Radar echo spatial downscaling method, device, equipment, medium and program product

Abstract

This application discloses a radar echo spatial downscaling method, apparatus, equipment, medium, and program product, relating to the field of meteorological information processing technology. The method includes: acquiring low-resolution radar echo data to be processed; upsampling the low-resolution radar echo data to be processed through interpolation to obtain high-resolution target echo data; extracting features from the high-resolution target echo data to obtain an initial high-frequency information representation map; and inputting the low-resolution radar echo data to be processed and the high-resolution target echo data into a downscaling model to obtain a high-resolution radar echo field corresponding to the low-resolution radar echo data to be processed. This application enhances the application value of radar echo data in refined weather analysis and severe weather early warning.

Description

Technical Field

[0001] This application relates to the field of meteorological information processing technology, and in particular to a method, apparatus, equipment, medium, and program product for spatial downscaling of radar echoes. Background Technology

[0002] Weather radar is a key tool for monitoring weather phenomena such as precipitation and wind fields. The resolution of its echo data directly affects the accuracy of numerical weather forecasts and severe weather warnings. However, due to limitations in radar hardware, detection range, and electromagnetic wave propagation characteristics, the resolution of the acquired baseline data is often limited. Traditional interpolation methods (such as bilinear and bicubic interpolation) only perform upsampling through smoothing functions, failing to recover high-frequency components (such as strong echo fronts and fine textural structures) lost during downsampling. This results in blurred reconstruction results and distorted physical field details.

[0003] In recent years, deep learning-based single-image super-resolution techniques have made significant progress. However, most methods are designed for natural images, and their loss functions, such as mean squared error, tend to generate smooth results, which are inconsistent with the sharp nonlinear structure and clear physical boundary features in radar echo data. Furthermore, existing methods typically use low-resolution images as the sole input, lacking explicit guidance for high-frequency information recovery. When dealing with highly structured data like radar echoes, the fidelity and physical consistency of the generated results are insufficient. While generative adversarial networks (GANs) can synthesize details, they suffer from problems such as training instability and pattern collapse. Denoising diffusion probability models, as an emerging generative model, generate data through a progressive denoising process, offering advantages such as stable training and good generative diversity, providing a new solution for high-quality radar echo downscaling. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, device, storage medium, and program product for spatial downscaling of radar echoes, aiming to solve the technical problem of low quality in spatial downscaling of radar echoes in related technologies.

[0005] Firstly, to achieve the above objectives, this application provides a radar echo spatial downscaling method, the method comprising: Acquire low-resolution radar echo data to be processed; The low-resolution radar echo data to be processed is upsampled by interpolation to obtain high-resolution target echo data. Feature extraction is performed on the high-resolution echo data of the target to obtain an initial high-frequency information representation map; Input the low-resolution radar echo data to be processed and the initial high-frequency information characterization map into the downscaling model to obtain the high-resolution radar echo field corresponding to the low-resolution radar echo data to be processed. The downscaling model is constructed based on a generative network of a denoising diffusion probability model. The downscaling model is trained based on multiple sets of paired samples. The paired samples include low-resolution radar echo samples, ground truth values ​​of high-resolution radar echo samples corresponding to low-resolution radar echo samples, and reference high-frequency information representation maps extracted from the ground truth values ​​of high-resolution radar echo samples.

[0006] In one embodiment, the step of extracting features from the target high-resolution echo data to obtain an initial high-frequency information characterization map includes: The high-resolution echo data of the target is processed by convolution operation using a Laplacian convolution kernel to obtain an initial high-frequency information representation map; the core calculation method is as follows: in, This represents the input radar echo image. This represents the high-frequency information representation diagram obtained from the calculation. For the Laplace operator.

[0007] In one embodiment, the downscaling model is used for: In the reverse denoising process of the model, the interpolation results of the low-resolution radar echo data to be processed and the initial high-frequency information characterization map are used as conditions. Starting from a randomly sampled Gaussian noise tensor, a multi-scale edge enhancement network is used to perform multi-step iterative noise prediction and removal to generate a high-resolution radar echo field.

[0008] In one embodiment, the multi-scale edge enhancement network is constructed based on a U-Net structure neural network, and the multi-scale edge enhancement network includes five processing layers; The multi-scale edge enhancement network is used to stitch the interpolation results of low-resolution radar echo data with the noise latent variables of the current step as input to the multi-scale edge enhancement network. The initial high-frequency information representation map is fused with the features of this layer through the edge information fusion module to enhance the ability to recover the edge structure of weak echo regions in radar echoes. The edge information fusion module includes: A multi-scale feature generation unit is used to perform multi-scale depth-separable convolution processing on the initial high-frequency information representation map to obtain multi-scale features of the original features and radar echo edge map. The bidirectional attention fusion unit is used to perform cross-attention calculation on multi-scale features as query and key-value pairs, and then performs residual connection between the attention calculation results and the corresponding query vectors to obtain the enhanced features.

[0009] In one embodiment, the edge information fusion module is used for: The initial high-frequency information representation map is downsampled to multiple different target resolutions to generate a set of multi-scale radar echo edge information feature maps. Each multi-scale radar echo edge information feature map is input into one or more processing layers corresponding to the resolution in the multi-scale edge enhancement network, and then fused with the features extracted by the multi-scale edge enhancement network at that layer.

[0010] In one embodiment, feature fusion is performed at multiple skip connections between the decoder and encoder of the multi-scale edge enhancement network to incorporate edge information.

[0011] Secondly, to achieve the above objectives, this application further provides a radar echo spatial downscaling device, the device comprising: The acquisition module is used to acquire low-resolution radar echo data to be processed. The upsampling module is used to perform upsampling processing on the low-resolution radar echo data to be processed by interpolation to obtain high-resolution echo data of the target. The feature extraction module is used to extract features from the target high-resolution echo data to obtain an initial high-frequency information representation map; The downscaling module is used to input the low-resolution radar echo data to be processed and the high-resolution target echo data into the downscaling model to obtain the high-resolution radar echo field corresponding to the low-resolution radar echo data to be processed. The downscaling model is constructed based on a generative network of a denoising diffusion probability model. The downscaling model is trained based on multiple sets of paired samples. The paired samples include low-resolution radar echo samples, ground truth values ​​of high-resolution radar echo samples corresponding to low-resolution radar echo samples, and reference high-frequency information representation maps extracted from the ground truth values ​​of high-resolution radar echo samples.

[0012] Thirdly, to achieve the above objectives, this application further provides a radar echo spatial downscaling device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the radar echo spatial downscaling method described above.

[0013] Fourthly, to achieve the above objectives, this application further provides a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the radar echo spatial downscaling method described above.

[0014] Fifthly, to achieve the above objectives, this application further provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the radar echo spatial downscaling method described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: This application introduces a generative framework based on a denoising diffusion probability model, explicitly fusing multi-scale high-frequency edge information extracted from interpolation results during the generation process. This effectively overcomes the limitations of traditional interpolation methods in recovering high-frequency details and the insufficient characterization of radar echo physical structure by existing deep learning methods. Leveraging the stable training and high-quality generation of the diffusion model, this method uses low-resolution echoes and their high-frequency priors as constraints during the inverse denoising process. Through a multi-scale edge enhancement network, it progressively reconstructs detailed structures, more accurately recovering strong echo fronts, fine textures, and clear physical boundaries. While improving spatial resolution, it significantly enhances the fidelity and physical consistency of the reconstruction results, thereby increasing the application value of radar echo data in refined weather analysis and severe weather warnings. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the radar echo spatial downscaling method of this application.

[0019] Figure 2 This is a schematic diagram of the first structure of the radar echo spatial downscaling method of this application.

[0020] Figure 3 This is a schematic diagram of the second structure of the radar echo spatial downscaling method model in this application.

[0021] Figure 4 This is a schematic diagram illustrating the effect of the radar echo spatial downscaling method of this application.

[0022] Figure 5 This is a schematic diagram of the radar echo spatial downscaling device of this application.

[0023] Figure 6 This is a schematic diagram of the radar echo spatial downscaling device of this application.

[0024] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0026] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0027] The main solution of this application embodiment is as follows: An initial high-frequency information representation map is extracted from the input low-resolution radar echo data through interpolation and Laplacian edge detection. Subsequently, this high-frequency information representation map and the low-resolution data itself constitute a joint condition, guiding a pre-trained denoising diffusion probability model. Finally, the model generates the final high-resolution radar echo field through a multi-step iterative inverse denoising process. The core of this method lies in using edge information to explicitly and precisely guide the generation process, enabling the model to learn to focus on and recover edge and texture details during reconstruction.

[0028] Specifically, this application provides a radar echo spatial downscaling method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the radar echo spatial downscaling method of this application.

[0029] In this embodiment, the radar echo spatial downscaling method includes steps S10 to S40: Step S10: Obtain the low-resolution radar echo data to be processed.

[0030] Step S20: The low-resolution radar echo data to be processed is upsampled by interpolation to obtain high-resolution echo data of the target.

[0031] Step S30: Extract features from the target high-resolution echo data to obtain an initial high-frequency information representation map.

[0032] Step S40: Input the low-resolution radar echo data to be processed and the initial high-frequency information characterization map into the downscaling model to obtain the high-resolution radar echo field corresponding to the low-resolution radar echo data to be processed.

[0033] The downscaling model is constructed based on a generative network of a denoising diffusion probability model. The downscaling model is trained based on multiple sets of paired samples. The paired samples include low-resolution radar echo samples, ground truth values ​​of high-resolution radar echo samples corresponding to low-resolution radar echo samples, and reference high-frequency information representation maps extracted from the ground truth values ​​of high-resolution radar echo samples.

[0034] In one possible implementation, step S30 includes step A10: Step A10 involves performing convolution operations on the high-resolution echo data of the target using a Laplacian convolution kernel to obtain an initial high-frequency information representation map; the core calculation method is as follows: in, This represents the input radar echo image. This represents the high-frequency information representation diagram obtained from the calculation. For the Laplace operator.

[0035] In this embodiment, the downscaling model is used for: In the reverse denoising process of the model, the interpolation results of the low-resolution radar echo data to be processed and the initial high-frequency information characterization map are used as conditions. Starting from a randomly sampled Gaussian noise tensor, a multi-scale edge enhancement network is used to perform multi-step iterative noise prediction and removal to generate a high-resolution radar echo field.

[0036] The multi-scale edge enhancement network is built on a U-Net neural network structure and includes five processing layers. The multi-scale edge enhancement network is used to stitch together the interpolation results of low-resolution radar echo data with the noise latent variables of the current step as input to the multi-scale edge enhancement network. The initial high-frequency information representation map is fused with the features of this layer through the edge information fusion module to enhance the ability to recover the edge structure of weak echo regions in radar echoes.

[0037] The edge information fusion module includes: A multi-scale feature generation unit is used to perform multi-scale depth-separable convolution processing on the initial high-frequency information representation map to obtain multi-scale features of the original features and radar echo edge map. The bidirectional attention fusion unit is used to perform cross-attention calculation on multi-scale features as query and key-value pairs, and then performs residual connection between the attention calculation results and the corresponding query vectors to obtain the enhanced features.

[0038] The edge information fusion module is used for: The initial high-frequency information representation map is downsampled to multiple different target resolutions to generate a set of multi-scale radar echo edge maps; Each multi-scale radar echo edge map is input into one or more processing layers corresponding to the resolution in the multi-scale edge enhancement network, and then fused with the features extracted by the multi-scale edge enhancement network at that layer.

[0039] Feature fusion is performed at multiple skip connections between the decoder and encoder of the multi-scale edge enhancement network to incorporate edge information.

[0040] Specifically, for ease of understanding, this embodiment takes the reconstruction of a low-resolution radar reflectivity image of 48×48 pixels into a high-resolution radar echo field of 192×192 pixels as an example. The core is to use a conditional diffusion model guided by edge information to perform the downscaling task.

[0041] This embodiment uses a single model to complete the process end-to-end, without the need for staged training and inference, so that the edge information of the weak echo region is given full attention during the generation process.

[0042] For training the downscaling model, this embodiment uses the SEVIR (Storm EVent ImagRy) dataset, which contains high spatiotemporal resolution radar echo image sequences of various weather events over the continental United States. The original data has a spatial resolution of 0.01 degrees (approximately 1000 meters) and a temporal resolution of 5 minutes.

[0043] The core of the downscaling model is a Conditional Denoising Diffusion Probability Model (Conditional DDPM), whose neural backbone adopts a U-Net structure that integrates an edge information fusion module.

[0044] At the start of model training, input conditions and training data are prepared for the model. High-resolution radar echo data of 1000 meters is obtained. Paired radar echo data is constructed by interpolation downsampling method, including the pairing of low-resolution data of 4000 meters and high-resolution data of 1000 meters to form paired samples.

[0045] For any input target low-resolution radar echo data To construct joint conditions for it that include content and edge information. The key step is to calculate the corresponding initial high-frequency information representation map. To represent edge information.

[0046] First, use bilinear interpolation to... Upsampled to a target resolution of 192×192, high-resolution echo data of the target is obtained. This step aims to align the spatial dimensions with the high-resolution image to be generated. A Laplacian operator with a Gaussian pyramid approximation is used for processing. It extracts edges in a stable and noise-resistant manner.

[0047] The specific steps are as follows: Initial high-frequency information representation diagram By applying Laplacian convolution kernels to the target's low-resolution radar echo data The initial high-frequency information representation map is obtained by convolving the interpolation results. Considered as guiding high-frequency information in the diffusion model, the core calculation formula for Laplacian edge detection is: in, This represents the input radar echo image. This represents the high-frequency information representation diagram obtained from the calculation. For the Laplace operator.

[0048] The Laplacian operator is approximated using a Gaussian pyramid: for the input radar image After Gaussian convolution and downsampling, the image is scaled by a factor of 4 and then convolved twice. Finally, the high-frequency representation is obtained by differencing the original image. ,Right now ,in It uses a 5×5 Gaussian convolution kernel.

[0049] For example, the Gaussian convolution kernel is composed of a one-dimensional vector. Generated through outer product, the specific matrix form is as follows: ; Then Joint conditions are defined as All data values ​​are min-max normalized to the [0,1] interval to stabilize model training.

[0050] This embodiment generates high-resolution data based on the interpolation results as input to the model, which simplifies the input structure. Compared with related technologies that rely on external data generation, this embodiment does not rely on external data sources, making the model training and inference process simpler.

[0051] For downscaling model construction, a parameter is constructed as follows: U-Net denoising network The network receives three inputs: time step Encoding, latent variables for adding noise and joint condition information . Figure 2 The network model structure diagram is shown. The key settings of the network structure are as follows: In the input layer of U-Net, the low-resolution image is upsampled to the target resolution. With noise latent variables The components are spliced ​​together along the channel dimension.

[0052] At the outputs of layers 1, 2, 3, and 4 of the U-Net encoder (i.e., the downsampled feature maps), an edge information fusion module is embedded, as shown in the module structure diagram below. Figure 3 The specific operation is as follows: Multi-scale feature generation unit: For the original input features Radar echo edge map Multi-branch depthwise separable convolution and group normalization are applied to both respectively, and the summation is followed by 1x1 convolution to generate their respective queries. ,key ,value The specific calculation method is as follows: The query generated from the original features, key, and value calculations are as follows: The query, key, and value calculation for generating the radar echo edge information feature map are as follows: in, This represents a one-to-one convolution. represents a depthwise separable convolution with kernel n. The three sizes of depthwise separable convolutions correspond to the edge structure features of small- to medium-scale convective cells, mesoscale systems, and large-scale layered clouds, respectively. GN represents group normalization.

[0053] Bidirectional attention fusion unit: Calculated radar echo edge map and original feature map Calculate the two cross-attentions: Using the original features as the query, we focus on key-value pairs of edge features, as shown in the following formula: Using radar echo edge features as the query, we focus on key-value pairs of the original features, as shown in the following formula: in This refers to the feature dimension of the query matrix involved in attention calculation, corresponding to the number of columns in the matrix, where T represents the matrix transpose operation. The normalization function maps the calculated attention score to probability weights, such that the sum of the weights at all positions is 1.

[0054] Cross-attention between two feature maps is calculated using a bidirectional query mechanism. For each direction, the attention output is residually concatenated with its corresponding original query vector, and then subjected to 1×1 convolution for feature transformation and fusion, resulting in enhanced features dominated by the original features and enhanced features dominated by radar echo edge features, respectively.

[0055] The final module output is the sum of the two plus the original feature, as shown in the following formula: in, and This is the result of fusing radar echo edge features with the original features. Original features The original skip connection features are replaced and fused with the features of the corresponding layer in the decoder. This allows radar echo edge information to be efficiently and adaptively injected into different scales of the generation process, resulting in enhanced output features. Passed to the decoder.

[0056] The training process of a downscaling model includes a forward noise-adding stage and a backward noise-denoising training stage: In the forward noise addition stage, the target high-resolution echo data Gaussian noise is added progressively; for each sample pair in the training batch , and its corresponding Randomly sample a time step ,in =1000. According to the forward process formula: Reparameterization yields: in, In time step Noise-added samples of radar echo data For noise scheduling variance, , Standard Gaussian noise. Construction conditions. ,Will Interpolation upsampling to ,get Will Input U-Net network Obtain the predicted noise ).

[0057] In the reverse denoising training phase, the corresponding low-resolution radar echo samples are used. Results after upsampling and reference high-frequency information representation diagram Given the condition, the multi-scale feature extraction and fusion network is trained to predict the noise added during the forward pass. The loss function used is L, where L is the error between the predicted noise and the actual added noise: Calculate the predicted noise and the actual noise The optimization objective for the L1 loss between them is as follows: in, This represents the output noise value of the network model. The true value of the noise added to the forward process. For the corresponding time step, Initial state of the diffusion model go through Noisy radar echo data with added noise. This indicates that in all radar echo data samples All possible time steps and noise samples The mean L1 error between the noise predicted by the model and the actual noise, calculated in the model. For guiding conditions.

[0058] Using the Adam optimizer, with an initial learning rate of 0.0001, the network parameters are updated via gradient descent. .

[0059] Finally, after the model training is complete, the trained model is used to perform spatial downscaling on the low-resolution radar echo data to be processed. During the training process, the model pays attention to both low-frequency background and high-frequency details, directly learning the complete mapping from low resolution to high resolution, and weak echo regions are given equal importance in the predicted targets.

[0060] Acquire the low-resolution radar echo data to be processed and perform the same normalization process as in the training phase. Upsample to the target's high-resolution size using bilinear interpolation, and calculate the corresponding initial high-frequency information representation map using the same Gaussian pyramid Laplacian method as in the previous steps.

[0061] Sample a random noise tensor from a standard Gaussian distribution. Its size is the same as the target high-resolution image. Guided by this method, noise is predicted and removed progressively, ultimately generating a high-resolution image. The process is defined as follows: in, and It is determined by the parameter as The mean and prediction variance of the denoising network prediction. For the corresponding time step, This represents a normal distribution.

[0062] Specifically, the reverse denoising process iterates T times to recover the high-resolution radar echo image, as shown in the following formula: in, For noise scheduling variance, For U-Net networks, , In time step The noisy data is reconstructed from the radar echo data. Finally, a high-resolution radar echo field is generated. .

[0063] The generated Denormalization is performed to restore it to the original range of radar echo physical quantities, resulting in a high-resolution radar echo product that can be used for operational analysis. The visualization results are as follows: Figure 4 The visualization results show that the edges of the convective cells and the fine structure of the echo bands in the images generated by this invention are relatively clear, and are close to the high-resolution true values.

[0064] This embodiment was validated in a real-world severe convective weather case, successfully reconstructing fine meteorological structures that are difficult to distinguish in low-resolution radar data, such as the details of the mesocyclone core and the bow echo leading edge. Compared with traditional interpolation methods, the generated high-resolution echo field shows significant advantages in visual clarity, structural integrity, and physical plausibility. More importantly, this result can effectively improve the initial field quality of weather forecasting models, thereby enhancing the accuracy and precision of short-term precipitation forecasts, demonstrating the practical application value of this invention in enhancing weather monitoring and early warning capabilities and refining meteorological services.

[0065] It should be noted that all the examples above are only for understanding this application and do not constitute a limitation on the radar echo spatial downscaling method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0066] This application also provides a radar echo spatial downscaling device; please refer to [reference needed]. Figure 5 The radar echo spatial downscaling device includes: The acquisition module 10 is used to acquire low-resolution radar echo data to be processed.

[0067] The upsampling module 20 is used to perform upsampling processing on the low-resolution radar echo data to be processed by interpolation to obtain high-resolution echo data of the target.

[0068] The feature extraction module 30 is used to extract features from the target high-resolution echo data to obtain an initial high-frequency information representation map.

[0069] The downscaling module 40 is used to input the low-resolution radar echo data to be processed and the high-resolution target echo data into the downscaling model to obtain the high-resolution radar echo field corresponding to the low-resolution radar echo data to be processed.

[0070] The downscaling model is constructed based on a generative network of a denoising diffusion probability model. The downscaling model is trained based on multiple sets of paired samples. The paired samples include low-resolution radar echo samples, ground truth values ​​of high-resolution radar echo samples corresponding to low-resolution radar echo samples, and reference high-frequency information representation maps extracted from the ground truth values ​​of high-resolution radar echo samples.

[0071] The radar echo spatial downscaling device provided in this application employs the radar echo spatial downscaling method described in the above embodiments, which can solve the technical problem of low accuracy in radar echo spatial downscaling in related technologies. Compared with related technologies, the beneficial effects of the radar echo spatial downscaling device provided in this application are the same as those of the radar echo spatial downscaling method provided in the above embodiments, and other technical features in the radar echo spatial downscaling device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0072] This application provides a radar echo spatial downscaling device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the radar echo spatial downscaling method in the above embodiments.

[0073] The following is for reference. Figure 6 The diagram illustrates a structural schematic suitable for implementing the radar echo spatial downscaling device in the embodiments of this application. The radar echo spatial downscaling device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The radar echo spatial downscaling device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0074] like Figure 6As shown, the radar echo spatial downscaling device includes a processing unit 1001 (e.g., a central processing unit, graphics processor, etc.) that can perform various appropriate actions and processes according to a program stored in read-only memory 1002 (ROM) or a program loaded from storage device 1003 into random access memory 1004 (RAM). The random access memory 1004 also stores various programs and data required for the operation of the radar echo spatial downscaling device. The processing unit 1001, read-only memory 1002, and random access memory 1004 are interconnected via bus 1005. An input / output interface 1006 (I / O interface) is also connected to bus 1005. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the radar echo spatial downscaling device to communicate wirelessly or wiredly with other devices to exchange data. Although radar echo spatial downscaling devices with various systems are shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0075] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0076] The radar echo spatial downscaling device provided in this application employs the radar echo spatial downscaling method described in the above embodiments, which can solve the technical problem of low accuracy in radar echo spatial downscaling in related technologies. Compared with related technologies, the beneficial effects of the radar echo spatial downscaling device provided in this application are the same as those of the radar echo spatial downscaling method provided in the above embodiments, and other technical features in this radar echo spatial downscaling device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0077] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0078] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0079] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the radar echo spatial downscaling method in the above embodiments.

[0080] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0081] The aforementioned computer-readable storage medium may be included in the radar echo spatial downscaling device; or it may exist independently and not assembled into the radar echo spatial downscaling device.

[0082] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by the radar echo spatial downscaling device, the radar echo spatial downscaling device performs the following actions: acquires low-resolution radar echo data to be processed; performs upsampling processing on the low-resolution radar echo data to be processed through interpolation to obtain high-resolution target echo data; extracts features from the high-resolution target echo data to obtain an initial high-frequency information representation map; and inputs the low-resolution radar echo data to be processed and the high-resolution target echo data into the downscaling model to obtain a high-resolution radar echo field corresponding to the low-resolution radar echo data to be processed.

[0083] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0084] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0085] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0086] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described radar echo spatial downscaling method, which can solve the technical problem of low accuracy in radar echo spatial downscaling in related technologies. Compared with related technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the radar echo spatial downscaling method provided in the above embodiments, and will not be repeated here.

[0087] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the radar echo spatial downscaling method described above.

[0088] The computer program product provided in this application can solve the technical problem of low accuracy in radar echo spatial downscaling in related technologies. Compared with related technologies, the beneficial effects of the computer program product provided in this application are the same as those of the radar echo spatial downscaling method provided in the above embodiments, and will not be repeated here.

[0089] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for spatial downscaling of radar echoes, characterized in that, The method includes: Acquire low-resolution radar echo data to be processed; The low-resolution radar echo data to be processed is upsampled by interpolation to obtain high-resolution target echo data; Feature extraction is performed on the high-resolution echo data of the target to obtain an initial high-frequency information characterization map; The low-resolution radar echo data to be processed and the initial high-frequency information characterization map are input into the downscaling model to obtain the high-resolution radar echo field corresponding to the low-resolution radar echo data to be processed. The downscaling model is constructed based on a generative network of a denoising diffusion probability model. The downscaling model is trained based on multiple sets of paired samples. The paired samples include low-resolution radar echo samples, ground truth values ​​of high-resolution radar echo samples corresponding to the low-resolution radar echo samples, and reference high-frequency information representation maps extracted from the ground truth values ​​of the high-resolution radar echo samples.

2. The radar echo spatial downscaling method as described in claim 1, characterized in that, The step of extracting features from the target high-resolution echo data to obtain an initial high-frequency information representation map includes: The high-resolution echo data of the target is processed by convolution operation using a Laplacian convolution kernel to obtain the initial high-frequency information representation map; wherein, the core calculation method is expressed as follows: in, This represents the input radar echo image. This represents the high-frequency information representation diagram obtained from the calculation. For the Laplace operator.

3. The radar echo spatial downscaling method as described in claim 1, characterized in that, The downscaling model is used for: In the reverse denoising process of the model, the interpolation results of the low-resolution radar echo data to be processed and the initial high-frequency information characterization map are used as conditions. Starting from a randomly sampled Gaussian noise tensor, multi-step iterative noise prediction and removal are performed through a multi-scale edge enhancement network to generate the high-resolution radar echo field.

4. The radar echo spatial downscaling method as described in claim 3, characterized in that, The multi-scale edge enhancement network is constructed based on a U-Net neural network structure and includes five processing layers. The multi-scale edge enhancement network is used to concatenate the interpolation result of the low-resolution radar echo data to be processed with the noise latent variable of the current step as the input of the multi-scale edge enhancement network. The initial high-frequency information representation map is fused with the features of this layer through the edge information fusion module to enhance the ability to recover the edge structure of the weak echo region in the radar echo. The edge information fusion module includes: A multi-scale feature generation unit is used to perform multi-scale depth-separable convolution processing on the initial high-frequency information representation map to obtain multi-scale features of the original features and radar echo edge map. The bidirectional attention fusion unit is used to perform cross-attention calculation on the multi-scale features as query and key-value pairs, and to perform residual connection between the attention calculation results and the corresponding query vectors to obtain the enhanced features.

5. The radar echo spatial downscaling method as described in claim 4, characterized in that, The edge information fusion module is used for: Each radar echo edge map is input into one or more processing layers corresponding to the resolution in the multi-scale edge enhancement network, and fused with the features extracted by the multi-scale edge enhancement network at that layer.

6. The radar echo spatial downscaling method as described in claim 5, characterized in that, The feature fusion is performed at multiple skip connections between the decoder and encoder of the multi-scale edge enhancement network to incorporate radar echo edge information.

7. A radar echo spatial downscaling device, characterized in that, The device includes: The acquisition module is used to acquire low-resolution radar echo data to be processed. The upsampling module is used to perform upsampling processing on the low-resolution radar echo data to be processed by interpolation to obtain high-resolution echo data of the target. The feature extraction module is used to extract features from the target high-resolution echo data to obtain an initial high-frequency information representation map; The downscaling module is used to input the low-resolution radar echo data to be processed and the initial high-frequency information characterization map into the downscaling model to obtain the high-resolution radar echo field corresponding to the low-resolution radar echo data to be processed. The downscaling model is constructed based on a generative network of a denoising diffusion probability model. The downscaling model is trained based on multiple sets of paired samples. The paired samples include low-resolution radar echo samples, ground truth values ​​of high-resolution radar echo samples corresponding to the low-resolution radar echo samples, and reference high-frequency information representation maps extracted from the ground truth values ​​of the high-resolution radar echo samples.

8. A radar echo spatial downscaling device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the radar echo spatial downscaling method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the radar echo spatial downscaling method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the radar echo spatial downscaling method as described in any one of claims 1 to 6.

Images