InSAR atmospheric delay correction method, device, equipment and medium

Abstract

The embodiment of the invention discloses an InSAR (Interferometric Synthetic Aperture Radar) atmospheric delay correction method, device and equipment and a medium. A specific embodiment of the method comprises the following steps: extracting atmospheric delay parameter data, and performing data dimension processing on target elevation data to generate elevation processing data; performing interference processing on the target SAR image to generate an interference pattern; for each wavelet transformation mode, performing high-frequency component feature extraction and low-frequency component feature extraction on the interferogram; generating auxiliary matrix data; performing feature information fusion on the at least one piece of high-frequency component feature information, and performing feature information fusion on the at least one piece of low-frequency component feature information; and generating an interference image after atmospheric delay correction by using the atmospheric delay correction neural network model. According to the embodiment, atmospheric correction can be quickly and accurately carried out on the target interferogram through the preset data preprocessing mode and the atmospheric delay correction neural network model, so that the high-quality interference image after atmospheric delay correction is generated.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of computer and remote sensing technology, and in particular to an InSAR atmospheric delay correction method, apparatus, device, and medium. Background Art

[0002] Currently, spaceborne synthetic aperture radar (InSAR) interferometry is an advanced space geodetic technique. By comparing the phase information between multiple coherent radar images, it can measure minute surface deformation changes. This technique has been widely used in areas such as surface deformation monitoring, plate tectonic movement, landslides, and glacier movement. However, the presence of atmospheric delay seriously hinders the further development and application of InSAR technology. Atmospheric delay refers to the differences in the propagation paths of electromagnetic waves through the Earth's atmosphere caused by spatiotemporal variations in atmospheric conditions (such as air pressure, water vapor content, free electron density, and temperature). Ultimately, these differences are reflected in the phase information of the interferogram, severely interfering with the accuracy of InSAR surface deformation monitoring. In extreme cases, it can even render InSAR deformation monitoring completely ineffective. To address atmospheric delay, a common approach is to establish a functional relationship between the tropospheric delay phase and surface elevation to eliminate the impact of the tropospheric delay phase.

[0003] However, when using the above method to solve the atmospheric delay problem, the following technical problems often occur:

[0004] There are certain limitations in solving the atmospheric delay problem through functional relationships. It can often only correct the delay phase that is highly correlated with the assumptions in the model. The correction effect on the turbulent part of the atmospheric delay with strong temporal and spatial variability is limited, resulting in the inability to achieve accurate atmospheric delay processing.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the Invention

[0006] The content of this disclosure is used to briefly introduce concepts that will be described in detail in the detailed description section below. The content of this disclosure is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0007] Some embodiments of the present disclosure provide methods, devices, equipment, and media for correcting InSAR atmospheric delays to solve the technical problems mentioned in the above background technology section.

[0008] In a first aspect, some embodiments of the present disclosure provide an InSAR atmospheric delay correction method, comprising: acquiring a target image, target meteorological data, and target elevation data taken by a target synthetic aperture radar for a target area; extracting atmospheric delay parameter data from the target meteorological data, and performing data dimension processing on the target elevation data to generate elevation processing data; performing interference processing on the target image to generate an interference pattern; for each wavelet transform method in at least one wavelet transform method, using the wavelet transform method to extract high-frequency component features and low-frequency component features from the interference pattern to generate high-frequency components. characteristic information and low-frequency component characteristic information; performing data interpolation and fusion on the above-mentioned elevation processing data and the above-mentioned atmospheric delay parameter data to generate auxiliary matrix data of the same size as the matrix corresponding to the above-mentioned interference pattern; performing feature information fusion on the obtained at least one high-frequency component characteristic information to generate first fused feature information, and performing feature information fusion on the obtained at least one low-frequency component characteristic information to generate second fused feature information; inputting the above-mentioned first fused feature information, the above-mentioned second fused feature information and the above-mentioned auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay corrected image for the above-mentioned target image.

[0009] In a second aspect, some embodiments of the present disclosure provide an InSAR atmospheric delay correction device, comprising: an acquisition unit configured to acquire a target image, target meteorological data and target elevation data taken by a target synthetic aperture radar for a target area; a processing unit configured to extract atmospheric delay parameter data from the target meteorological data data, and perform data dimension processing on the target elevation data to generate elevation processing data; an interference processing unit configured to perform interference processing on the target image to generate an interference pattern; an extraction unit configured to perform high-frequency component feature extraction and low-frequency component feature extraction on the interference pattern using the wavelet transform method for each wavelet transform method in at least one wavelet transform method to generate an interference pattern. High-frequency component feature information and low-frequency component feature information; a first fusion unit is configured to perform data interpolation and fusion on the above-mentioned elevation processing data and the above-mentioned atmospheric delay parameter data to generate auxiliary matrix data of the same size as the matrix corresponding to the above-mentioned interference pattern; a second fusion unit is configured to perform feature information fusion on the at least one high-frequency component feature information obtained to generate first fused feature information, and perform feature information fusion on the at least one low-frequency component feature information obtained to generate second fused feature information; a generating unit is configured to input the above-mentioned first fused feature information, the above-mentioned second fused feature information and the above-mentioned auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay corrected image for the above-mentioned target image.

[0010] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0011] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.

[0012] The above-described embodiments of the present disclosure have the following beneficial effects: The InSAR atmospheric delay correction methods of some embodiments of the present disclosure can accurately perform atmospheric correction on target images to generate high-quality synthetic aperture radar images. Specifically, the reason for the inaccurate atmospheric correction is that the use of functional relationships to address atmospheric delay has certain limitations. This often only corrects delay phases that are highly correlated with model assumptions. The correction effect on the turbulent portion of the atmospheric delay, which exhibits strong temporal and spatial variability, is limited, resulting in an inability to accurately process atmospheric delay. Based on this, the InSAR atmospheric delay correction methods of some embodiments of the present disclosure first obtain a target image, target meteorological data, and target elevation data for a target area captured by a target synthetic aperture radar. Acquiring a synthetic aperture radar image of the target area facilitates the subsequent generation of an interferogram within the target area. Acquiring the target meteorological data can provide meteorological information within the target area, which can be used to provide meteorological auxiliary data during the subsequent atmospheric delay correction process. Acquiring the target elevation data can provide terrain information within the target area, which can be used to provide meteorological auxiliary data during the subsequent atmospheric delay correction process. Then, atmospheric delay parameter data is extracted from the target meteorological data, and the target elevation data is subjected to data dimension processing to generate elevation processing data. Here, by extracting atmospheric delay parameter data from the target meteorological data, key meteorological parameter content can be extracted. By performing data dimension processing on the target elevation data, the phase information corresponding to the subsequent matching interference pattern can be matched. Next, interference processing is performed on the target image to generate an interference pattern for subsequent wavelet transform processing. For each of at least one wavelet transform method, the wavelet transform method is used to extract high-frequency component features and low-frequency component features from the interference pattern to generate high-frequency component feature information and low-frequency component feature information. Here, based on the large frequency differences in the interference pattern, by extracting high-frequency component feature information and low-frequency component feature information, the subsequent atmospheric delay correction neural network model can accurately generate an atmospheric delay correction image based on feature information of different frequencies. Data interpolation and fusion are performed on the processed elevation data and the atmospheric delay parameter data to generate auxiliary matrix data of the same size as the matrix corresponding to the interference pattern, which serves as auxiliary feature information to subsequently accurately generate an atmospheric delay correction image. Furthermore, feature information fusion is performed on the obtained at least one high-frequency component feature information to generate first fused feature information, and feature information fusion is performed on the obtained at least one low-frequency component feature information to generate second fused feature information, thereby obtaining more accurately composite feature content of the high-frequency features and feature content of the low-frequency features.Finally, the first fused feature information, the second fused feature information, and the auxiliary matrix data are input into a pre-trained atmospheric delay correction neural network model to accurately generate an atmospheric delay correction image for the target image. In summary, by pre-processing the target image, target meteorological data, and target elevation data, and extracting composite high-frequency and low-frequency feature information, and based on the atmospheric delay correction neural network model, an atmospheric delay correction image can be accurately generated. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0014] Figure 1 is a flow chart of some embodiments of the InSAR atmospheric delay correction method according to the present disclosure;

[0015] Figure 2 is a schematic structural diagram of some embodiments of the InSAR atmospheric delay correction device according to the present disclosure;

[0016] Figure 3 is a schematic structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure;

[0017] Figure 4 is a schematic diagram illustrating model training and model application of an atmospheric delay correction neural network model according to the present disclosure;

[0018] Figure 5 is a schematic diagram illustrating generation of second fused feature information according to the present disclosure;

[0019] Figure 6 is a schematic diagram illustrating generation of a corrected interferogram according to the present disclosure. DETAILED DESCRIPTION

[0020] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0021] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other.

[0022] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0023] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0024] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0025] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0026] refer to Figure 1 , shows a process 100 of some embodiments of the InSAR atmospheric delay correction method according to the present disclosure. The InSAR atmospheric delay correction method includes the following steps:

[0027] Step 101: Acquire a target image, target meteorological data, and target elevation data of a target area captured by a target synthetic aperture radar.

[0028] In some embodiments, in response to receiving session communication request information for the first user terminal and the second user terminal, the execution subject (e.g., an electronic device) of the above-mentioned InSAR atmospheric delay correction method can obtain a target image, target meteorological data, and target elevation data for the target area captured by the target synthetic aperture radar through a wired connection or a wireless connection. The target image can be an image captured by a synthetic aperture radar (SAR) for the target area to be subjected to atmospheric delay correction. Atmospheric delay refers to the difference in the path of electromagnetic wave signals propagating in the atmosphere due to the spatiotemporal changes in atmospheric conditions (parameters such as air pressure, water vapor content, free electron density, and temperature) when electromagnetic waves pass through the Earth's atmosphere. Ultimately, this difference will be reflected in the phase information of the interferogram, seriously interfering with the accuracy of surface deformation monitoring using InSAR technology, and in extreme cases even causing the complete failure of InSAR deformation monitoring. The target meteorological data can be meteorological data for the current area. In practice, meteorological data for the target area can be obtained through the Generic Atmospheric Correction Online Service (GACOS, an online service platform for atmospheric correction). The target meteorological data can have a temporal resolution of minutes and a spatial resolution of 90 meters. The target elevation data can be elevation data within the target area. The elevation data can be DEM (Digital Elevation Model) data. DEM data is a digital terrain model used to represent the elevation of the Earth's surface. DEM data is a model that digitally represents the topography of the Earth's surface, simulating changes in ground elevation through a series of ordered numerical arrays.

[0029] Step 102 : extracting atmospheric delay parameter data from the target meteorological data, and performing data dimension processing on the target elevation data to generate elevation processing data.

[0030] In some embodiments, the execution entity may extract atmospheric delay parameter data from the target meteorological data, and perform data dimension processing on the target elevation data to generate elevation processing data. The atmospheric delay parameter data may be at least one parameter data that significantly affects atmospheric delay. In practice, the atmospheric delay parameter data may include water vapor content, temperature, and air pressure parameters. The data dimension processing may be a phase height conversion of the elevation data to unify the dimension of the elevation data with the dimension of the phase data in the interferogram. That is, the dimension corresponding to the elevation processing data is the same as the dimension of the phase data in the interferogram.

[0031] In some optional implementations of some embodiments, the execution entity may extract atmospheric delay parameter data from the target meteorological data, and perform data dimension processing on the target elevation data to generate elevation processed data, including the following steps:

[0032] In the first step, for each raster data in the target elevation data, perform the following generation steps:

[0033] Sub-step 1: Divide the target value by the radar wavelength to generate a first division result. The target value may be 4π. The radar wavelength refers to the wavelength of the electromagnetic wave used in the radar system. The target elevation data may be in raster format. Specifically, the target elevation data includes multiple raster data.

[0034] Sub-step 2: Multiply the satellite distance corresponding to the raster data by the sine value corresponding to the radar wave incident angle to obtain a multiplication result. The satellite distance may be the distance between the satellite and the ground corresponding to the raster data.

[0035] Sub-step 3: Divide the vertical baseline length corresponding to the above-mentioned raster data by the above-mentioned multiplication result to obtain a second division result. The vertical baseline length refers to the distance component between two antennas (or radar sensors) in the vertical line of sight direction (i.e., perpendicular to the surface target) in an InSAR system. The vertical baseline length is relative to the parallel baseline (i.e., the component of the baseline in the line of sight slant range direction).

[0036] Sub-step 4: multiplying the first division result, the second division result, and the ground height corresponding to the raster data to obtain processed raster data. The ground height may be the height of a ground protrusion corresponding to the raster data.

[0037] In the second step, the processed raster dataset is determined as the above-mentioned elevation processing data.

[0038] Step 103: performing interference processing on the target image to generate an interference pattern.

[0039] In some embodiments, the execution entity may perform interference processing on the target image by a conventional interference processing method to generate an interference pattern.

[0040] Step 104 : For each of at least one wavelet transform mode, use the wavelet transform mode to extract high-frequency component features and low-frequency component features from the interference pattern to generate high-frequency component feature information and low-frequency component feature information.

[0041] In some embodiments, the execution entity may utilize each of at least one wavelet transform method to extract high-frequency and low-frequency component features from the interferogram, thereby generating high-frequency and low-frequency component feature information. The high-frequency feature information may be feature content corresponding to the high-frequency signal in the interferogram. The low-frequency feature information may be feature content corresponding to the low-frequency signal in the interferogram. The at least one wavelet transform method may include, but is not limited to, at least one of the following: a Daubechies wavelet function, a coiflet (Coiflet Wavelet) wavelet function, or a Haar (Haar Wavelet Basis Function) wavelet function. The Haar wavelet basis function not only provides good temporal resolution but also exhibits greater sensitivity to sudden changes in atmospheric delay signals, which are often present in atmospheric delay data. The Daubechies wavelet function has higher smoothness and can process smooth signals, corresponding to the stationary portion of atmospheric delay signals. Furthermore, the Daubechies wavelet function has stronger analytical capabilities for signals rich in detail, preserving the rich detail in atmospheric delay data. The coiflet wavelet function is well-suited for analyzing complex signals due to its superior smoothness. Considering that atmospheric delay patterns are often complex in areas with rapidly changing meteorological conditions, the coiflet wavelet basis function offers unique advantages in capturing complex signal patterns. High-frequency component features provide rich atmospheric delay details within the interferogram, while low-frequency component features provide background information on the large-scale atmospheric delay profile.

[0042] Here, considering that the atmospheric delay signals in the obtained interferogram have significant differences in the frequency domain, the significant signals are separated through preprocessing to improve the accuracy of subsequent atmospheric correction.

[0043] Step 105 : performing data interpolation and fusion on the elevation processing data and the atmospheric delay parameter data to generate auxiliary matrix data of the same size as the matrix corresponding to the interference pattern.

[0044] In some embodiments, the execution entity may perform data interpolation and fusion on the elevation processing data and the atmospheric delay parameter data to generate auxiliary matrix data having the same size as the matrix corresponding to the interference pattern.

[0045] Step 106 : Fusing the obtained at least one high-frequency component feature information to generate first fused feature information, and fusing the obtained at least one low-frequency component feature information to generate second fused feature information.

[0046] In some embodiments, the execution entity may fuse the obtained at least one high-frequency component feature information to generate first fused feature information, and fuse the obtained at least one low-frequency component feature information to generate second fused feature information.

[0047] As an example, the execution entity may perform feature information averaging processing on at least one high-frequency component feature information to generate first fused feature information, and perform averaging processing on at least one low-frequency component feature information to generate second fused feature information.

[0048] Step 107 : Input the first fused feature information, the second fused feature information and the auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay correction image for the target image.

[0049] In some embodiments, the execution entity may input the first fused feature information, the second fused feature information, and the auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay corrected image for the target image. The atmospheric delay corrected image may be a synthetic aperture radar image after atmospheric delay correction. The atmospheric delay correction neural network model may be a neural network model that performs atmospheric delay correction. For example, the atmospheric delay correction neural network model may be a plurality of convolutional layers connected in series.

[0050] In some optional implementations of some embodiments, the atmospheric delay correction neural network model includes: a multi-branch feature code extractor, a feature fusion network, and a multi-scale regression network. The multi-branch feature code extractor may be a feature code extractor with multiple branches. The feature code extractor may be an encoder that extracts semantic content of features. The multi-branch feature code extractor may include: a first feature code extractor corresponding to the first fused feature information, a second feature code extractor corresponding to the second fused feature information, and a third feature code extractor corresponding to the auxiliary matrix data. The number of convolutional layers corresponding to the first, second, and third feature code extractors may vary and may be adaptively adjusted. The feature fusion network may be a network layer that fuses feature information from each input feature information. The multi-branch feature code extractor extracts high-frequency detail component channels, low-frequency background component channels, and auxiliary component channels of meteorological and terrain information from the input data. Considering the multi-scale information present in the atmospheric delay features, as well as large-scale transformation information with a wide spatial distribution range and localized change information in small areas, a corresponding multi-scale regression network is provided to perform multi-scale regression analysis on the features to accurately capture the changes in the corresponding patterns. The multi-scale regression network consists of different branch structures with convolution kernel sizes of 1, 3, and 5. These kernels process information at different scales, thereby obtaining richer feature information at different scales. Finally, the features obtained from different branches are fused, normalized, and activated to obtain the final output features of the model.

[0051] Optionally, the execution entity may input the first fused feature information, the second fused feature information, and the auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay corrected image for the target image, comprising the following steps:

[0052] In the first step, the first fused feature information, the second fused feature information, and the auxiliary matrix data are input into the multi-branch feature code extractor to generate a multi-channel feature information set. The multi-channel feature information set may include feature code information corresponding to the first fused feature information, feature code information corresponding to the second fused feature information, and feature code information corresponding to the auxiliary matrix data.

[0053] In the second step, the multi-channel feature information set is input into the feature fusion network to generate fused feature information.

[0054] The third step is to input the above fusion feature information into the above multi-scale regression network to generate the above atmospheric delay correction image.

[0055] In some optional implementations of some embodiments, the multi-branch feature code extractor includes: a low-frequency feature extraction module, a high-frequency feature extraction module, and an auxiliary data feature extraction module. The low-frequency feature extraction module may be a network layer that extracts low-frequency feature semantic content from low-frequency signals. The high-frequency feature extraction module may be a network layer that extracts high-frequency feature semantic content from high-frequency signals. The auxiliary data feature extraction module may be a network layer that extracts specific feature semantic content from auxiliary data. The low-frequency feature extraction module is used to extract large-scale contour information from the interferometric phase image, requiring branches with large-scale feature capture capabilities. Therefore, the basic module of the low-frequency feature extraction module consists of a 7x7 convolution kernel, a normalization layer, an activation function, and a pooling layer in sequence. Since low-frequency feature information is mostly large-scale, contour-level information with relatively simple features, a deeper network structure is not required to extract high-level, abstract features. Therefore, to balance computational efficiency and fitting accuracy, the basic module of the low-frequency feature extraction module is repeated three times to obtain the low-frequency feature extraction module. Since the high-frequency signal primarily contains information about the detailed changes in the interferometric pattern, the spatial variation scale of this detail information is relatively small, so a larger convolution kernel is not required. On the other hand, detailed information contains a large amount of information, requiring a deeper network structure to extract more abstract, advanced, and robust features. The high-frequency feature extraction module is optimized using a 3x3 convolution kernel, a deeper network structure, and a larger number of basic channels. The basic module of the high-frequency feature extraction module consists of a 3x3 convolution layer, a batch normalization (BN) layer, a Gaussian Error Linear Unit (Gelu) activation function, and a channel attention layer. The high-frequency feature extraction module uses a basic channel number of 64 and a repetition depth of 5. The input data corresponding to the auxiliary data feature extraction module has not only higher spatial resolution but also significantly increased data complexity. Therefore, the auxiliary data feature extraction module uses a combination of dilated convolution, batch normalization, Gaussian activation function, and global average pooling as the basic module of the auxiliary information branch. The dilated convolution can limit the expansion of the receptive field, maintain data resolution, and perform multi-scale feature extraction, making full and efficient use of the input data. The input data of the auxiliary data feature extraction module contains a large amount of information. Therefore, the number of basic channels corresponding to the auxiliary data feature extraction module is 64. The branch basic module is repeated 5 times to construct a larger branch structure to obtain high-level abstract features. After each basic module is executed, a channel attention mechanism module is used to adjust the importance of different feature channels, determine the optimal weight ratio relationship between the target image, target meteorological data and target elevation data, and ensure the accuracy of subsequent atmospheric correction.

[0056] Optionally, the execution entity may input the first fused feature information, the second fused feature information, and the auxiliary matrix data into the multi-branch feature code extractor to generate a multi-channel feature information set, including the following steps:

[0057] In the first step, the first fusion feature information is input into the low-frequency feature extraction module to generate low-frequency channel feature information.

[0058] In the second step, the second fusion feature information is input into the high-frequency feature extraction module to generate high-frequency channel feature information.

[0059] The third step is to input the auxiliary matrix data into the auxiliary data feature extraction module to generate auxiliary channel feature information.

[0060] The fourth step is to generate the multi-channel feature information set according to the low-frequency channel feature information, the high-frequency channel feature information and the auxiliary channel feature information.

[0061] As an example, the execution entity may directly combine the low-frequency channel feature information, the high-frequency channel feature information, and the auxiliary channel feature information to generate a multi-channel feature information set.

[0062] In some optional implementations of some embodiments, the execution entity may input the multi-channel feature information set into the feature fusion network to generate fused feature information, including the following steps:

[0063] In the first step, the multi-channel feature information set is input into the average pooling layer included in the feature fusion network to generate an average pooling result.

[0064] In the second step, the above average pooling results are input into the attention weight calculation layer included in the above feature fusion network to generate attention feature information.

[0065] In the third step, the attention feature information is input into the multi-layer perceptron included in the feature fusion network to generate a feedforward output result.

[0066] In the fourth step, the above feedforward output result is input into the batch normalization layer included in the above feature fusion network to generate a normalized result.

[0067] The fifth step is to input the above normalized result into the activation function layer included in the above feature fusion network to generate the above fusion feature information. And

[0068] In some optional implementations of some embodiments, the execution entity may input the fused feature information into the multi-scale regression network to generate the atmospheric delay corrected image, including the following steps:

[0069] In the first step, the fused feature information is input into the convolution layer at the first convolution kernel scale included in the multi-scale regression network to generate a first convolution result.

[0070] In the second step, the fused feature information is input into a convolution layer at a second convolution kernel scale included in the multi-scale regression network to generate a second convolution result, wherein the second convolution kernel scale is higher than the first convolution kernel scale.

[0071] In the third step, the fused feature information is input into the convolution layer at the third convolution kernel scale included in the multi-scale regression network to generate a third convolution result, wherein the third convolution kernel scale is higher than the second convolution kernel scale.

[0072] In the fourth step, the fused feature information is input into the convolution layer at the fourth convolution kernel scale included in the multi-scale regression network to generate a fourth convolution result, wherein the fourth convolution kernel scale is higher than the third convolution kernel scale.

[0073] In the fifth step, the second convolution result is input into the convolution layer at the first convolution kernel scale included in the multi-scale regression network to generate a fifth convolution result.

[0074] In the sixth step, the third convolution result is input into the convolution layer at the first convolution kernel scale included in the multi-scale regression network to generate a sixth convolution result.

[0075] In the seventh step, the fourth convolution result is input into the convolution layer at the first convolution kernel scale included in the multi-scale regression network to generate a seventh convolution result.

[0076] In the eighth step, the first convolution result, the fifth convolution result, the sixth convolution result and the seventh convolution result are combined to generate combined feature information.

[0077] In the ninth step, the combined feature information is input into the batch normalization layer and the activation function layer to output the atmospheric delay correction image.

[0078] In some optional implementations of some embodiments, the atmospheric delay correction neural network model is trained by the following steps:

[0079] The first step is to obtain original synthetic aperture radar image sets, meteorological data datasets, and elevation datasets under different atmospheric imaging environments. The original synthetic aperture radar image sets can be image sets under a variety of imaging meteorological conditions to ensure that they contain a rich variety of atmospheric delay signal samples. Furthermore, the original synthetic aperture radar image sets can be image sets under different imaging area type factors. For example, different imaging area type factors can be mountainous, plain, etc. The patterns of atmospheric delay signals in these different regions are significantly different. Sample sets produced in relevant areas help enhance the model's atmospheric delay correction results under various terrain conditions. The meteorological data dataset here can use GACOS meteorological data.

[0080] The second step is to perform interferometric processing on the original synthetic aperture radar images in the original synthetic aperture radar image set to generate radar interferograms, thereby obtaining a radar interferogram set. The specific implementation method will not be described in detail.

[0081] In the third step, for each radar interferogram in the above radar interferogram set, the following generation steps are performed:

[0082] Sub-step 1: For each of at least one wavelet transform method, perform high-frequency component feature extraction and low-frequency component feature extraction on the radar interferogram using the wavelet transform method to generate radar high-frequency component feature information and radar low-frequency component feature extraction information. The specific implementation method is not further described.

[0083] Sub-step 2: performing feature information fusion on the obtained at least one radar high-frequency component feature information to generate third fused feature information, and performing feature information fusion on the obtained at least one radar low-frequency component feature information to generate fourth fused feature information.

[0084] Sub-step 3: pre-processing the corresponding meteorological data and the corresponding elevation data to generate auxiliary matrix data. The data pre-processing may include extracting atmospheric delay parameter data from the meteorological data and performing data dimension processing on the target elevation data.

[0085] Sub-step 4: combining the third fused feature information, the fourth fused feature information and the auxiliary matrix data to generate combined data.

[0086] Sub-step 5, utilizes at least one atmospheric correction method, carries out atmospheric correction to the above-mentioned radar interferogram, to generate at least one atmospheric correction interferogram. Wherein, the atmospheric correction method can be a method for performing atmospheric delay correction. In practice, at least one atmospheric correction method can include: based on external meteorological data method, WRF (Weather Research and Forecasting Model, Weather Research and Forecasting Model) atmospheric model, based on phase-elevation empirical relationship, based on MERIS spectrometer (Medium Resolution Imaging Spectrometer), based on spatial-temporal filtering method atmospheric delay correction method. Wherein, based on external meteorological data method and WRF atmospheric model, when meteorological changes are strong, there is better correction effect. Based on phase-elevation empirical relationship, based on MERIS spectrometer and based on spatial-temporal filtering method atmospheric delay correction method has better correction result for atmospheric delay caused by altitude.

[0087] Sub-step 6: Determine terrain roughness (TR) and atmospheric variability (AV) information corresponding to the radar interferogram. The terrain roughness information may represent the average terrain roughness within the area corresponding to the radar interferogram. The atmospheric variability information may represent the severity of atmospheric variability within the area.

[0088] As an example, the execution entity may determine the terrain undulation information and the atmospheric change severity information corresponding to the radar interference pattern through a standard deviation calculation formula.

[0089] Sub-step 7: Determine at least one correction weight corresponding to the at least one atmospheric correction method based on the terrain relief information and the atmospheric change severity information. The correction weight can represent the effectiveness of the atmospheric correction method, i.e., the degree to which the atmospheric correction method will be subsequently used. The correction weight can be a value between 0 and 1. A higher value indicates a higher effectiveness of the atmospheric correction method. Each atmospheric correction method has a corresponding correction weight.

[0090] As an example, the execution entity may match at least one correction weight corresponding to at least one atmospheric correction method according to the interval corresponding to the terrain undulation information and the atmospheric change severity information.

[0091] Sub-step 8: performing interferogram fusion on the at least one atmospheric correction interferogram according to the at least one correction weight to generate a correction interferogram.

[0092] As an example, the execution entity may perform weighted fusion on at least one atmospheric correction interferogram according to at least one correction weight to generate a correction interferogram.

[0093] Sub-step 9: using the combined data as training data, using the corrected interferogram as an interferogram label, and combining the combined data and the corrected interferogram to generate training data.

[0094] The fourth step is to train the initial atmospheric delay correction neural network model based on the obtained training data set to generate an atmospheric delay correction neural network model. The initial atmospheric delay correction neural network model can be an atmospheric delay correction neural network model that has not yet been trained.

[0095] As an example, the above-mentioned execution entity can use a conventional model training method (such as a model parameter update method based on back propagation) to train the initial atmospheric delay correction neural network model according to the obtained training data set to generate an atmospheric delay correction neural network model.

[0096] The above-mentioned "first step to fourth step" as one of the inventive points of the present disclosure solves another technical problem "how to accurately train the atmospheric delay correction neural network to obtain a more accurate model for atmospheric delay correction". Based on this, the present disclosure implements preprocessing and feature extraction of basic data by performing interference processing on the original synthetic aperture radar image, performing various transformations on the radar interference map, and extracting feature information and elevation data at each feature level for data preprocessing. At least one atmospheric correction method is used to generate atmospheric correction maps under each method, and the correction weights corresponding to each atmospheric correction method are determined according to the degree of terrain undulation and the severity of atmospheric changes, so as to accurately correct the interference map. Based on the above processing method, training data can be accurately generated to accurately train the initial large atmospheric delay correction neural network model to obtain a more accurate model for atmospheric delay correction.

[0097] Optionally, the execution entity may determine at least one correction weight corresponding to the at least one atmospheric correction method based on the terrain relief information and the atmospheric change severity information, which may include the following steps:

[0098] In the first step, the atmospheric change severity information is multiplied by a target value to obtain a first multiplication result. The target value may be a preset value, for example, 4.

[0099] In the second step, the terrain relief information is multiplied by the target value to obtain a second multiplication result.

[0100] In a third step, in response to determining that the atmospheric change severity information is greater than or equal to the second multiplication result, a correction weight ratio corresponding to the at least one atmospheric correction method is determined as a first correction weight ratio. The first correction weight ratio may be a preset ratio value. For example, the first correction weight ratio may be "4:2:2:1:1."

[0101] In the fourth step, the first correction weight ratio is normalized to generate at least one correction weight corresponding to the at least one atmospheric correction method.

[0102] In a fifth step, in response to determining that the second multiplication result is greater than the atmospheric change severity information and the atmospheric change severity information is greater than the terrain relief information, a correction weight ratio corresponding to the at least one atmospheric correction method is determined to be a second correction weight ratio. The second correction weight ratio may be a preset ratio. For example, the second correction weight ratio may be "3:2:3:1:1."

[0103] In a sixth step, the second correction weight ratio is normalized to generate at least two correction weights corresponding to the at least one atmospheric correction method.

[0104] In step 7, in response to determining that the first multiplication result is greater than or equal to the terrain relief information, a correction weight ratio corresponding to the at least one atmospheric correction method is determined to be a third correction weight ratio. The third correction weight ratio may be a preset ratio. For example, the third correction weight ratio may be "5:2:1:1:1."

[0105] In the eighth step, the third correction weight ratio is normalized to generate at least one correction weight corresponding to the at least one atmospheric correction method.

[0106] In step 9, in response to determining that the first multiplication result is greater than the terrain relief information and the terrain relief information is greater than the atmospheric change severity information, a correction weight ratio corresponding to the at least one atmospheric correction method is determined to be a fourth correction weight ratio. The fourth correction weight ratio may be a preset ratio value. For example, the fourth correction weight ratio may be "4:2:2:1:1."

[0107] The above-mentioned "first step to ninth step" as one of the inventive points of the present disclosure solves another technical problem "how to accurately fuse the atmospheric correction results corresponding to at least one atmospheric correction method to obtain an accurate atmospheric correction label". Based on this, the present disclosure determines the main source of atmospheric delay of samples in the sample library by analyzing two indicators: information on terrain undulation and information on the severity of atmospheric changes, and allocates the weight ratios of different methods. The principle of weight distribution is that when the atmospheric delay caused by drastic meteorological changes is dominant, a larger integration weight is set for the improved method based on external meteorological data and meteorological models. When in an area where the atmospheric delay is dominated by terrain undulation, a larger weight is designed for the method based on the phase-elevation relationship. In this way, at least one correction weight can be accurately generated in various situations, and the fusion of the corresponding output results of at least one atmospheric correction method can be achieved to obtain an accurate atmospheric correction label.

[0108] like Figure 4 As shown, a schematic diagram of model training and model application of the atmospheric delay correction neural network model is shown.

[0109] like Figure 4 As shown, after data preprocessing, the SAR interferogram, meteorological data (corresponding to meteorological data data) and DEM data (corresponding to elevation data) corresponding to the original synthetic aperture radar image are sequentially input into the multi-branch feature extraction module (corresponding to the multi-branch feature code extractor), the channel attention mechanism fusion module (feature fusion network) and the multi-scale regression head (multi-scale regression network) to generate the predicted image after atmospheric correction (i.e., Data_pred). In the application stage, the atmospheric correction image (i.e., Data) is directly generated. In the process of generating the interferogram label, first, for the SAR interferogram, filtering methods, phase-elevation relationship methods, meteorological models, etc. are used to generate corrected data1, corrected data2, corrected data3, etc. respectively. Then, according to the weights corresponding to at least one correction method, corrected data1, corrected data2, corrected data3, etc., an interferogram label is generated. Based on the interferogram label and the predicted image after atmospheric correction, model training for the atmospheric delay correction neural network model can be achieved.

[0110] like Figure 5 As shown, a schematic diagram of generating the second fusion feature information is shown.

[0111] like Figure 6 , which shows a schematic diagram of generating the corrected interference pattern.

[0112] The above-described embodiments of the present disclosure have the following beneficial effects: The atmospheric correction methods of some embodiments of the present disclosure can accurately perform atmospheric correction on target images to generate high-quality synthetic aperture radar images. Specifically, the reason for the inaccurate atmospheric correction is that the use of functional relationships to address atmospheric delay has certain limitations. This often only corrects delay phases that are highly correlated with model assumptions. The correction effect on the turbulent portion of the atmospheric delay, which exhibits strong temporal and spatial variability, is limited, resulting in an inability to accurately process atmospheric delay. Based on this, the large InSAR atmospheric delay correction methods of some embodiments of the present disclosure first obtain a target image, target meteorological data, and target elevation data for the target area, captured by a target synthetic aperture radar. Acquiring a synthetic aperture radar image of the target area facilitates the subsequent generation of an interferogram within the target area. Acquiring the target meteorological data can provide meteorological information within the target area, which can be used to provide meteorological auxiliary data during the subsequent atmospheric delay correction process. Acquiring the target elevation data can provide terrain information within the target area, which can be used to provide meteorological auxiliary data during the subsequent atmospheric delay correction process. Then, atmospheric delay parameter data is extracted from the target meteorological data, and the target elevation data is subjected to data dimension processing to generate elevation processing data. Here, by extracting atmospheric delay parameter data from the target meteorological data, key meteorological parameter content can be extracted. By performing data dimension processing on the target elevation data, the phase information corresponding to the subsequent matching interference pattern can be matched. Next, interference processing is performed on the target image to generate an interference pattern for subsequent wavelet transform processing. For each of at least one wavelet transform method, the wavelet transform method is used to extract high-frequency component features and low-frequency component features from the interference pattern to generate high-frequency component feature information and low-frequency component feature information. Here, based on the large frequency differences in the interference pattern, by extracting high-frequency component feature information and low-frequency component feature information, the subsequent atmospheric delay correction neural network model can accurately generate an atmospheric delay correction image based on feature information of different frequencies. Data interpolation and fusion are performed on the processed elevation data and the atmospheric delay parameter data to generate auxiliary matrix data of the same size as the matrix corresponding to the interference pattern, which serves as auxiliary feature information to subsequently accurately generate an atmospheric delay correction image. Furthermore, feature information fusion is performed on the obtained at least one high-frequency component feature information to generate first fused feature information, and feature information fusion is performed on the obtained at least one low-frequency component feature information to generate second fused feature information, thereby obtaining more accurately composite feature content of the high-frequency features and feature content of the low-frequency features.Finally, the first fused feature information, the second fused feature information, and the auxiliary matrix data are input into a pre-trained atmospheric delay correction neural network model to accurately generate an atmospheric delay correction image for the target image. In summary, by pre-processing the target image, target meteorological data, and target elevation data, and extracting composite high-frequency and low-frequency feature information, and based on the atmospheric delay correction neural network model, an atmospheric delay correction image can be accurately generated.

[0113] Further references Figure 2 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of an InSAR atmospheric delay correction device. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the InSAR atmospheric delay correction device can be specifically applied to various electronic devices.

[0114] like Figure 2 As shown, an InSAR atmospheric delay correction device 200 includes: an acquisition unit 201, a processing unit 202, an interference processing unit 203, an extraction unit 204, a first fusion unit 205, a second fusion unit 206 and a generation unit 207. The acquisition unit 201 is configured to acquire a target image, target meteorological data and target elevation data taken by a target synthetic aperture radar for a target area; the processing unit 202 is configured to extract atmospheric delay parameter data from the target meteorological data and perform data dimension processing on the target elevation data to generate elevation processing data; the interference processing unit 203 is configured to perform interference processing on the target image to generate an interference pattern; the extraction unit 204 is configured to perform high-frequency component feature extraction and low-frequency component feature extraction on the interference pattern using the wavelet transform method for each wavelet transform method in at least one wavelet transform method to generate high-frequency component feature information and low-frequency component feature information. information; a first fusion unit 205 is configured to perform data interpolation fusion on the elevation processing data and the atmospheric delay parameter data to generate auxiliary matrix data of the same size as the matrix corresponding to the interference pattern; a second fusion unit 206 is configured to perform feature information fusion on the at least one high-frequency component feature information obtained to generate first fused feature information, and to perform feature information fusion on the at least one low-frequency component feature information obtained to generate second fused feature information; a generation unit 207 is configured to input the first fused feature information, the second fused feature information and the auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay corrected image for the target image.

[0115] It is understood that the various units described in the InSAR atmospheric delay correction device 200 are similar to those described in the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the InSAR atmospheric delay correction device 200 and the units included therein, and will not be described in detail here.

[0116] Reference below Figure 3 , which shows a structural schematic diagram of an electronic device (eg, an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0117] like Figure 3 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. Various programs and data required for the operation of the electronic device 300 are also stored in the RAM 303. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0118] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as needed.

[0119] In particular, according to some embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from a network via the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are performed.

[0120] It should be noted that in some embodiments of the present disclosure, the computer-readable medium mentioned above may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, 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, device, or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable 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.

[0121] In some embodiments, the client and server can communicate using any currently known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0122] The computer-readable medium may be included in the electronic device, or may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains a target image, target meteorological data, and target elevation data taken by a target synthetic aperture radar for the target area; extracts atmospheric delay parameter data from the target meteorological data, and performs data dimension processing on the target elevation data to generate elevation processing data; performs interference processing on the target image to generate an interference pattern; and for each wavelet transform method in at least one wavelet transform method, uses the wavelet transform method to extract high-frequency component features and low-frequency component features from the interference pattern. To generate high-frequency component feature information and low-frequency component feature information; perform data interpolation and fusion on the above-mentioned elevation processing data and the above-mentioned atmospheric delay parameter data to generate auxiliary matrix data of the same size as the matrix corresponding to the above-mentioned interference pattern; perform feature information fusion on the obtained at least one high-frequency component feature information to generate first fused feature information, and perform feature information fusion on the obtained at least one low-frequency component feature information to generate second fused feature information; input the above-mentioned first fused feature information, the above-mentioned second fused feature information and the above-mentioned auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay corrected image for the above-mentioned target image.

[0123] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone 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 a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0124] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0125] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The units described may also be provided in a processor. For example, they may be described as follows: a processor includes an acquisition unit, a processing unit, an interference processing unit, an extraction unit, a first fusion unit, a second fusion unit, and a generation unit. The names of these units do not, in some cases, constitute limitations on the units themselves. For example, the acquisition unit may also be described as a "unit for acquiring target images, target meteorological data, and target elevation data captured by a target synthetic aperture radar for a target area."

[0126] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0127] The above description is only an illustration of some preferred embodiments of the present disclosure and the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. An InSAR atmospheric delay correction method, comprising: Acquire target images, target meteorological data and target elevation data taken by target synthetic aperture radar for the target area; Extracting atmospheric delay parameter data from the target meteorological data, and performing data dimension processing on the target elevation data to generate elevation processing data; performing interference processing on the target image to generate an interference pattern; For each of at least one wavelet transform mode, perform high-frequency component feature extraction and low-frequency component feature extraction on the interference pattern using the wavelet transform mode to generate high-frequency component feature information and low-frequency component feature information; Performing data interpolation and fusion on the elevation processing data and the atmospheric delay parameter data to generate auxiliary matrix data having the same size as the matrix corresponding to the interference pattern; Performing feature information fusion on the obtained at least one high-frequency component feature information to generate first fused feature information, and performing feature information fusion on the obtained at least one low-frequency component feature information to generate second fused feature information; The first fused feature information, the second fused feature information and the auxiliary matrix data are input into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay correction image for the target image.

2. The method according to claim 1, wherein The atmospheric delay correction neural network model includes: a multi-branch feature code extractor, a feature fusion network and a multi-scale regression network; and The step of inputting the first fused feature information, the second fused feature information, and the auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay correction image for the target image includes: Inputting the first fused feature information, the second fused feature information and the auxiliary matrix data into the multi-branch feature code extractor to generate a multi-channel feature information set; Inputting the multi-channel feature information set into the feature fusion network to generate fused feature information; The fused feature information is input into the multi-scale regression network to generate the interference image after the atmospheric delay correction.

3. The method according to claim 2, wherein: The multi-branch feature code extractor includes: a low-frequency feature extraction module, a high-frequency feature extraction module and an auxiliary data feature extraction module; and The step of inputting the first fused feature information, the second fused feature information, and the auxiliary matrix data into the multi-branch feature code extractor to generate a multi-channel feature information set includes: Inputting the first fusion feature information into the low-frequency feature extraction module to generate low-frequency channel feature information; Inputting the second fusion feature information into the high-frequency feature extraction module to generate high-frequency channel feature information; Inputting the auxiliary matrix data into the auxiliary data feature extraction module to generate auxiliary channel feature information; The multi-channel feature information set is generated according to the low-frequency channel feature information, the high-frequency channel feature information, and the auxiliary channel feature information.

4. The method according to claim 2, wherein: Inputting the multi-channel feature information set into the feature fusion network to generate fused feature information includes: Inputting the multi-channel feature information set into the average pooling layer included in the feature fusion network to generate an average pooling result; Inputting the average pooling result into the attention weight calculation layer included in the feature fusion network to generate attention feature information; Inputting the attention feature information into a multilayer perceptron included in the feature fusion network to generate a feedforward output result; Inputting the feedforward output result into a batch normalization layer included in the feature fusion network to generate a normalized result; Inputting the normalized result into the activation function layer included in the feature fusion network to generate the fused feature information; and Inputting the fused feature information into the multi-scale regression network to generate the atmospheric delay corrected image includes: Inputting the fused feature information into a convolution layer at a first convolution kernel scale included in the multi-scale regression network to generate a first convolution result; Inputting the fused feature information into a convolution layer at a second convolution kernel scale included in the multi-scale regression network to generate a second convolution result, wherein the second convolution kernel scale is higher than the first convolution kernel scale; Inputting the fused feature information into a convolution layer at a third convolution kernel scale included in the multi-scale regression network to generate a third convolution result, wherein the third convolution kernel scale is higher than the second convolution kernel scale; Inputting the fused feature information into a convolution layer at a fourth convolution kernel scale included in the multi-scale regression network to generate a fourth convolution result, wherein the fourth convolution kernel scale is higher than the third convolution kernel scale; Inputting the second convolution result into the convolution layer at the first convolution kernel scale included in the multi-scale regression network to generate a fifth convolution result; Inputting the third convolution result into the convolution layer at the first convolution kernel scale included in the multi-scale regression network to generate a sixth convolution result; Inputting the fourth convolution result into the convolution layer at the first convolution kernel scale included in the multi-scale regression network to generate a seventh convolution result; combining the first convolution result, the fifth convolution result, the sixth convolution result, and the seventh convolution result to generate combined feature information; The combined feature information is input into a batch normalization layer and an activation function layer to output the atmospheric delay correction image.

5. The method according to claim 1, wherein The step of extracting atmospheric delay parameter data from the target meteorological data and performing data dimension processing on the target elevation data to generate elevation processing data includes: For each grid data in the target elevation data, the following generation steps are performed: Dividing the target value by the radar wavelength to generate a first division result; Multiplying the satellite distance corresponding to the raster data by the sine value corresponding to the radar wave incident angle to obtain a multiplication result; Dividing the vertical baseline length corresponding to the grid data by the multiplication result to obtain a second division result; multiplying the first division result, the second division result, and the ground height corresponding to the raster data to obtain processed raster data; The obtained processed raster dataset is determined as the elevation processed data.

6. The method according to claim 3, wherein: The multi-branch feature code extractor further includes: a first multi-head attention mechanism model, a second multi-head attention mechanism model and a third multi-head attention mechanism model; and Generating the multi-channel feature information set according to the low-frequency channel feature information, the high-frequency channel feature information, and the auxiliary channel feature information includes: Inputting the low-frequency channel feature information and the high-frequency channel feature information into the first multi-head attention mechanism model to generate first low-frequency channel feature information and first high-frequency channel feature information; Inputting the high-frequency channel feature information and the auxiliary channel feature information into the second multi-head attention mechanism model to generate first auxiliary channel feature information and second high-frequency channel feature information; Inputting the low-frequency channel feature information, the high-frequency channel feature information, and the auxiliary channel feature information into the third multi-head attention mechanism model to generate second auxiliary channel feature information, second low-frequency channel feature information, and third high-frequency channel feature information; The low-frequency channel feature information, the high-frequency channel feature information, the auxiliary channel feature information, the first low-frequency channel feature information, the first high-frequency channel feature information, the first auxiliary channel feature information, the second high-frequency channel feature information, the second auxiliary channel feature information, the second low-frequency channel feature information and the third high-frequency channel feature information are determined as the multi-channel feature information set.

7. An InSAR atmospheric delay correction device, comprising: an acquisition unit configured to acquire a target image, target meteorological data, and target elevation data taken by a target synthetic aperture radar for a target area; a processing unit configured to extract atmospheric delay parameter data from the target meteorological data and perform data dimension processing on the target elevation data to generate elevation processed data; an interference processing unit configured to perform interference processing on the target image to generate an interference pattern; an extraction unit configured to perform high-frequency component feature extraction and low-frequency component feature extraction on the interference pattern using each of at least one wavelet transform mode to generate high-frequency component feature information and low-frequency component feature information; a first fusion unit configured to perform data interpolation fusion on the elevation processing data and the atmospheric delay parameter data to generate auxiliary matrix data of the same size as the matrix corresponding to the interference pattern; a second fusion unit configured to perform feature information fusion on the obtained at least one high-frequency component feature information to generate first fused feature information, and to perform feature information fusion on the obtained at least one low-frequency component feature information to generate second fused feature information; A generating unit is configured to input the first fused feature information, the second fused feature information and the auxiliary matrix data into a pre-trained atmospheric delay correction neural network model to generate an atmospheric delay correction image for the target image.

8. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer-readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images