A method and device for detecting shallow sea underwater terrain based on SAR image

Abstract

The present application provides a kind of shallow sea underwater terrain detection method and device based on SAR image, belong to marine remote sensing application field, including: data preprocessing;Calculate water depth background field gradient;Calculate the position code of each sub-image;SAR underwater terrain inversion training data set is constructed;Build shallow sea underwater terrain inversion model based on multi-source feature input and residual structure;The precision of the shallow sea underwater terrain inversion model based on multi-source feature input and residual structure obtained by training is verified using data not participating in training, the migration of the shallow sea underwater terrain inversion model based on multi-source feature input and residual structure is evaluated by cross-region verification, and the shallow sea underwater terrain inversion model based on multi-source feature input and residual structure is applied to the water depth inversion of target area, and the water depth distribution result is output.The present application can improve the inversion precision and migration of model.

Description

Technical Field

[0001] This invention belongs to the field of marine remote sensing applications, specifically relating to a method and apparatus for shallow sea underwater topography detection based on SAR imagery. Background Technology

[0002] High-precision shallow water underwater topographic data is crucial for the utilization of marine resources and the development of the marine economy in coastal areas. Traditional shipborne or airborne depth measurement methods can only acquire observational data on the course of a vessel, making it difficult to conduct large-scale underwater topographic surveys. With the development of satellite remote sensing technology, satellite optical remote sensing and synthetic aperture radar (SAR) remote sensing have become the main means of large-scale shallow water underwater topographic observation.

[0003] Optical remote sensing sensors mainly include active lidar and passive hyperspectral imagers. The former offers high measurement accuracy but low spatial coverage, while the latter provides higher spatial coverage but lower accuracy. Furthermore, optical remote sensing is highly sensitive to water quality, limiting the range of water depth measurements. In contrast, SAR imagery can capture changes in sea surface features caused by underwater topographic modulation and can perform underwater topographic inversion based on the bright and dark stripes or regular stripe features on the image.

[0004] Current underwater topography inversion methods based on SAR imagery include simulation inversion models, theoretical inversion models, and surge inversion models. These methods rely on SAR imaging simulation theory, underwater topography modulation mechanisms, and prior wave parameters, resulting in computational complexity and limited effective inversion depths. In contrast, deep learning's unique advantages in image feature extraction make it well-suited for processing high spatial resolution SAR imagery. By constructing deep learning models to describe the nonlinear relationship between image features and water depth, shallow-water underwater topography inversion can be achieved.

[0005] On the one hand, existing optical remote sensing underwater topographic detection technologies are easily affected by factors such as water transparency, atmospheric conditions, and different seabed and biological optical characteristics, resulting in a relatively shallow detection range. On the other hand, SAR-based theoretical inversion models often simplify complex ocean processes, while wave inversion methods rely on prior parameters such as wave period, leading to insufficient detection depth and inversion accuracy.

[0006] However, existing deep learning inversion algorithms based on SAR image stripes lack in-depth exploration of common physical mechanisms across different regions. The features extracted by the model are strongly coupled with the feature parameters of local sea areas, resulting in poor model transferability and difficulty in applying them to the inversion of underwater topography in different sea areas. Existing inversion methods focus on constructing inversion models through physical mechanisms (such as geometric modeling, electromagnetic scattering modeling, and SAR imaging simulation). That is, they theoretically explain physical processes by revealing physical mechanisms (how water depth modulates wave spectra, how the interaction between flow fields and topography affects the sea surface, and how wind fields affect wave spectra and scattering characteristics), and then construct inversion models to detect underwater topography. The shortcomings of these methods are that they rely on a series of system assumptions and prior knowledge, are computationally complex, have a limited applicable water depth range, and generally have poor model transferability. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a shallow water underwater topography detection method and apparatus based on SAR imagery. It utilizes the feature extraction capabilities of deep learning to process the bright and dark stripe features on SAR images that are closely related to underwater topography, constructing a shallow water depth inversion model (MFIR-WDI) based on multi-source feature input and residual structure. In addition to the SAR image backscattering coefficient, incident angle, and geographic latitude and longitude, the model input includes the background depth gradient (BDG), sea surface wind field, and current field, which describe environmental and sea state conditions. The model's inversion accuracy and transferability are significantly improved compared to existing models, and the inversion depth can reach over 50 meters.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A shallow water underwater topography detection method based on SAR imagery includes the following steps:

[0010] Step 1: Collect polarimetric SAR images, sea surface wind and current field data, water depth background field data, and reference water depth data, and preprocess the data;

[0011] Step 2: Based on GEBCO global DEM data, calculate the east-west and north-south directional gradients. Then, calculate the initial gradient values ​​based on the two directional gradients, and process them using Gaussian filtering and the Sigmoid function to obtain the water depth gradient. Multiply the gradient by the water depth normalization coefficient to obtain the water depth background field gradient, which serves as prior knowledge of the water depth distribution changes in different regions. GEBCO represents the ocean depth map, and DEM represents the digital elevation model.

[0012] Step 3: Divide the polarimetric SAR image into several sub-images to form a sample sequence, calculate the location code of each sub-image, improve the strong regional coupling characteristics brought by geographic latitude and longitude information, and preprocess the SAR sub-images.

[0013] Step 4: Using sub-images, water depth background field gradient, radar incident angle, geographic latitude and longitude, sea surface wind field and current field data as input, construct a SAR underwater terrain inversion training dataset;

[0014] Step 5: Build a shallow water underwater topography inversion model based on multi-source feature input and residual structure, and train the shallow water underwater topography inversion model using the mean square error loss function;

[0015] Step 6: Validate the accuracy of the shallow water underwater topography inversion model obtained from the training using data not used in the training. Evaluate the transferability of the shallow water underwater topography inversion model through cross-regional validation. Apply the shallow water underwater topography inversion model to the water depth inversion of the target area and output the water depth distribution results.

[0016] The present invention also provides a shallow sea underwater topography detection device based on SAR imagery, comprising the following modules:

[0017] The preprocessing module collects polarimetric SAR images, sea surface wind and current field data, water depth background field data, and reference water depth data, and preprocesses the data.

[0018] The gradient calculation module, based on GEBCO global DEM data, calculates the east-west and north-south directional gradients. Then, it calculates the initial gradient values ​​based on the two directional gradients, and obtains the water depth gradient through Gaussian filtering and the Sigmoid function. Multiplying it by the water depth normalization coefficient yields the water depth background field gradient, which serves as prior knowledge of the water depth distribution changes in different regions. GEBCO represents the ocean depth map, and DEM represents the digital elevation model.

[0019] The coding and calculation module divides the polarimetric SAR image into several sub-images to form a sample sequence, calculates the location code of each sub-image, and improves the strong regional coupling characteristics brought about by geographic latitude and longitude information.

[0020] The dataset construction module takes sub-images, water depth background field gradients, radar incident angles, geographical latitude and longitude, sea surface wind field and current field data as inputs to construct a SAR underwater terrain inversion training dataset.

[0021] The model building module constructs a shallow sea underwater topography inversion model based on multi-source feature input and residual structure, and trains the shallow sea underwater topography inversion model using the mean square error loss function.

[0022] The inversion module uses data not used in training to verify the accuracy of the trained shallow water underwater topography inversion model, evaluates the transferability of the shallow water underwater topography inversion model through cross-regional verification, and applies the shallow water underwater topography inversion model to the water depth inversion of the target area, outputting water depth distribution results.

[0023] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method for shallow sea underwater topography detection based on SAR imagery.

[0024] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described shallow seawater topographic exploration method based on SAR imagery.

[0025] Beneficial effects:

[0026] 1. This invention addresses the issues of insufficient model accuracy and inversion depth by employing multi-source feature input and residual structures. It integrates data features (backscattering coefficient, local incident angle, latitude and longitude, wind field, current field, and water depth background field gradient) and image features (SAR backscattering coefficient image and water depth background field gradient image). By introducing sea surface wind and current field parameters, the model's applicability to different sea conditions is improved. A residual module is used to extract high-dimensional features from the image, and skip connections alleviate gradient vanishing, reducing the risk of overfitting the model to local training data. Compared with existing wave models and existing models in this region, the underwater topography inversion model based on multi-source feature input and residual structures can more accurately describe the nonlinear relationship between water depth and SAR signals, significantly improving inversion accuracy and extending the effective detection range to 70 meters.

[0027] 2. This invention addresses the poor model transferability problem by introducing prior knowledge of the water depth background field gradient and performing positional encoding within SAR images. The water depth background field gradient calculated based on GEBCO (Gross Ocean Depth Map) global DEM (Digital Elevation Model) data serves as the image feature input, characterizing the water depth distribution trends in different regions. This facilitates the model's learning of the spatial correlation between water depth and SAR signals, reduces feature shifts caused by regional differences, and significantly improves the model's transferability. Positional encoding within the SAR image suppresses the strong regional coupling caused by geographic latitude and longitude information, enhancing the model's generalization ability. Even with limited SAR data available for training in the target area, features can still be extracted from a small amount of SAR data using the first two modules of the MFIR-WDI model, and the output module can be retrained to achieve high-precision water depth inversion results, greatly reducing model training parameters and improving computational efficiency.

[0028] 3. This invention addresses the robustness issue of spatiotemporally dynamic SAR image feature interpretation by combining sea surface wind field, current field, and water depth background field gradient. Sea surface wind field and current field represent different sea state conditions, affecting the salience of underwater topographic stripes on SAR images and providing a physical basis for SAR feature interpretation. Background field gradient serves as prior knowledge of underwater topography, representing water depth changes in different regions and providing geographical constraints for SAR feature interpretation. The combination of these three effectively helps the model distinguish and fuse spatiotemporally dynamic information (sea surface wind field and current field) and spatial background information (BDG), better describing the relationship between SAR features (especially bright and dark stripes) and underwater topography in different time phases and regions. Attached Figure Description

[0029] Figure 1 This is a framework diagram of a shallow water depth inversion model based on multi-source feature input and residual structure.

[0030] Figure 2 This is a map showing the water depth inversion results for the northeastern waters of a certain island; where A is the reference water depth, B is the mean water depth of the four predicted results, and C is the error statistics between the predicted water depth and the reference water depth.

[0031] Figure 3 This is a statistical chart of water depth inversion errors in a certain island's sea area; where A is the comparison result of the MFIR-WDI model, and B is the comparison result of the optimized MFIR-WDI model.

[0032] Figure 4 The image shows the water depth inversion results for a certain island's sea area; where A is the reference water depth, B is the water depth inversion result of the MFIR-WDI model, and C is the optimized water depth inversion result of the MFIR-WDI model.

[0033] Figure 5 This is a flowchart of a shallow water underwater topography detection method based on SAR imagery according to the present invention;

[0034] Figure 6 This is a schematic diagram of a shallow seawater topographic detection device based on SAR imagery according to the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0036] like Figure 5As shown, this invention provides a shallow water underwater topography detection method based on SAR imagery. The underwater topography inversion model it provides, based on multi-source input and residual structure, significantly improves the inversion accuracy, with an effective inversion depth of approximately 70 meters. The model input includes Sentinel-1 VV polarimetric SAR imagery, water depth background field gradient, radar incidence angle, sea surface wind field, and sea surface current field. High-dimensional image features of the SAR image and background field gradient are extracted using a residual module; data features of radar backscattering coefficient, radar incidence angle, sea surface wind field, sea surface current field, and water depth background field gradient are extracted using a multilayer perceptron. The two types of features are combined and calculated using a fully connected layer, ultimately outputting the water depth inversion result. The specific steps include:

[0037] Step 1: Data Collection and Preprocessing

[0038] (1) Data collection included acquiring Sentinel-1A (S1A satellite of Sentinel-1) VV co-polarized images at different time phases, and performing preprocessing such as radiometric calibration, thermal noise removal, topographic correction, image cropping, and resampling; acquiring sea surface wind field and sea surface current field data characterizing sea conditions in the study area; acquiring ETOPO (Global Topographic Data) 2022 global ocean bathymetry data provided by the National Centers for Environmental Information (NCEI) and water depth interpolation data for a certain sea area provided by the National Geophysical Data Center (NGDC), which were used as the true reference values ​​for water depth in the first and second sea areas, respectively. The spatial resolution of all data was resampled to 10 meters.

[0039] (2) Calculation of water depth background field gradient:

[0040] This invention calculates the BDG (depth background gradient) based on the General Bathymetric Chart of the Ocean (GEBCO) global DEM (Digital Elevation Model) data provided by the Global Earth System Project, using it as prior knowledge of water depth distribution in different regions. First, the directional gradients of the GEBCO depth background field in the east-west and north-south directions are calculated separately. and Calculate the initial gradient value based on the gradients in two directions. Then, based on the Gaussian filter function Smoothing is performed to suppress noise; finally, the Sigmoid function is used for nonlinear mapping to obtain the water depth gradient. The specific calculations are as follows:

[0041] ;

[0042] ;

[0043] ;

[0044] ;

[0045] ;

[0046] ;

[0047] Where i and j represent the row and column numbers of the image, respectively. Indicates the first Line number The water depth values ​​of the column, Let be the standard deviation of the Gaussian filter function, k be the distance from the center of the convolution kernel to the edge, the kernel size be (2k+1, 2k+1), m be the offset in the row direction, and n be the offset in the column direction. This represents the gradient value after processing with the Gaussian filter function. The parameter is indicated by the subscript "min", which means taking the minimum value.

[0048] In this invention, the water depth background field gradient (BDG) is defined as the water depth gradient. Normalized coefficient of water depth The product of these terms is used to characterize the changes in water depth characteristics in different regions. The calculation process is as follows:

[0049] ;

[0050] ;

[0051] In the formula, This represents the maximum water depth in the area. This represents the minimum water depth in the region.

[0052] (3) Perform position encoding:

[0053] The SAR image is divided into several sub-images of 8×8 size, and all sub-images constitute a sample sequence. Location coding is used to describe the relative positional relationships between the sub-image samples, which can improve the strong regional coupling characteristics caused by geographic latitude and longitude information. First, the sub-images are unfolded into one-dimensional vectors, and all sub-image vectors obtained based on the original SAR image form an image sequence; then, the location-coded vector PE of each sub-image vector in the sequence is calculated according to the following formula:

[0054] ;

[0055] ;

[0056] Finally, the location-encoded features and image features are added and fused together to obtain the composite image features containing location information, denoted as follows: .

[0057] ;

[0058] In the formula, pos represents the position of the sub-image in the sequence. and These represent the position encoding vectors of two adjacent dimensions, where the dimensions of the position vectors are... The value is 64, where p represents the position index of the sub-image. Represents image features.

[0059] (4) Data quality control:

[0060] Data quality control follows the following three principles: (1) Remove land areas with water depth values ​​greater than 0 in the reference true values; (2) Perform data augmentation on sub-image samples with depths exceeding 30 meters to solve the problem of uneven distribution of training data caused by the scarcity of deep-water samples; (3) Remove sub-image samples with missing environmental parameters such as wind field and flow field.

[0061] Step 2: Construct a SAR underwater terrain inversion training dataset:

[0062] The input of the SAR underwater terrain inversion training dataset constructed in this invention consists of three parts. First, VV polarimetric SAR sub-images of size 8×8 and the corresponding water depth background field gradients are acquired and used as the image input of the model. Then, radar incident angle, geographical latitude and longitude, sea surface wind field and current field information of size 8×8 are acquired and their average values ​​are used as the radar observation parameters and environmental parameters input of the model.

[0063] The multi-source feature input consists of three parts: first, it consists of radar backscattering coefficient and BDG feature sub-map, with a total of 2 feature channels, which are input into the convolutional layer; second, it consists of backscattering coefficient, radar incident angle, latitude and longitude, wind speed UV and current speed UV feature parameters, which are input into the linear layer; and third, the water depth background field gradient parameters are input into the linear layer, and the output is the corresponding reference water depth value.

[0064] Step 3: Construct a shallow water depth inversion model based on multi-source feature input and residual structure:

[0065] like Figure 1 As shown, the construction of the shallow water depth inversion model based on multi-source feature input and residual structure includes:

[0066] Step 3-1: Construct an image feature extraction module based on residual modules:

[0067] like Figure 1 As shown, firstly, a 3×3 convolution kernel is used to process the input backscattering coefficient image and the water depth background field gradient image. Figure 1 The input 1 is used for convolution calculation; then two residual modules are connected sequentially ( Figure 1 Residual modules 1 and 2 extract high-dimensional features from the image; each residual module consists of three convolutional layers, namely a 1×1 channel compression layer, a 3×3 feature extraction layer, and a 1×1 convolutional channel expansion layer (which are respectively processed by...). Figure 1 The algorithm consists of batch normalization layers 1, 2, and 3 and their corresponding activation functions. Skip connections are used to directly pass gradients to alleviate the gradient vanishing problem. Finally, an average pooling layer is used to obtain a 512×1 feature vector.

[0068] Step 3-2: Construct an auxiliary parameter extraction module based on a multilayer perceptron:

[0069] First, the input backscattering coefficient, incident angle, longitude, latitude, wind speed UV component and current velocity UV component, and backscattering coefficient ( Figure 1 Input 2) is normalized, and then a Linear layer is used for feature mapping to obtain a 64×1 feature vector; then, the gradient value of the water depth background field ( Figure 1 Input 3) is mapped to a 4×1 terrain prior feature vector through a Linear layer. This terrain prior feature vector contains the differences in underwater topography between different sea areas, effectively enhancing the model's transferability.

[0070] Step 3-3: Construct the output module:

[0071] The feature vectors extracted in steps 3-1 and 3-2 are combined to obtain a 580×1 feature vector, which is used as the input vector for the output module. First, a Batch Normalization (BN) layer is used to normalize this vector; then, two fully connected layers (…) are used… Figure 1 The fully connected layer 1 and fully connected layer 2 in the process are used to perform nonlinear mapping calculations, and finally the water depth inversion results are output.

[0072] Step 4: Train the underwater terrain inversion model:

[0073] The SAR underwater topography inversion training dataset obtained in step two and the shallow water underwater topography inversion model based on multi-source feature input built in step three are used to train the model. The training requires multiple iterations to adjust the model weights to the optimal level so that the model can converge.

[0074] The mean squared error (MSE) loss function is used during training. :

[0075] ;

[0076] In the formula, This indicates the number of samples in the training set. The model inversion results represent the water depth. represents the reference truth value, and n represents the index value.

[0077] Step 5: Verify model accuracy.

[0078] The shallow water underwater topography inversion model based on multi-source feature input, obtained from step four training, was tested using data from four scenes of the first sea area that were not included in the training. Accuracy evaluation metrics included the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) between the inverted water depth and the reference water depth. Figure 2 As shown ( Figure 2 A is the reference water depth. Figure 2 B represents the mean water depth of the four predicted scenarios. Figure 2 (where C represents the statistical error between the predicted and reference water depths). The results show that the model's MAE, RMSE, and MAPE are 0.82m, 1.02m, and 10.80%, respectively. Compared with existing underwater topography inversion models applicable to this region, the proposed shallow water depth inversion model (MFIR-WDI model) based on multi-source feature input and residual structure reduces MAE, RMSE, and MAPE by 36.92%, 47.69%, and 6.49%, respectively.

[0079] Step Six: Perform model transferability validation:

[0080] The MFIR-WDI model (a shallow water depth inversion model based on multi-source feature input and residual structure) obtained in step four was directly used for water depth inversion in the second sea area. The water depth prediction results for the four sets of data are as follows: Figure 3 As shown ( Figure 3 A represents the comparison results of the MFIR-WDI model. Figure 3(B represents the comparison results of the optimized MFIR-WDI model): MAE, MRE, and MAPE are 2.34m, 2.97m, and 15.61%, respectively. Two additional SAR data sets from this sea area were selected. Based on the two feature extraction modules of the MFIR-WDI model, features were extracted first, and then the output module was retrained. The MAE, MRE, and MAPE of the model were 1.47m, 1.89m, and 12.30%, respectively. Compared with directly applying the MFIR-WDI model to the second sea area, the optimized shallow water depth inversion model (MFIR-WDI model) based on multi-source feature input and residual structure, obtained by retraining the output module, achieves better water depth inversion results. Specifically, the optimized model reduces MAE, RMSE, and MAPE by 37.18%, 36.36%, and 21.20%, respectively, confirming that the proposed shallow water depth inversion model (MFIR-WDI model) based on multi-source feature input and residual structure possesses excellent transferability.

[0081] like Figure 4 The image shown is a map illustrating the water depth inversion results for a certain island's sea area. Figure 4 A is the reference water depth. Figure 4 B represents the water depth inversion result from the MFIR-WDI model. Figure 4 C represents the water depth inversion result of the optimized MFIR-WDI model.

[0082] This invention adds a priori water depth background field gradient describing the underwater topography distribution pattern to the input, solving the problem of poor mobility of underwater topography inversion models; it introduces sea surface wind field and sea surface current field parameters to suppress the negative impact of temporal and sea state differences on model stability; and the location encoding inside the SAR image suppresses the strong regional coupling caused by geographic latitude and longitude information, improving the model's generalization ability.

[0083] Preferably, Sentinel-1A imagery is replaced by other SAR satellite data, such as Gaofen-3 and RADARSAT-2 VV polarimetric imagery.

[0084] like Figure 6 As shown, the present invention also provides a shallow sea underwater topography detection device based on SAR imagery, comprising the following modules:

[0085] The preprocessing module collects polarimetric SAR images, sea surface wind and current field data, water depth background field data, and reference water depth data, and preprocesses the above data.

[0086] The gradient calculation module, based on GEBCO global DEM data, calculates the east-west and north-south directional gradients. Then, it calculates the initial gradient value based on these two directional gradients, and obtains the water depth gradient by processing it with Gaussian filtering and the Sigmoid function. Multiplying it by the water depth normalization coefficient, it obtains the water depth background field gradient, which serves as prior knowledge of the changes in water depth distribution in different regions.

[0087] The coding and calculation module divides the polarimetric SAR image into several sub-images to form a sample sequence, calculates the location code of each sub-image, and improves the strong regional coupling characteristics brought about by geographic latitude and longitude information.

[0088] The dataset construction module takes preprocessed SAR sub-images, water depth background field gradient, radar incident angle, geographic latitude and longitude, sea surface wind field and current field data as input to construct a SAR underwater topography inversion training dataset.

[0089] The model building module constructs a shallow sea underwater topography inversion model based on multi-source feature input and residual structure, and trains the model using the mean squared error loss function.

[0090] The inversion module uses untrained data to verify the accuracy of the shallow sea underwater topography inversion model based on multi-source feature input and residual structure. It evaluates the transferability of the shallow sea underwater topography inversion model based on multi-source feature input and residual structure through cross-regional verification, applies the shallow sea underwater topography inversion model based on multi-source feature input and residual structure to the target area for water depth inversion, outputs water depth distribution results, and stores the water depth distribution map and numerical results to a specified database or visualization terminal.

[0091] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method for shallow sea topographic exploration based on SAR imagery.

[0092] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described shallow seawater topographic exploration method based on SAR imagery.

[0093] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0094] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0095] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0096] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0097] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0098] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A shallow water underwater topography detection method based on SAR imagery, characterized in that, Includes the following steps: Step 1: Collect polarimetric SAR images, sea surface wind and current field data, water depth background field data, and reference water depth data, and preprocess the data; Step 2: Based on GEBCO global DEM data, calculate the east-west and north-south directional gradients. Then, calculate the initial gradient values ​​based on the two directional gradients, and process them using Gaussian filtering and the Sigmoid function to obtain the water depth gradient. Multiply the gradient by the water depth normalization coefficient to obtain the water depth background field gradient, which serves as prior knowledge of the water depth distribution changes in different regions. GEBCO represents the ocean depth map, and DEM represents the digital elevation model. Step 3: Divide the polarimetric SAR image into several sub-images to form a sample sequence, calculate the location code of each sub-image, and improve the strong regional coupling characteristics brought about by geographic latitude and longitude information; Step 4: Using sub-images, water depth background field gradient, radar incident angle, geographic latitude and longitude, sea surface wind field and current field data as input, construct a SAR underwater terrain inversion training dataset; Step 5: Build a shallow water underwater topography inversion model based on multi-source feature input and residual structure, and train the shallow water underwater topography inversion model using the mean square error loss function; Step 6: Validate the accuracy of the shallow water underwater topography inversion model obtained from the training using data not used in the training. Evaluate the transferability of the shallow water underwater topography inversion model through cross-regional validation. Apply the shallow water underwater topography inversion model to the water depth inversion of the target area and output the water depth distribution results.

2. The shallow sea underwater topography detection method based on SAR imagery according to claim 1, characterized in that, In step 1, the polarimetric SAR images include Gaofen-3 polarimetric images, RADARSAT-2 VV polarimetric images, or Sentinel-1 VV polarimetric SAR images; preprocessing includes radiometric calibration, thermal noise removal, terrain correction, image cropping, and resampling, removing land areas with water depth values ​​greater than 0 in the reference ground truth, performing data augmentation on sub-image samples with depths greater than 30 meters, and removing sub-image samples with missing wind and flow field parameters; the spatial resolution of all data is unified to 10 meters.

3. The shallow sea underwater topography detection method based on SAR imagery according to claim 1, characterized in that, In step 2, the formula for calculating the water depth background field gradient (BDG) is as follows: ； Among them, water depth gradient The water depth normalization coefficient is obtained by smoothing the surface using a Gaussian filter and then nonlinearly mapping it using a Sigmoid function. It is obtained by normalizing the maximum and minimum values.

4. The shallow sea underwater topography detection method based on SAR imagery according to claim 1, characterized in that, In step 3, the formula for calculating the location code is: ； ； The location-coded features are added to and fused with the image features, resulting in a composite image feature containing location information, denoted as follows: ; ； In the formula, pos represents the position of the sub-image in the sequence. and These represent the position encoding vectors of two adjacent dimensions, where the dimensions of the position encoding vectors are... The value is 64, where p represents the position index of the sub-image; Represents image features.

5. The shallow sea underwater topography detection method based on SAR imagery according to claim 1, characterized in that, In step 5, the shallow sea underwater topography inversion model based on multi-source feature input and residual structure includes an image feature extraction module, an auxiliary parameter extraction module, and an output module. The image feature extraction module uses the residual module to extract high-dimensional features of SAR images and water depth background field gradients.

6. The shallow sea underwater topography detection method based on SAR imagery according to claim 5, characterized in that, The image feature extraction module uses a 3×3 convolution kernel to perform convolution calculations on the input SAR image and the water depth background field gradient image; it employs two residual modules, each consisting of a 1×1 channel compression layer, a 3×3 feature extraction layer, and a 1×1 channel expansion layer, and uses skip connections to alleviate the gradient vanishing problem; The auxiliary parameter extraction module uses a linear layer to perform feature mapping on the backscattering coefficient, incident angle, sea surface wind field and flow field data to obtain a 64×1 feature vector; it also uses a linear layer to map the water depth background field gradient data into a 4×1 terrain prior feature vector. The output module uses a batch normalization layer to normalize the fused feature vectors; it uses two fully connected layers to perform nonlinear mapping calculations, and finally outputs the water depth inversion result.

7. The shallow sea underwater topography detection method based on SAR imagery according to claim 1, characterized in that, In step 6, the accuracy verification metrics include mean absolute error, root mean square error, and mean absolute percentage error.

8. A shallow sea underwater topographic detection device based on SAR imagery, characterized in that, Includes the following modules: The preprocessing module collects polarimetric SAR images, sea surface wind and current field data, water depth background field data, and reference water depth data, and preprocesses the data. The gradient calculation module, based on GEBCO global DEM data, calculates the east-west and north-south directional gradients. Then, it calculates the initial gradient values ​​based on the two directional gradients, and obtains the water depth gradient through Gaussian filtering and the Sigmoid function. Multiplying it by the water depth normalization coefficient yields the water depth background field gradient, which serves as prior knowledge of the water depth distribution changes in different regions. GEBCO represents the ocean depth map, and DEM represents the digital elevation model. The coding and calculation module divides the polarimetric SAR image into several sub-images to form a sample sequence, calculates the location code of each sub-image, and improves the strong regional coupling characteristics brought about by geographic latitude and longitude information. The dataset construction module takes sub-images, water depth background field gradients, radar incident angles, geographical latitude and longitude, sea surface wind field and current field data as inputs to construct a SAR underwater terrain inversion training dataset. The model building module constructs a shallow sea underwater topography inversion model based on multi-source feature input and residual structure, and trains the shallow sea underwater topography inversion model using the mean square error loss function. The inversion module uses data not used in training to verify the accuracy of the trained shallow water underwater topography inversion model, evaluates the transferability of the shallow water underwater topography inversion model through cross-regional verification, and applies the shallow water underwater topography inversion model to the water depth inversion of the target area, outputting water depth distribution results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of a shallow sea underwater topography detection method based on SAR imagery as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of a shallow sea underwater topography exploration method based on SAR imagery as described in any one of claims 1 to 7.

Images