SAR remote sensing data interpretation method based on environment cognition

Abstract

The embodiment of the invention relates to the technical field of SAR scene recognition and interpretation, in particular to an SAR remote sensing data interpretation method based on environmental cognition. The embodiment of the invention provides an SAR remote sensing data interpretation method based on environmental cognition, and the method comprises the steps: carrying out the region division of SAR scene data, and obtaining a plurality of region scenes; environment electromagnetic characteristic calculation is carried out on the obtained SAR data of each area scene, the characteristic quantity of the area scene is obtained, and the characteristic quantity comprises a backscattering coefficient and probability distribution; and inputting the feature quantity into a trained neural network recognition model, and outputting the specific type of the region scene corresponding to the feature quantity after the neural network recognition model performs feature extraction on the feature quantity. The embodiment of the invention provides an SAR (Synthetic Aperture Radar) remote sensing data interpretation method based on environmental cognition, which can realize real-time simulation of infrared characteristics of a scene with a participating medium.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of SAR scene recognition and interpretation, and in particular to a SAR remote sensing data interpretation method based on environment cognition. Background Art

[0002] Synthetic aperture radar (SAR) is widely used in the field of remote sensing reconnaissance and identification technology because it has all-day and all-weather imaging capabilities that optical remote sensing imaging systems do not have. The SAR imaging process mainly reflects the electromagnetic scattering characteristics of the target and the environment. The material and structure of the target, the composition and distribution of the environment show a high sensitivity to electromagnetic scattering in the SAR image. Therefore, SAR remote sensing images have extremely high application value in target and environment interpretation and reconnaissance.

[0003] In recent years, synthetic aperture radar (SAR) recognition and interpretation technology has made great progress. Most studies focus on the electromagnetic scattering characteristics of targets with metal materials such as aircraft and ships. However, the targets in the measured SAR data must be placed in the environment. The coupling between the target and the environment is very complex, and the environmental characteristics are numerous, rich in variety, and complex in distribution. Ultimately, the environmental cognition research and application of SAR data are in a dilemma, and the data are difficult to interpret effectively. Summary of the invention

[0004] The embodiment of the present invention provides a SAR remote sensing data interpretation method based on environmental cognition, which can realize real-time simulation of infrared characteristics of scenes with participating media.

[0005] In a first aspect, an embodiment of the present invention provides a SAR remote sensing data interpretation method based on environmental cognition, comprising:

[0006] Divide the SAR scene data into regions to obtain multiple regional scenes;

[0007] For each of the obtained SAR data of the regional scene, the environmental electromagnetic characteristics are calculated to obtain the characteristic quantity of the regional scene, wherein the characteristic quantity includes the backscattering coefficient and the probability distribution;

[0008] The feature quantity is input into a trained neural network recognition model, and after the neural network recognition model extracts features from the feature quantity, it outputs the specific type of the regional scene corresponding to the feature quantity.

[0009] In a second aspect, an embodiment of the present invention further provides a SAR remote sensing data interpretation device based on environmental cognition, which is used to implement any method described in the above claims, and the device comprises:

[0010] A division unit, used for dividing the SAR scene data into regions to obtain multiple regional scenes;

[0011] A data conversion unit is used to calculate the environmental electromagnetic characteristics of each of the obtained SAR data of the regional scene to obtain the characteristic quantity of the regional scene, wherein the characteristic quantity includes a backscattering coefficient and a probability distribution;

[0012] The recognition unit is used to input the feature quantity into a trained neural network recognition model, and after the neural network recognition model extracts features from the feature quantity, it outputs the specific type of the regional scene corresponding to the feature quantity.

[0013] In a third aspect, an embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in any embodiment of this specification is implemented.

[0014] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, enables the computer to execute the method described in any embodiment of this specification.

[0015] The embodiment of the present invention provides a SAR remote sensing data interpretation method, device, electronic device and storage medium based on environmental cognition. First, the environment of the SAR scene data is preprocessed and divided into regions; secondly, the environmental electromagnetic characteristics cognition research is carried out to obtain the main characteristic quantities describing the environmental electromagnetic characteristics, which are backscattering coefficient and probability distribution; finally, the overall architecture of SAR environmental characteristics intelligent fusion interpretation based on deep learning technology is used to realize the recognition and interpretation of targets and environments in SAR scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a navigation diagram based on the cognitive architecture of SAR environmental characteristics provided by one embodiment of the present invention;

[0018] Figure 2a This is a schematic diagram before filtering provided by an embodiment of the present invention;

[0019] Figure 2b is a filtered schematic diagram provided by an embodiment of the present invention;

[0020] Figure 3 is a schematic diagram of a binarization result provided by an embodiment of the present invention;

[0021] Figure 4 A SAR environment characteristic feature fusion interpretation diagram provided by an embodiment of the present invention;

[0022] Figure 5 This is a monitoring result diagram of a target in a complex environment provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0024] The specific implementation of the above concept is described below.

[0025] An embodiment of the present invention provides a SAR remote sensing data interpretation method based on environmental cognition, the method comprising:

[0026] Divide the SAR scene data into regions to obtain multiple regional scenes;

[0027] For each of the obtained SAR data of the regional scene, the environmental electromagnetic characteristics are calculated to obtain the characteristic quantity of the regional scene, wherein the characteristic quantity includes the backscattering coefficient and the probability distribution;

[0028] The feature quantity is input into a trained neural network recognition model, and after the neural network recognition model extracts features from the feature quantity, it outputs the specific type of the regional scene corresponding to the feature quantity.

[0029] First, the environment of the SAR scene data is preprocessed and divided into regions, and the logical reference Figure 1 A cognitive architecture for environmental characteristic data is proposed; secondly, research on the cognitive environment electromagnetic characteristics is conducted to obtain the main characteristic quantities describing the environmental electromagnetic characteristics, namely the backscattering coefficient and probability distribution; finally, an overall architecture for the intelligent fusion and interpretation of SAR environmental characteristics based on deep learning technology is established to realize the recognition and interpretation of targets and environments in SAR scenarios.

[0030] In some embodiments of the present invention, the SAR scene data is divided into regions to obtain multiple regional scenes, including:

[0031] Filtering the SAR scene data to obtain filtered data;

[0032] The filtered data is binarized to divide a plurality of scene areas.

[0033] Specifically, SAR images contain a large amount of speckle noise, which will affect the image quality. Filtering the image to remove speckle noise is the main content of SAR image preprocessing. Here, the Gaussian filter algorithm is used for filtering to reduce the impact of clutter on port extraction. Gaussian filtering convolves the image pixel by pixel through the Gaussian kernel to obtain the value of each pixel. Gaussian filtering is a low-pass filtering algorithm, or a data smoothing algorithm. During the convolution process, the Gaussian filter uses the values ​​of the surrounding pixels to calculate the pixel value at the center position. Therefore, the filtered image appears "blurred" visually and has a good effect on suppressing speckle noise. Figure 2 shows the results before and after filtering. It can be seen that the distinction between water and land in the filtered image is improved.

[0034] Image segmentation is a technique and process to divide an image into regions with different characteristics and extract the target of interest. The characteristics here can be grayscale, texture characteristics, etc. Only after the segmentation is completed can the segmented targets be located, classified and identified.

[0035] This patent uses the OTSU algorithm to determine the image adaptive threshold, and performs binarization processing on the filtered image to quantize the image into two gray levels, 0 and 1. The OTSU method is based on the one-dimensional grayscale histogram of the image, and uses the maximum inter-class variance of the target and background as the threshold selection criterion.

[0036] The experimental results are as follows Figure 3 As shown in the figure, the image after sea-land segmentation has theoretically separated the water area from the land area (the water area is marked black in the binary image, and the land area is marked white). Since the speckle noise of the SAR image cannot be completely removed, and the radar illumination angle causes the connected land or breakwater area to be disconnected on the current binary image. In addition, due to the influence of bright-toned targets (such as ships, islands, etc.) in dark-toned water areas and dark-toned targets (such as lakes, ponds, roads, etc.) in bright-toned land areas, there are still many small areas in the water area and holes on the land in the segmented area, and there are grid-shaped breeding areas, which brings trouble to subsequent processing. Therefore, it is necessary to use morphological processing methods to further remove speckle noise and eliminate holes.

[0037] In some embodiments of the present invention, the SAR data of each of the regional scenes obtained are subjected to environmental electromagnetic characteristic calculation to obtain a characteristic quantity of the regional scene, wherein the characteristic quantity includes a backscattering coefficient and a probability distribution, including:

[0038] For each of the obtained regional scenes, using the SAR scene data of the regional scene to determine the mean backscatter coefficient thereof;

[0039] The scene data of the area is used to model the amplitude distribution and obtain the probability distribution.

[0040] Specifically, the main characteristic quantities describing the electromagnetic characteristics of the environment are backscattering coefficient and probability distribution.

[0041] The backscatter coefficient is the radar cross section per unit area and is a dimensionless quantity. Due to the diversity of the terrain and vegetation cover on the ground, the movement of the waves, the movement of the scattering points in the radar beam illumination area and the changes in the scattering points will cause fluctuations in time and space. Therefore, ground / sea clutter is a random signal. The intensity of ground / sea clutter is usually characterized by its backscatter coefficient mean, which is generally a relationship between the scattering coefficient established uniformly in a small area and the rubbing angle (or incident angle), frequency (or band) and polarization.

[0042] The amplitude distribution modeling of ground and sea clutter usually adopts exponential distribution, gamma distribution, lognormal distribution, Weibull distribution, Rayleigh distribution, etc. In addition, K distribution is also a well-known distribution for describing SAR clutter. At present, generalized gamma distribution (GΓD) is also widely used in the field of signal processing, and has been proved by relevant experiments to be a probability distribution that can well fit the statistical behavior of SAR clutter. Therefore, this patent adopts GΓD for clutter distribution fitting.

[0043] The PDF (probability density function) of GΓD is:

[0044]

[0045] In the formula, σ, ν, κ are scale parameter, energy parameter and shape parameter respectively; Γ(·) represents the gamma function. In recent years, its improved form has appeared:

[0046]

[0047] It can be seen that when σ in formula (1) is When it is replaced, it is transformed into formula (2), that is, formula (2) is the generalization of formula (1).

[0048] After selecting the GΓD that fits the SAR clutter distribution, the next important step is to accurately estimate the three parameters of GΓD. Traditional probability distribution parameter estimation methods mainly include moment estimation (method of moments, MoM), maximum likelihood estimation (maximum likelihood estimation, MLE) and expectation maximization (expectation maximization, EM). However, many studies have shown that the best GΓD parameter estimation method is the logarithmic cumulants estimation method (method oflogarithmic cumulants, MoLC), which was proposed by Noicolas. Therefore, this method is used to estimate the GΓD distribution parameters of clutter.

[0049] After derivation by scholars, the third-order logarithmic cumulant of GΓD is obtained as:

[0050] c1=lnσ+(Ψ(κ)-lnκ) / ν (3)

[0051] c2=Ψ(1,κ) / ν 2 (4)

[0052] c3=Ψ(2,κ) / ν 3 (5)

[0053] Among them, Ψ(m,κ) is the m-th order polygamma function.

[0054] In practice, the logarithmic cumulant can be replaced by an estimate of the logarithmic cumulant of the observed data. For the observed data {xii=1,…,N}, the third-order logarithmic cumulant can be estimated by the following formula:

[0055]

[0056] Combining formulas (3)-(5) and (6), we get

[0057]

[0058] in, is the estimated value of κ. Based on formula (7), we can use numerical method to estimate and solve Then, two more parameters are estimated:

[0059]

[0060] Among them, sgn(·) is the sign function, and are the estimated values ​​of σ and ν, respectively.

[0061] The present invention adopts standard normal distribution as kernel function:

[0062]

[0063] Where exp(·) is an exponential function. The kernel density at any point x is estimated as:

[0064]

[0065] Among them, x1,x2,…,x N is the sample point, i.e. the value of the pixel in the SAR image, N is the total number of pixels, and h is the bandwidth of the kernel function. Since the choice of bandwidth h will affect Here, we select the optimal window width h according to the literature (43) 15 σN -15 , σ is the sample standard deviation.

[0066] In some embodiments of the present invention, the training method of the neural network recognition model includes:

[0067] Acquire multiple training samples; the multiple training samples include: SAR data of multiple regional scenes of a specific regional scene type;

[0068] The neural network is trained using the multiple training samples to obtain the neural network recognition model.

[0069] The overall architecture of intelligent fusion interpretation of SAR environmental features based on deep learning technology is shown in Figure 4.

[0070] (1) Data augmentation

[0071] First, the original SAR data set is augmented to expand the samples and overcome the problem of data imbalance. Data augmentation can not only improve the generalization ability of the model to prevent overfitting, but also improve the robustness of the model. Data augmentation is particularly suitable for target classification and recognition tasks because its category information does not change with the transformation of the image. Due to the characteristics of SAR images, the data augmentation method of geometric transformation is used in the experiment to expand the original data set. The process is as follows: crop the center of the image and only keep the central 480*480 pixel area; then rotate it by 45°, 120°, and -45° to obtain samples rotated at three different angles.

[0072] In addition, the mixup data enhancement method is also used. The core idea is to randomly select two images from each batch during training and mix them in a certain ratio to generate a new image. The formula is as follows:

[0073]

[0074] Among them, x i ,x j are two images randomly selected from a batch of the training set, y i ,y j is the corresponding one-hot label. Through the prior knowledge: the linear interpolation of the feature vector and the linear interpolation of the corresponding target are still corresponding, a new sample is constructed. Where λ is obtained through β(α,α) distribution and α is a hyperparameter.

[0075] (2) Recognition framework based on deep learning

[0076] The SAR dataset after data augmentation is input into the neural network to obtain the corresponding feature representation. The image feature extraction model used is the ResNet network pre-trained on the public dataset. The first layer of the network is a convolution layer with a convolution kernel of 7×7; in order to extract features useful for recognition, a CBAM attention module is added after the convolution layer, and the dimension of its output feature map is the same as the input dimension. Then the ResNet network is connected, where Layer1 consists of 3 Building Blocks; Layer2 consists of 4 Building Blocks; Layer3 consists of 6 Building Blocks; Layer4 consists of 4 Building Blocks. The ResNet network deepens the depth of the network by stacking Building Blocks, giving the network stronger expression capabilities. The Building Block is composed of two 3×3 convolutional layers stacked together, and each convolutional layer is followed by a batch normalization (batchnorm) layer and ReLU function. The role of batch normalization is to unify the distribution of each batch of data so that the distribution of input data can fall into the sensitive area of ​​the nonlinear activation function to accelerate network convergence and improve stability. For the ReLU function, batch normalization can more effectively prevent gradient explosion and gradient disappearance.

[0077] (3) Multi-dimensional feature fusion

[0078] Radar images are high-resolution images containing rich detail information. Therefore, using only ResNet to extract high-level semantic features will cause the details of the image to be ignored, thus affecting the recognition accuracy. Considering that the deep feature map has a large scale, it can capture the high-level semantic information of the salient object; while the shallow feature map has a small scale and can extract the fine semantic information of the target. Inspired by this method, the designed network structure not only extracts the global features of the image when obtaining the feature representation of the SAR ship image, but also obtains the feature maps of each layer, thus obtaining a set of multi-scale feature maps.

[0079] Specifically, the output feature maps of Layer1 and Layer3 of ResNet are upsampled respectively, and then the feature maps are spliced ​​to obtain the bottom-level features and high-level features respectively. Then the convolution layer is used to sample the bottom-level features so that their size matches the high-level features. The high-level features are scaled by convolution. The scaled bottom-level and high-level features are directly spliced ​​and passed through the Sigmoid activation function to obtain the detailed features of the SAR image.

[0080] Interpretation experiment and result analysis

[0081] (1) Dataset Introduction

[0082] SSDD (SAR Ship Detection Datasets) is the first publicly available dataset for ship target detection in SAR images. This dataset mainly covers ship targets in ocean and port environments.

[0083] It can be used to train and test detection algorithms. SSDD is produced by drawing on the production process of the PASCAL VOC dataset. The data mainly comes from SAR images acquired by RadarSat-2, TerraSAR-X and Sentinel-1 sensors, with four polarization modes of HH, HV, VV and VH, and a resolution range of 1m-15m. There are ship targets in large areas of open sea and nearshore areas. In the SSDD dataset, there are a total of 1160 images and 2456 ships, with an average of 2.12 ships per image. The suffix indexes 1 and 9 are selected as the test set (a total of 232 images), and the rest are used as the training set (a total of 928 images). In particular, the SSDD dataset provides two annotation methods, namely horizontal box annotation and rotated box annotation.

[0084] Table 1 SSDD dataset division

[0085]

[0086] (2) Object Detection Evaluation Criteria

[0087] The most commonly used evaluation criteria for target detection is mean Average Precision (mAP), which is the average of APs of each category. AP is the average precision of IoU from 0.50:0.05:0.95. APs, APm, and AP1 represent small targets, medium targets, and large targets, respectively. Average precision is the area under the precision-recall curve. Since there is only one category for ship detection, all APs are equivalent to mAP. Generally speaking, the better the detector, the higher the AP value. The main evaluation indicator of this patent is AP50, which refers to the AP with an IoU ratio of the detection box to the true box of more than 0.5. The evaluation criteria also include precision (Precision, p) and recall (Recall, r). Precision refers to the proportion of all positive samples that are correctly identified as positive samples. Recall is the proportion of predicted samples that are correctly identified as positive samples. The calculation method of the two evaluation indicators is as follows.

[0088]

[0089] Among them, TP (True Positive) is the number of correctly detected ships, FP (False Positive) is the number of ships incorrectly classified as positive, and FN (False Negative) is the number of ships incorrectly classified as negative.

[0090]

[0091] Where p represents precision, r represents recall, and p is a function with r as a parameter.

[0092] (3) Parameter settings

[0093] The software environment for this part of the experiment is Pytorch1.7 and Cuda11.0, and the GPU used is NVIDIA GeForceRTX 3070. The network is initialized using the Imagenet pre-trained weights, and the model is trained using the Stochastic Gradient Descent (SGD) algorithm, with the learning rate set to 0.005, the weight decay set to 0.0005, and the momentum set to 0.9.

[0094] (4) Experimental results and analysis

[0095] Figure 5 The prediction results of network detection of targets in different environments are given. The first row is the real label of the corresponding scene, and the second to fifth rows are Faster R-CNN (FPN), RetinaNet, and SSD respectively.

[0096] The above environmental recognition and interpretation method has clear logic. The interpretation architecture and method were tested in ocean scenes. The experimental results show that the SAR scene data is preprocessed based on the environmental characteristic classification architecture method proposed in this patent, and combined with the statistical prior knowledge of environmental characteristics. Based on the interpretation architecture proposed in this patent, it is feasible to identify and interpret targets in environmental scenes in measured SAR data.

[0097] An embodiment of the present invention provides a SAR remote sensing data interpretation device based on environmental cognition. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. From a hardware perspective, a hardware architecture diagram of an electronic device in which a SAR remote sensing data interpretation device based on environmental cognition provided in an embodiment of the present invention is located, in addition to a processor, memory, network interface, and non-volatile memory, the electronic device in the embodiment where the device is located can generally include other hardware, such as a forwarding chip responsible for processing messages, etc. Taking software implementation as an example, as a device in a logical sense, it is formed by the CPU of the electronic device in which it is located reading the corresponding computer program in the non-volatile memory into the memory and running it. A SAR remote sensing data interpretation device based on environmental cognition provided in this embodiment includes:

[0098] A division unit, used for dividing the SAR scene data into regions to obtain multiple regional scenes;

[0099] A data conversion unit is used to calculate the environmental electromagnetic characteristics of each of the obtained SAR data of the regional scene to obtain the characteristic quantity of the regional scene, wherein the characteristic quantity includes a backscattering coefficient and a probability distribution;

[0100] The recognition unit is used to input the feature quantity into a trained neural network recognition model, and after the neural network recognition model extracts features from the feature quantity, it outputs the specific type of the regional scene corresponding to the feature quantity.

[0101] It is to be understood that the structure illustrated in the embodiment of the present invention does not constitute a specific limitation on a SAR remote sensing data interpretation device based on environmental cognition. In other embodiments of the present invention, a SAR remote sensing data interpretation device based on environmental cognition may include more or fewer components than those shown in the figure, or combine some components, or split some components, or arrange the components differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0102] The information interaction, execution process and other contents between the modules in the above-mentioned device are based on the same concept as the embodiment of the method of the present invention. For the specific contents, please refer to the description in the embodiment of the method of the present invention, and no further description is given here.

[0103] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, a SAR remote sensing data interpretation method based on environmental cognition in any embodiment of the present invention is implemented.

[0104] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the processor executes a SAR remote sensing data interpretation method based on environmental cognition in any embodiment of the present invention.

[0105] Specifically, a system or device equipped with a storage medium can be provided, on which software program code that implements the functions of any of the above-mentioned embodiments is stored, and a computer (or CPU or MPU) of the system or device can be enabled to read and execute the program code stored in the storage medium.

[0106] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.

[0107] The storage medium embodiments for providing the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer by a communication network.

[0108] In addition, it should be clear that the functions of any of the above embodiments can be implemented not only by executing the program code read by the computer, but also by enabling an operating system operating on the computer to complete part or all of the actual operations based on instructions from the program code.

[0109] In addition, it can be understood that the program code read from the storage medium is written to a memory provided in an expansion board inserted into the computer or to a memory provided in an expansion module connected to the computer, and then based on the instructions of the program code, a CPU installed on the expansion board or expansion module is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above-mentioned embodiments.

[0110] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical factors in the process, method, article or device including the elements.

[0111] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A SAR remote sensing data interpretation method based on environmental cognition, characterized in that: include: Divide the SAR scene data into regions to obtain multiple regional scenes; For each of the obtained SAR data of the regional scene, the environmental electromagnetic characteristics are calculated to obtain the characteristic quantity of the regional scene, wherein the characteristic quantity includes the backscattering coefficient and the probability distribution; The feature quantity is input into a trained neural network recognition model, and after the neural network recognition model extracts features from the feature quantity, it outputs the specific type of the regional scene corresponding to the feature quantity.

2. The method according to claim 1, characterized in that The SAR scene data is divided into regions to obtain multiple regional scenes, including: Filtering the SAR scene data to obtain filtered data; The filtered data is binarized to divide a plurality of scene areas.

3. The method according to claim 2, characterized in that The filtering of the SAR scene data to obtain filtered data includes: Perform Gaussian filtering on the SAR scene data to obtain filtered data.

4. The method according to claim 1, characterized in that: The SAR data of each of the regional scenes is calculated for environmental electromagnetic characteristics to obtain feature quantities of the regional scene, wherein the feature quantities include backscatter coefficients and probability distributions, including: For each of the obtained regional scenes, using the SAR scene data of the regional scene to determine the mean backscatter coefficient thereof; The amplitude distribution model is built using the scene data of the area to obtain the probability distribution.

5. The method according to claim 4, characterized in that The method of using the regional scene data to perform amplitude distribution modeling to obtain probability distribution includes: According to the scene data of the area, the amplitude distribution is modeled using the generalized gamma distribution model to obtain the probability distribution.

6. The method according to claim 1, characterized in that The training method of the neural network recognition model includes: Acquire multiple training samples; the multiple training samples include: SAR data of multiple regional scenes of a specific regional scene type; The neural network is trained using the multiple training samples to obtain the neural network recognition model.

7. The method according to claim 6, characterized in that The multiple training samples also include augmented data of the SAR data of the multiple regional scenes, and the augmented data is obtained by cropping and rotating images in the SAR data, or by using a mixup data enhancement method.

8. A SAR remote sensing data interpretation device based on environmental cognition, characterized in that: For implementing the method described in any one of claims 1 to 7, the device comprises: A division unit, used for dividing the SAR scene data into regions to obtain multiple regional scenes; A data conversion unit is used to calculate the environmental electromagnetic characteristics of each of the obtained SAR data of the regional scene to obtain the characteristic quantity of the regional scene, wherein the characteristic quantity includes a backscattering coefficient and a probability distribution; The recognition unit is used to input the feature quantity into a trained neural network recognition model, and after the neural network recognition model extracts features from the feature quantity, it outputs the specific type of the regional scene corresponding to the feature quantity.

9. An electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 7.

Images