A position self-adaptive method and device for microwave brightness temperature fast simulation and a storage medium

Abstract

The application discloses a position adaptive method for microwave brightness temperature fast simulation, and relates to the technical field of remote sensing radiation transmission and artificial intelligence. The application adopts a position adaptive U-Net structure, while keeping the advantages of cross-scale feature fusion and skip connection of an encoder-decoder, introduces an LC non-shared convolution layer before the decoding end output, establishes a position-radiation response relationship in units of pixels, and thus significantly enhances the adaptation capability to complex land surface spatial heterogeneity and non-stationarity without explicitly introducing static priors such as surface emissivity and vegetation type. The method of the application takes meteorological variables and observation viewing angle parameters as inputs, takes satellite microwave brightness temperature as a supervision signal, is trained and verified based on the adopted L1 index, and obtains optimal weights by using a learning rate adaptation and an early stopping strategy. The application significantly reduces the processing complexity, improves the spatial structure consistency, and especially performs outstandingly in scenes such as abrupt slope direction, dry-wet transition zone and scanning edge angle.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing radiation transmission and artificial intelligence technology, and in particular to a position adaptive method, device and storage medium for rapid microwave brightness temperature simulation. Background Technology

[0002] Microwave radiative transfer models (RTMs) are the core link between satellite brightness and temperature and atmospheric-surface conditions. Existing fast RTMs (such as RTTOV / CRTM) perform well in oceans and relatively homogeneous regions, but in land areas with complex surface types and significant spatial non-stationarity, they rely on static priors such as surface emissivity and vegetation type, making them susceptible to limitations in prior resolution and accuracy, leading to problems such as large errors, spatial structure distortion, and excessively smoothed boundaries. Furthermore, polar-orbiting satellite sensors have natural coverage holes at low latitudes or at the boundary between ascending and descending orbits, further exacerbating the discontinuity and representativeness errors of the training data. Therefore, there is an urgent need for a fast simulation method for land surface microwave brightness and temperature that has stronger adaptability to surface heterogeneity and lower dependence on static priors. Summary of the Invention

[0003] This invention proposes a location-adaptive method, device, and storage medium for rapid simulation of microwave brightness temperature. It is a location-adaptive U-Net (PA-UNet) scheme, which introduces a locally-connected non-shared convolutional layer at the end of the U-Net decoding to learn independent convolutional weights and biases for each pixel location, modeling the location response relationship, reducing the dependence on surface priors (emissivity / vegetation type), realizing rapid and high-precision simulation of satellite microwave brightness temperature, and supporting operational "re-projection to observation space" evaluation and assimilation interface docking.

[0004] On the one hand, the present invention provides a position adaptive method for rapid microwave brightness temperature simulation, which includes the following steps:

[0005] Step 1, Data Construction:

[0006] A grid is established in the target area according to the set step value. Using the land reanalysis dataset (e.g., ERA5-Land data) and observation view parameters as input, six channels of data are extracted: near-surface air temperature, zonal component of horizontal wind vector, meridional component of horizontal wind vector, surface skin temperature, soil temperature, and soil moisture. The data is then matched with satellite point-by-point brightness temperature according to timestamp and latitude and longitude. The observation data is mapped to the grid to generate sample labels. Input channel data for the samples is generated based on the extracted six channels of data and satellite observation view parameters.

[0007] Step 2, Normalization and Partitioning:

[0008] The labels and input channel data of the samples are normalized to construct training and validation sets;

[0009] Step 3, construct the radiative transfer alternative model:

[0010] The radiative transfer alternative model includes an encoder and a decoder. The encoder's downsampling path consists of multiple coding units, with the first coding unit consisting of a double convolutional block (DoubleConv) and the remaining coding units consisting of cascaded downsampling layers (DOWN) and double convolutional blocks. The decoder's upsampling path consists of multiple decoding units, with the last decoding unit consisting of a locally-connected LC header and the remaining decoding units consisting of cascaded upsampling layers (UP) and double convolutional blocks.

[0011] Finally, a 1×1 convolutional layer is used to output the simulated single-channel brightness temperature;

[0012] The dual convolutional block consists of, in sequence: a 3×3 convolutional layer, a group normalization layer, and a SiLU activation function;

[0013] The downsampling layer consists of: a 3×3 convolutional layer with a stride of 2, a group normalization layer, and a SiLU activation function.

[0014] The upsampling layer uses bilinear upsampling;

[0015] The number of encoding units and decoding units remains the same, the number of channels in each encoding unit increases progressively, and the number of channels in each decoding unit decreases progressively. Furthermore, skip connections are set between encoding units and decoding units at the same level, so that the output feature map of the current level encoding unit is spliced ​​with the output feature map of the previous level decoding unit through the splicing layer to obtain the input of the current level decoding unit.

[0016] The LC head is used to learn an independent 3×3 filter and bias for each spatial location, achieving a position-dependent linear mapping;

[0017] Step 4: Train the constructed radiative transfer substitution model based on the training set and validation set. When the preset training convergence condition is met, a radiative transfer simulation model for the target area is obtained.

[0018] Furthermore, in step 1, the step value is set to 0.1° × 0.1°.

[0019] Furthermore, in step 2, the labels of the samples and the data of each input channel are normalized to [0,1] based on the statistics of the sample dataset, and then the training set and validation set are randomly divided in the time dimension according to the specified ratio.

[0020] Furthermore, the processing procedure for double convolutional blocks is as follows:

[0021]

[0022] in, , These are the input and output of the current double convolutional block, respectively, and the layer identifier. Used to distinguish different levels, functions , For the Sigmoid function, The independent variable of the function, The symbol for convolution. For grouping normalization operation, For fill operation, For a constant 1×1 projection, The weights are 3×3 cores. This is a bias term.

[0023] Furthermore, the LC head processing procedure is as follows:

[0024]

[0025] in, For the final stage decoding feature map In spatial location The pixel values ​​are given by B, where B is the batch size, C is the number of channels, and H and W are the height and width of the image. For the real number field; Indicates spatial location Pixel-level output at the location, and all The output image is composed of , This represents the initial number of channels, i.e., the number of channels in the output of the radio transmission substitution model; To avoid sharing weights, bias , To decode the feature map from the final stage The number of domains extracted.

[0026] Furthermore, in step 4, the model training uses the mean absolute error loss (MAE).

[0027] Furthermore, it also includes step 5, evaluation and business reinvestment:

[0028] On the grid established in step 1, based on the label of each grid (i.e., the gridded satellite microwave brightness temperature data), several evaluation indicators (such as MAE, RMSE, PCC, etc.) are calculated pixel by pixel.

[0029] Then, in the application scenario, the single-channel brightness temperature prediction value obtained from the radiative transfer simulation model is used for service backhaul: the distance-weighted average of grid points is performed within a specified radius (e.g., the radius is set to 50km), and then backhauled to the satellite point-by-point observation position to generate point-by-point products for interfacing with the assimilation interface.

[0030] Furthermore, the business re-projection uses a distance-weighted average, with the weight decreasing as the spherical distance from the pixel center increases.

[0031] On the other hand, the present invention also provides a computer device, the computer device including a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the position adaptive method for rapid microwave brightness temperature simulation provided by the present invention described above.

[0032] On the other hand, the present invention also provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to implement the position adaptive method for rapid microwave brightness temperature simulation provided by the present invention described above.

[0033] The technical solution provided by this invention brings at least the following beneficial effects:

[0034] (1) Low prior dependence: LC learns independent weights and biases for each pixel location, implicitly carrying the spatial differences of static attributes such as surface emissivity, vegetation type, and roughness. It can obtain low MAE simulation accuracy without explicitly inputting the above priors.

[0035] (2) Strong non-stationary adaptation: In areas with abrupt slope changes, irrigation-bare land mosaic, urban fringe, etc., PA-UNet significantly reduces over-smoothing and boundary distortion compared to shared convolution;

[0036] (3) Business-friendly: Spatial observation back projection reduces representativeness error and facilitates smooth integration with existing O-B statistical and assimilation observation operators. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0039] Figure 2This is a schematic diagram of the PA-UNet network structure;

[0040] Figure 3 This is a schematic diagram of a Locally-Connected non-shared convolutional layer (each pixel location uses an independent 3×3 kernel);

[0041] Figure 4 This is a schematic diagram of spatial projection of observations (descaled mapping from grid points to satellite geolocation).

[0042] Figure 5 This is a comparison chart of the effects on two target areas. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be described in detail and completely below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Generally, the components of the embodiments of the present invention described and shown in the accompanying drawings can be arranged and designed using different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of the present invention.

[0044] This invention provides a location-adaptive method for rapid microwave brightness temperature simulation, which is a method for a location-adaptive PA-UNet radiative transfer substitution model for rapid microwave brightness temperature simulation. It achieves rapid brightness temperature simulation based on multi-source data for land areas, a symmetric encoding and decoding U-Net backbone network, and location-adaptive non-shared convolutional layers. This method integrates six channels (near-surface air temperature, zonal component of the horizontal wind vector, meridional component of the horizontal wind vector, surface skin temperature, soil temperature, and soil moisture) of the global high-resolution land reanalysis dataset ERA5-Land with satellite observation viewpoint parameters on a 0.1°×0.1° (64×64) regular grid. By introducing LocallyConnected non-shared convolutions at the end of the U-Net decoding, an independent 3×3 filter and bias are learned for each pixel location, thereby modeling the location response relationship and reducing dependence on surface priors. A lightweight LC (LocallyConnected) head is used to adaptively characterize the non-stationary and heterogeneous features of the land surface. Training employs L1 (MAE) loss and AdamW optimization, combined with learning rate scheduling and early stopping. Evaluation is performed on a 64×64 grid, comparing the results pixel-by-pixel with gridded satellite microwave brightness temperature labels (MAE / RMSE / PCC). In application scenarios, grid predictions can be averaged and back-projected to point-by-point observations within a range of approximately 50km, achieving seamless integration with the assimilation interface. This significantly improves the accuracy of brightness temperature simulation and the consistency of spatial structure while ensuring efficiency.

[0045] like Figure 1 As shown, as one possible implementation, the specific implementation steps of a position adaptive method for rapid microwave brightness temperature simulation provided by this embodiment of the invention include:

[0046] Step S1: Acquire raw satellite data and REA5-Land data. In this embodiment, acquire satellite passive microwave brightness temperature data, satellite geolocation data, satellite observation angle parameters, and REA5-Land data (near-surface air temperature, zonal component of horizontal wind vector, meridional component of horizontal wind vector, surface skin temperature, soil temperature, and soil moisture).

[0047] Step S2: Preprocess the satellite data and ERA5-Land data.

[0048] Based on the required simulated brightness temperature areas, the satellite passive microwave brightness temperature data obtained by the microwave imager is preprocessed to remove outlier dates and unwanted areas.

[0049] This embodiment uses satellite passive microwave brightness temperature data products as the data source. First, an index is created by date. Data files containing daily global coverage (both ascent and descent) are archived and parsed to extract accessible indexes and data records. These records include environmental / instrument information, satellite brightness temperature data, and metadata such as available quality indicators and observation angle parameters. Then, spatial and temporal filtering is performed based on the mission area boundaries, retaining only relevant observations. The MWRI scan time is matched with the most recent time period in the ERA5-Land hourly field to ensure consistent driving label timing. To adapt to network training, point-by-point observations are mapped to a 0.1°×0.1° regular grid (64×64) according to latitude and longitude. If multiple pixels exist within the same grid, a distance-weighted average is taken, and bilinear interpolation is used to form a 1×64×64 gridded brightness temperature label. Simultaneously, a distance-weighted average of the incident angle is performed and used as the 7th input channel. Finally, brightness temperature, latitude and longitude, timestamps, and incident angles are organized and saved as data, forming a sample set that can be directly used for PA-UNet training and evaluation.

[0050] Step S3, normalization and data partitioning: perform min-max normalization to [0,1] on all input channels and labels according to the training set statistics; partition the training / validation / test sets along the time dimension in a 3:1:1 ratio.

[0051] Step S4: Construct and train a brightness temperature simulation network.

[0052] Encoding end: 3×3 convolution with stride=2 and reflection padding are used instead of max pooling. GroupNorm+SiLU is used for normalization / activation to enhance the stability and edge preservation ability of mini-batch training.

[0053] Decoding end: Bilinear upsampling is used and skipped connections are made with symmetric coding layers to fuse multi-scale context and details;

[0054] Position Adaptive Header: After the last upsampling, a LocallyConnected non-shared convolutional layer is cascaded to learn an independent 3×3 convolutional kernel and bias (without weight sharing) for each spatial location. Then, a 1×1 convolutional output is used to simulate brightness temperature in a single channel to model the position response relationship.

[0055] The L1 (MAE) loss function is used as the primary loss. The optimizer uses AdamW (weight decay = 1e-4, initial learning rate = 1e-4), combined with ReduceLROnPlateau (patience = 4, minimum learning rate = 1e-6) and early stopping (stopping after 20 epochs of validation without improvement). The weights with the lowest MAE in the validation are saved as the final model for testing and inference.

[0056] Step S5: Evaluation and application of feedback.

[0057] Assessment criteria: On a 64×64 grid, MAE, RMSE, and PCC are calculated pixel by pixel with the gridded satellite microwave brightness temperature data, and can be grouped and statistically analyzed according to surface type, etc.

[0058] Application scope: To facilitate product release or interface assimilation, for each satellite passive microwave brightness temperature observation point, grid prediction values ​​within a spherical radius of approximately 50km are collected, and a distance-weighted average is performed to obtain the return brightness temperature of that point. O-B statistics and operational monitoring are used; this result is not used as the main evaluation indicator in this paper.

[0059] In the position adaptive method for rapid microwave brightness temperature simulation proposed in this embodiment, the specific implementation of the radiative transfer substitution model is as follows: Figure 2As shown, the network consists of a symmetric encoder-decoder backbone, an LC non-shared convolutional position adaptive head, and a 1×1 output mapping. The encoder extracts semantic features by progressively downsampling 3×3 convolutions with a stride of 2. First, multi-scale semantic features are extracted by progressively downsampling through the encoder branches. Each basic unit is a "double convolutional block (Conv3×3→GroupNorm→SiLU→Conv3×3→GroupNorm→SiLU)," and max pooling is replaced with 3×3 convolutions with a stride of 2 and reflection padding to reduce boundary artifacts. The number of channels at each level increases from 64 to 128 to 256 to 512. The design employs incremental (example configuration) methods. At the decoding end, progressive bilinear upsampling is used, and the features from the symmetric coding layer are fused at the splicing layer via skip connections (Skip-Fusion). After fusion, cross-scale information reconstruction is achieved through "convolution + GroupNorm + SiLU," resulting in a feature map that combines global semantics and local texture. To characterize the statistical differences between different geographical locations, a non-shared convolutional LC position adaptive head is introduced after the final upsampling. This head learns an independent 3×3 filter and bias for each pixel location, and finally, a 1×1 convolution is used to compress the channels to 1, yielding a single-channel brightness temperature prediction map. This design retains the advantages of U-Net's cross-scale feature fusion while modeling the non-stationary relationship of positional radiation response through the position adaptive head. This structure, while maintaining the advantages of U-Net's multi-scale feature fusion, achieves pixel-level position adaptive correction through Locally-Connected non-shared convolutional layers. This effectively improves numerical accuracy and spatial structure consistency in complex land surface scenes without explicitly introducing static priors such as emissivity / vegetation type, thus achieving higher fitting accuracy and spatial structure fidelity in areas with terrain undulations and surface type mosaics.

[0060] In this invention, each convolutional unit is filled with a pre-reflection filter, followed by a 3×3 convolution, GroupNorm, and SiLU activation. When the input / output channels are inconsistent, the bypass uses a 1×1 convolution for linear projection and is added to the main channel to enhance gradient stability. Its mathematical expression is (as shown in the example below). (For example, a layer)

[0061]

[0062] in, Represents convolution. For grouping normalization operation, , For the Sigmoid function, For a constant 1×1 projection, The weights are 3×3 cores. For bias terms, , The first Layers and The output feature map of the layer, which is the input and output of the current double convolution; This is a fill operation.

[0063] In this invention example, the position-adaptive LC layer is a non-shared convolutional structure, such as... Figure 3 As shown. Let the final-stage decoding feature map... (Default H=W=64). Where B is the batch size, C is the number of channels, and H×W is the spatial scale, i.e., height and width. For real numbers. After reflection filling, use PyTorch's Unfold function to... Extract each position Neighborhood ( =3) Vector, denoted as For each spatial location Learn the non-shared weights separately With bias The pixel-level output is:

[0064]

[0065] in, Indicates the initial number of channels. This is the final output image.

[0066] Because the weights are explicitly sensitive to spatial location, the LC layer can be viewed as a pixel-level adaptive correction operator at the decoding end. Its position-dependent linear mapping implicitly carries spatial differences in static attributes such as terrain aspect, surface type / roughness, and vegetation density during the learning process, reducing spatial averaging and excessive boundary smoothing caused by shared convolution. Therefore, for each satellite passive microwave brightness temperature observation point, grid prediction values ​​within a spherical radius of approximately 50km are collected, and a distance-weighted average is performed to obtain the projected brightness temperature at that point. Figure 4 The grid points shown represent the descaling mapping to satellite geolocation.

[0067] Based on the method of the embodiments of the present invention, the following can be obtained: Figure 5The comparison charts of the two target regions show that, in test region one (120°E–126.3°E, 45°N–51.3°N), compared to RTTOV, MAE / RMSE decreased from 5.657 / 7.345 K to 2.019 / 2.638 K (a decrease of approximately 64%), while PCC increased to 0.962; in test region two (120°W–113.7°W, 45°N–51.3°N), MAE / RMSE decreased to 1.323 / 1.694 K (a decrease of 67%), while PCC increased to 0.984. This means that the method in this embodiment significantly improves the corresponding indicators. The evaluation indicators include, but are not limited to, MAE, RMSE, and PCC, and compared to RTTOV in the two regions, MAE and RMSE are reduced by at least 60%, and PCC is increased by at least 0.05.

[0068] In this embodiment of the invention, the pixel-level loss (L1 / MAE) adopted is as follows:

[0069]

[0070] in, For the first The spatial location of each sample The predicted values ​​(pixel values ​​output by the model) on the model. This corresponds to the actual value (pixel value of the reference image).

[0071] In this example, MAE, RMSE, and Pearson correlation coefficient (PCC) are used to objectively evaluate the model's accuracy and spatial structure consistency.

[0072]

[0073]

[0074]

[0075] in, For the first The predicted value for each sample, This corresponds to the actual value. , These are the standard deviations of the predicted value and the actual value, respectively.

[0076] In an exemplary embodiment, the present invention also provides a computer device including a processor and a memory, the memory storing at least one computer program. The at least one computer program is loaded and executed by one or more processors to implement any of the above-described position adaptive methods for rapid microwave brightness temperature simulation.

[0077] In an exemplary embodiment, the present invention also provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor of a computer device to implement any of the above-described position adaptive methods for rapid microwave brightness temperature simulation.

[0078] In one possible implementation, the aforementioned computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, and an optical data storage device, etc.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0080] The above descriptions are merely some embodiments of the present invention. Those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A position-adaptive method for rapid microwave brightness temperature simulation, characterized in that, Includes the following steps: Step 1, Data Construction: A grid is established in the target area according to the set step value. Using the land reanalysis dataset and observation view parameters as input, six channels of data are extracted: near-surface air temperature, zonal component of horizontal wind vector, meridional component of horizontal wind vector, surface skin temperature, soil temperature, and soil moisture. The data is then matched with satellite point-by-point brightness temperature according to timestamp and latitude and longitude. The observation data is mapped to the grid to generate sample labels. Input channel data for the samples is generated based on the extracted six channels of data and satellite observation view parameters. Step 2, Normalization and Partitioning: The labels and input channel data of the samples are normalized to construct training and validation sets; Step 3, construct the radiative transfer alternative model: The radiative transfer alternative model includes an encoder and a decoder. The encoder's downsampling path consists of multiple coding units, with the first coding unit consisting of a double convolutional block and the remaining coding units consisting of cascaded downsampling layers and double convolutional blocks. The decoder's upsampling path consists of multiple decoding units, with the last decoding unit consisting of an LC header and the remaining decoding units consisting of cascaded upsampling layers and double convolutional blocks. Finally, a 1×1 convolutional layer is used to output the simulated single-channel brightness temperature; The dual convolutional block consists of, in sequence: a 3×3 convolutional layer, a group normalization layer, and a SiLU activation function; The downsampling layer consists of: a 3×3 convolutional layer with a stride of 2, a group normalization layer, and a SiLU activation function. The upsampling layer uses bilinear upsampling; The number of encoding units and decoding units remains the same, the number of channels in each encoding unit increases progressively, and the number of channels in each decoding unit decreases progressively. Furthermore, skip connections are set between encoding units and decoding units at the same level, so that the output feature map of the current level encoding unit is spliced ​​with the output feature map of the previous level decoding unit through the splicing layer to obtain the input of the current level decoding unit. The LC head is used to learn an independent 3×3 filter and bias for each spatial location to achieve a location-dependent linear mapping; Step 4, the constructed radiative transfer substitution model is trained based on the training set and validation set. When the preset training convergence condition is met, a radiative transfer simulation model for the target region is obtained.

2. The method as described in claim 1, characterized in that, In step 1, the step value is set to 0.1° × 0.1°.

3. The method as described in claim 1, characterized in that, In step 2, the labels of the samples and the data of each input channel are normalized to [0,1] based on the statistics of the sample dataset. Then, the training set and the validation set are randomly divided in the time dimension according to the specified ratio.

4. The method as described in claim 1, characterized in that, The processing procedure for double convolutional blocks is as follows: ； in, , These are the input and output of the current double convolutional block, respectively, and the layer identifier. Used to distinguish different levels, functions , For the Sigmoid function, The independent variable of the function, The symbol for convolution. For grouping normalization operation, For fill operation, For a constant 1×1 projection, The weights are 3×3 cores. This is a bias term.

5. The method as described in claim 1, characterized in that, The processing procedure for the LC head is as follows: ； in, For the final stage decoding feature map In spatial location The pixel values ​​are given by B, where B is the batch size, C is the number of channels, and H and W are the height and width of the image. For the real number field; Indicates spatial location Pixel-level output at the location, and all The output image is composed of , Indicates the initial number of channels; To avoid sharing weights, bias , To decode the feature map from the final stage The number of domains extracted.

6. The method as described in claim 1, characterized in that, In step 4, the model training uses the mean absolute error loss.

7. The method as described in claim 1, characterized in that... It also includes step 5, evaluation and business reinvestment: On the specification grid established in step 1, several evaluation metrics are calculated pixel by pixel based on the label of each grid. Then, in the application scenario, the single-channel brightness temperature prediction value obtained based on the radiative transfer simulation model is used for service retransmission: the distance-weighted average of grid points within a specified radius is performed, and then the data is retransmitted to the satellite point-by-point observation position to generate point-by-point products.

8. The method as described in claim 7, characterized in that, The business return projection uses a distance-weighted average, with the corresponding weight decreasing as the spherical distance from the pixel center increases.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method of claim 1.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 1.

Images