-resolution surface temperature prediction result to obtain the final high-resolution surface temperature data; wherein, the sum of the weights of the high-resolution intra-block weighted map in each non-overlapping scale block is 1.

[0037] The solutions of the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific application examples.

[0038] This embodiment first uses a lightweight state-space super-resolution backbone network to generate initial high-resolution land surface temperature prediction results. Then, by combining block-level conservation correction and learnable intra-block weight allocation mechanism, the initial prediction results are corrected for scale consistency, and finally, high-resolution land surface temperature results with prediction accuracy, lightweight characteristics and regional scale reliability are obtained.

[0039] Specifically, the low-resolution land surface temperature is first input into the backbone network to obtain the initial high-resolution prediction results:

[0040] in, This indicates a low-resolution surface temperature input. This represents a super-resolution backbone network based on a state-space model. This represents the initial high-resolution prediction result. The backbone network includes a shallow convolutional feature extraction module, a multi-level residual state space module, and a reconstruction upsampling module. The residual state space module is used to enhance the ability to model long-range spatial dependencies and achieve efficient feature representation with a low parameter scale.

[0041] To ensure scale consistency between the super-resolution results and the original low-resolution observations, this embodiment performs average downsampling on a scale-block basis on the initial high-resolution prediction results to obtain the re-aggregated low-resolution results:

[0042] in, The scale factor is represented as The block average downsampling operator is then used. The residual between the input low-resolution observations and the back-aggregation result is then calculated.

[0043] Based on this, the low-resolution residuals are mapped back to the high-resolution space, and the initial high-resolution prediction results are conservatively corrected by combining intra-block weights to obtain the final high-resolution output:

[0044] in, This represents a high-resolution residual map obtained by expanding the low-resolution residual. Represents a high-resolution intra-block weight map. This represents element-wise multiplication. This represents the final high-resolution output. The intra-block weights satisfy the condition that the sum of the weights within each non-overlapping scale block is 1, thereby ensuring the total amount of low-resolution residuals is conserved in the corresponding high-resolution blocks.

[0045] Regarding the construction of intra-block weights, this embodiment uses a lightweight convolutional weight prediction head to construct intra-block weights based on the initial high-resolution prediction results, deep features of the backbone network, and high-resolution guiding information, thereby achieving adaptive allocation of intra-block residuals. In a preferred embodiment, the high-resolution guiding information is further constructed as auxiliary index features to reflect differences in land cover, vegetation, and moisture, thereby enhancing the ability to recover temperature details under complex surface conditions.

[0046] To ensure that the weight allocation satisfies the conservation constraint, the learnable intra-block weights are obtained through intra-block normalization, which satisfies:

[0047] in, Indicates the first This allows for the redistribution of low-resolution residuals within high-resolution blocks while maintaining overall consistency and adaptively adjusting to local surface features.

[0048] During training, this embodiment uses a joint loss function to optimize the network. This joint loss function includes at least high-resolution reconstruction loss, scale-physical consistency loss, and gradient constraint loss, and its expression can be written as:

[0049] in, Used to constrain the error between the final high-resolution output and the true high-resolution surface temperature. Used to constrain the consistency between the final output after aggregation and the original low-resolution input. Used to constrain the consistency between predicted and actual results in spatial gradient. , and These represent the weighting coefficients of the corresponding loss terms.

[0050] This embodiment uses the Heihe River Basin surface temperature dataset to train, validate, and test the method described in this embodiment.

[0051] (1) Dataset and Preprocessing The dataset includes low-resolution (120m) land surface temperature data, high-resolution (30m) land surface temperature data, and high-resolution guide data (such as NDVI, NDWI, and NDMI). The ratio of the training set, validation set, and test set is 6:1:3. To improve the stability of model evaluation and reduce the impact of randomness from a single partition, K-fold cross-validation is used for model training and validation during the training and validation phases. In this embodiment, K is 5. After completing K rounds of cross-validation, the model parameters with the best validation results are selected as the final model, and tested on an independent test sample set to obtain the final performance evaluation results of the method of this invention.

[0052] The input data is normalized so that its values ​​fall within the range required for model training. The input image patch size is set to 128×128 pixels.

[0053] (2) Model structure like Figure 2 As shown, the super-resolution reconstruction model used in this embodiment includes a backbone network based on a state-space model and a scale consistency correction module.

[0054] 2.1) Backbone Network The backbone network adopts a lightweight structure based on a selective state space model, which includes a shallow convolutional feature extraction module, a multi-level residual state space module, and a reconstruction upsampling module connected in sequence.

[0055] a) Shallow convolutional feature extraction module: Consists of a 3×3 convolutional layer, used to map the input low-resolution surface temperature data to a high-dimensional feature space.

[0056] b) Multi-level residual state space module: This module consists of multiple stacked residual state space blocks, each containing a Mamba module and residual connections. In this embodiment, the network depth is set to [2,2,2,2], and the feature embedding dimension is set to 60. This module utilizes Mamba's dynamic input dependency selection mechanism to maintain the global receptive field while achieving linear growth in computational complexity and sequence length, effectively capturing the long-range spatial dependencies of surface temperature.

[0057] c) Reconstruction Upsampling Module: Composed of sub-pixel convolutional layers (PixelShuffle) or deconvolutional layers, used to reconstruct the high-dimensional feature map into an initial high-resolution land surface temperature prediction result. .

[0058] 2.2) Scale Consistency Correction Module The scale consistency correction module is used to correct the initial prediction results. Constraints are applied to ensure that the scale remains consistent with the original low-resolution input under the scale transformation. The specific process is as follows: First, block-averaged downsampling is performed on the initial high-resolution prediction result Y0 to obtain the back-aggregated low-resolution result. :

[0059] in, The scale factor is represented as The block average downsampling operator. In this embodiment... .

[0060] Then, the input low-resolution observations are calculated. With back aggregation results The residuals between:

[0061] Next, the low-resolution residuals By extending to a high-resolution space through repetition or interpolation, a high-resolution residual map is obtained. .

[0062] To distribute the residuals reasonably across each pixel within a high-resolution block, this embodiment introduces a learnable intra-block weight map W. This weight map is generated using a lightweight convolutional weight prediction head, whose input includes: the initial high-resolution prediction result. The backbone network outputs deep features and high-resolution guiding information (such as NDVI, NDWI, NDMI, etc.). After the prediction head outputs the initial weight map, the weights within each non-overlapping scale block are normalized so that:

[0063] Finally, the normalized weighted map W is used to analyze the high-resolution residual map. The weighted allocation is then added to the initial prediction results. To obtain the final high-resolution output :

[0064] in, This indicates element-wise multiplication. This operation ensures the conservation of the total amount of low-resolution residuals within high-resolution blocks, while also enabling adaptive detail recovery based on local surface features (such as vegetation and water bodies).

[0065] (3) Model training This embodiment uses a joint loss function to optimize the network. The joint loss function includes high-resolution reconstruction loss. Scale-based physical consistency loss and gradient constraint loss Its expression is:

[0066] in, , and These represent the weighting coefficients of the corresponding loss terms, which are set to 1.0, 0.1, and 0.05 respectively in this embodiment.

[0067] The model training device uses a single V100-32GB graphics card.

[0068] (4) Tests and Results For each test sample, firstly, low-resolution land surface temperature input, high-resolution land surface temperature labels, and high-resolution guiding data are loaded, and normalization processing consistent with the training phase is performed. Then, the low-resolution land surface temperature is input into the model to obtain the final high-resolution prediction result. Next, the prediction result is compared with the actual high-resolution land surface temperature labels, and the root mean square error (RMSE), mean absolute error (MAE), correlation coefficient (Corr), relative standard deviation (RSD), and re-aggregation consistency error (RAE) are calculated. The re-aggregation consistency error is obtained by averaging the predicted high-resolution land surface temperature results back to the low-resolution space according to a 4-fold scale block, and then comparing it with the original low-resolution input. It is used to characterize the scale consistency of the output results of this invention in the sense of scale transformation.

[0069] The test results of this embodiment are as follows: Figure 3 As shown. Specifically, in the super-resolution task of surface temperature in the Heihe River Basin, the model in this embodiment has a root mean square error (RMSE) of 0.848650, a mean absolute error (MAE) of 0.576083, a correlation coefficient (Corr) of 0.951870, a relative standard deviation (RSD) of 0.048696, and a re-aggregation consistency error (RAE) of 1.45 × 10⁻⁶. -5 Compared to the baseline model without scale conservation constraints, the model of this invention achieves a significant reduction in the reconvergence consistency error index, while showing no significant degradation in indices such as root mean square error, mean absolute error, correlation coefficient, and relative standard deviation. Experimental results demonstrate that this invention significantly enhances the regional scale conservation and physical reliability of the results while maintaining the accuracy of super-resolution prediction.

[0070] This embodiment also provides a surface temperature super-resolution reconstruction and 3D visualization device based on state space and scale consistency. Building upon the surface temperature super-resolution reconstruction method described above, this embodiment further constructs a 3D visualization device to intuitively display the super-resolution results.

[0071] This 3D visualization device adopts a three-layer architecture: a data transformation layer, a data management layer, and a front-end rendering layer. Data conversion layer: Reads super-resolution temperature results, low-resolution temperature data, digital elevation data, and optional RGB and NDVI guide data, and converts them into standardized asset files (such as 3D Tiles or GLTF format) that can be called by the front end.

[0072] Data Management Layer: Responsible for the storage, indexing, and scheduling of standardized asset files, supporting on-demand loading, and improving rendering efficiency.

[0073] Front-end rendering layer: Based on graphics libraries such as WebGL or Three.js, a three-dimensional terrain surface is constructed using digital elevation data, and temperature data is mapped into a semi-transparent thermal texture according to preset color bands (such as heat map color bands) and superimposed on the three-dimensional terrain surface.

[0074] like Figure 4 and Figure 5 As shown, the device offers a wealth of interactive functions, including: switching between super-resolution and low-resolution results, switching between RGB / NDVI base maps, temperature range adjustment, transparency adjustment, viewpoint reset, and pixel temperature query. Through the above technical solutions, a three-dimensional, interactive, and switchable display of temperature super-resolution results is achieved, greatly improving the intuitiveness and ease of use of the results, facilitating the analysis of thermal environments by researchers and decision-makers.

[0075] This application also provides a surface temperature super-resolution reconstruction system that can implement the above-described method. The system includes: The data acquisition module is used to acquire low-resolution surface temperature data; The super-resolution reconstruction module is used to input the low-resolution surface temperature data into a pre-constructed super-resolution reconstruction model to obtain an initial high-resolution surface temperature prediction result; wherein, the super-resolution reconstruction model includes a backbone network based on a state-space model; The scale consistency correction module is used to perform scale consistency correction on the initial high-resolution land surface temperature prediction results to obtain the final high-resolution land surface temperature data. The scale consistency correction module includes: The scaling unit is used to perform block-averaged downsampling on the initial high-resolution surface temperature prediction results to obtain aggregated low-resolution data. The residual calculation unit is used to calculate the low-resolution residual based on the low-resolution surface temperature data and the aggregated low-resolution data, and to extend the low-resolution residual to the high-resolution space to obtain a high-resolution residual map. The weight generation unit is used to generate a high-resolution intra-block weight map based on the initial high-resolution surface temperature prediction result, wherein the sum of the weights of the high-resolution intra-block weight map in each non-overlapping scale block is 1. The correction unit is used to apply the weighted distribution of the high-resolution residual map to the initial high-resolution surface temperature prediction result using the high-resolution intra-block weight map, so as to obtain the final high-resolution surface temperature data.

[0076] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0077] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0078] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0079] Please see Figure 6 , Figure 6 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 601 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 602 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 602 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 602 and is called and executed by the processor 601 using the methods described in the embodiments of this application. The input / output interface 603 is used to implement information input and output; The communication interface 604 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 605 transmits information between various components of the device (e.g., processor 601, memory 602, input / output interface 603, and communication interface 604); The processor 601, memory 602, input / output interface 603, and communication interface 604 are connected to each other within the device via bus 605.

[0080] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0081] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0082] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0083] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0084] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0085] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0086] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0087] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0088] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0089] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0090] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0091] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0092] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0093] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0094] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for land surface temperature super-resolution reconstruction, characterized in that, The method includes the following steps: Acquire low-resolution surface temperature data; The low-resolution land surface temperature data is input into a pre-constructed super-resolution reconstruction model to obtain an initial high-resolution land surface temperature prediction result; wherein, the super-resolution reconstruction model includes a backbone network based on a state-space model; The initial high-resolution land surface temperature prediction results are subjected to scale consistency correction to obtain the final high-resolution land surface temperature data; the scale consistency correction includes: The initial high-resolution surface temperature prediction results are downsampled by block averaging to obtain aggregated low-resolution data. Calculate the low-resolution residual based on the low-resolution surface temperature data and the reconstituted low-resolution data; The low-resolution residuals are extended to a high-resolution space to obtain a high-resolution residual map; A high-resolution intra-block weighted map is generated based on the initial high-resolution surface temperature prediction results, and the high-resolution residual map is weighted and distributed to the initial high-resolution surface temperature prediction results using the high-resolution intra-block weighted map to obtain the final high-resolution surface temperature data; wherein, the sum of the weights of the high-resolution intra-block weighted map in each non-overlapping scale block is 1.

2. The method of claim 1, wherein, The state-space model-based backbone network is a lightweight backbone network constructed based on a selective state-space model. It includes a shallow convolutional feature extraction module, a multi-level residual state-space module, and a reconstruction upsampling module connected in sequence. The multi-level residual state-space module is used to enhance the ability to model long-range spatial dependencies.

3. The method of claim 1, wherein, The step of generating a high-resolution intra-block weighted map based on the initial high-resolution surface temperature prediction results includes: An initial weight map is generated using the weight prediction head based on the initial high-resolution surface temperature prediction results, the deep features output by the backbone network, and the high-resolution guidance information. The initial weight map is subjected to intra-block normalization so that the sum of the weights in each non-overlapping scale block is 1, thereby obtaining the high-resolution intra-block weight map.

4. The method of claim 3, wherein, The high-resolution guidance information includes one or more of the following: normalized vegetation index, normalized water index, and normalized difference water-vegetation index.

5. The method of claim 1, wherein, The super-resolution reconstruction model is trained using a joint loss function, which includes: High-resolution reconstruction loss used to constrain the error between the final high-resolution surface temperature data and the true high-resolution surface temperature data; Scale physical consistency loss used to constrain the consistency between the final high-resolution land surface temperature data after back-aggregation and the low-resolution land surface temperature data; And a gradient constraint loss used to constrain the consistency of the final high-resolution land surface temperature data with the real high-resolution land surface temperature data in terms of spatial gradient.

6. The method of claim 1, wherein, The final high-resolution surface temperature data is calculated using the following formula: in, This indicates the final high-resolution output. This represents the initial high-resolution prediction result. Indicates the scale factor. Represents a high-resolution intra-block weight map. This represents a high-resolution residual map obtained by expanding the low-resolution residual. This indicates element-wise multiplication.

7. A land surface temperature super-resolution reconstruction system characterized by, include: The data acquisition module is used to acquire low-resolution surface temperature data; The super-resolution reconstruction module is used to input the low-resolution surface temperature data into a pre-constructed super-resolution reconstruction model to obtain an initial high-resolution surface temperature prediction result; wherein, the super-resolution reconstruction model includes a backbone network based on a state-space model; The scale consistency correction module is used to perform scale consistency correction on the initial high-resolution land surface temperature prediction results to obtain the final high-resolution land surface temperature data. The scale consistency correction module includes: The scaling unit is used to perform block-averaged downsampling on the initial high-resolution surface temperature prediction results to obtain aggregated low-resolution data. The residual calculation unit is used to calculate the low-resolution residual based on the low-resolution surface temperature data and the aggregated low-resolution data, and to extend the low-resolution residual to the high-resolution space to obtain a high-resolution residual map. The weight generation unit is used to generate a high-resolution intra-block weight map based on the initial high-resolution surface temperature prediction result, wherein the sum of the weights of the high-resolution intra-block weight map in each non-overlapping scale block is 1. The correction unit is used to apply the weighted distribution of the high-resolution residual map to the initial high-resolution surface temperature prediction result using the high-resolution intra-block weight map, so as to obtain the final high-resolution surface temperature data.

8. An electronic device, comprising: The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images