Target RCS intelligent prediction method and device based on SAR image, and storage medium

Abstract

The invention discloses a target RCS intelligent prediction method and device based on an SAR image, and a storage medium, relates to the field of microwave detection, and is used for improving the generalization ability of target RCS detection based on the SAR image. According to the method, the SAR image and the total RCS of the sample model are acquired, each group of scattering centers of the sample model in the SAR image are extracted, the first echo data and the second echo data are acquired after the echo signals of the scattering centers are superposed and stacked, the RCS of the second echo data is acquired based on the RCS calibration result, and then the PINN is trained to acquire the prediction model. And predicting the total RCS according to the SAR image of the target to be measured by using the prediction model. The method has good RCS prediction accuracy and generalization performance.

Description

Technical Field

[0001] The present invention relates to the field of microwave detection, and in particular to a method, a device and a storage medium for intelligently predicting a target RCS (Radar Cross Section) based on a SAR (Synthetic Aperture Radar) image. Background Art

[0002] Detecting the target RCS is of great significance for target identification (such as ship type, etc.). Correspondingly, detecting the target RCS also has great application value for studying target stealth technology, etc.

[0003] SAR can acquire high-resolution SAR images, and SAR images can reflect the backscattering characteristics of the target. Therefore, it is possible to consider detecting the target RCS through SAR images.

[0004] At present, most of the research on SAR images is based on obtaining the RCS of the target at some known angles through simulation and other means to obtain the RCS of the target at unknown angles. The RCS prediction accuracy of the target is relatively high. However, this method needs to obtain the RCS of the target at some angles first, which is not suitable for the detection of targets with undetected RCS and has insufficient generalization ability. Summary of the invention

[0005] The object of the present invention is to provide a method, device and storage medium for intelligent prediction of target RCS based on SAR images to improve the generalization ability of detecting target RCS based on SAR images in order to solve all or part of the above problems.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A target RCS intelligent prediction method based on SAR images, comprising:

[0008] Get the facet RCS and total RCS of the sample model at different azimuths;

[0009] Acquire a SAR image based on the bin RCS;

[0010] Performing RCS calibration on the SAR image;

[0011] Extracting each group of scattering centers of the sample model in the SAR image, each group of scattering centers correspondingly generates an echo signal;

[0012] Superimposing and stacking each group of echo signals respectively to obtain first echo data and second echo data;

[0013] Calculate the RCS of the second echo data based on the RCS calibration result;

[0014] With the goal of minimizing the total loss of the embedded physical information neural network, taking the total RCS and the first echo data as true values ​​and the RCS of the second echo data as input, training the embedded physical information neural network to obtain a prediction model;

[0015] The prediction model is used to predict the total RCS of the target to be measured based on the SAR image of the target to be measured.

[0016] In order to solve all or part of the above problems, the present invention also provides a storage medium, which stores a computer program. When the computer program is run, the above-mentioned target RCS intelligent prediction method based on SAR images is executed.

[0017] In order to solve all or part of the above problems, the present invention also provides a device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the processor runs the computer program, it executes the above-mentioned target RCS intelligent prediction method based on SAR images.

[0018] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0019] The target RCS intelligent prediction method based on SAR images in this application adopts an embedded physical information neural network as the prediction model architecture, which can train a prediction model with excellent performance under the conditions of limited target detection samples, and can accurately and comprehensively predict the unknown angles of sample targets and the RCS of unknown targets, with strong generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0021] Figure 1 The embodiment of the present application provides a training flow chart of a target RCS intelligent prediction method based on SAR images.

[0022] Figure 2 It is a schematic diagram of the RCS calibration results in the embodiment of the present application.

[0023] Figure 3 It is a schematic diagram of constructing the second echo data in an embodiment of the present application.

[0024] Figure 4 It is an architecture diagram of the embedded physical information neural network provided in an embodiment of the present application.

[0025] Figure 5 This is a sample example provided in the embodiments of the present application.

[0026] Figure 6 This is a comparison chart of the total RCS predicted by samples 1, 2, and 3 under different azimuth interval sampling in the embodiment of the present application.

[0027] Figure 7 This is a comparison chart of the total RCS predicted for sample No. 4 at different azimuth sampling intervals according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] All features disclosed in this specification, or steps in all methods or processes disclosed, except mutually exclusive features and / or steps, can be combined in any manner.

[0029] Any feature disclosed in this specification (including any additional claims and abstract), unless otherwise stated, may be replaced by other equivalent or alternative features having similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

[0030] In order to solve the problem that the current target RCS prediction method based on SAR images has insufficient generalization ability, the embodiments of the present application provide a target RCS intelligent prediction method, device and storage medium based on SAR images to improve the generalization ability of all-round prediction of target RCS based on SAR images.

[0031] In some embodiments, Figure 1 As shown in FIG. 1 , the target RCS intelligent prediction method based on SAR images includes the following processes:

[0032] S1. Obtain the facet RCS and total RCS of the sample model at different azimuth angles.

[0033] As an optional implementation, the GO (geometric optics)-PO (physical optics) hybrid method is used to calculate the backward RCS of the sample model under the HH polarization of the radar wave, and the surface element RCS and total RCS (Total RCS) of the sample model at different azimuth angles (0-360° range) are obtained.

[0034] In addition, as an optional implementation manner of RCS calibration, at least two small targets are arranged around the sample model as calibration points, and the RCS of each calibration point is known.

[0035] S2. Acquire SAR images based on the facet RCS.

[0036] Using the grid model and the back-projection algorithm (BPA), the signal echo of the radar under the direct side view is calculated according to the surface element RCS to obtain the SAR image.

[0037] As an optional implementation of RCS calibration, the obtained SAR image includes grayscale images of at least two set calibration points.

[0038] S3. Perform RCS calibration on the SAR image.

[0039] The so-called RCS calibration is to obtain the mapping relationship between RCS and amplitude value in the SAR image.

[0040] As an optional implementation, step S3 includes:

[0041] S3.1. Determine the amplitude value of each calibration point in the SAR image.

[0042] S3.2. Based on the amplitude value of each calibration point and the RCS of each calibration point, a mapping relationship between the RCS and the amplitude value of the SAR image is calibrated.

[0043] In some feasible implementations, the method for calibrating the mapping relationship between the RCS and the amplitude value of the SAR image is to fit the mapping relationship between the RCS and the amplitude value of the SAR image based on the amplitude value of each calibration point and the RCS of each calibration point. In some specific implementations, the mapping relationship between the RCS and the amplitude value of the SAR image is fitted by fitting the amplitude value of each calibration point and the RCS of each calibration point using a linear regression method.

[0044] like Figure 2 (a) is an example of arranged calibration points. Assume that there are four arranged calibration points, namely a, b, c and d. As an optional implementation, all calibration points are located at: corner positions relative to the sample model when it is located in the middle of the SAR image, for example Figure 2 In (a), when the sample model is located in the middle of the image, the four calibration points are distributed at the upper left corner and the lower right corner, which can avoid mutual interference caused by side lobes.

[0045] In the generated SAR image, the amplitude value of each calibration point is read out respectively, and the p x Represented by, where x∈{a,b,c,d}. At the same time, the RCS corresponding to the four calibration points is obtained to form the data set required for RCS calibration. The linear regression method is used to fit the data points in the data set, and the linear equation is used as the fitting target to fit the linear equation, which is expressed as:

[0046] σ x_dB =k×p x +b,

[0047] In the formula, σ x_dB It represents RCS in dB, k and b are the coefficients obtained by fitting. Figure 2(b) shows the result of linear fitting using the amplitude values ​​and RCS of the four calibration points, where the overall amplitude value of the SAR image is normalized before fitting.

[0048] S4. extracting each group of scattering centers of the sample model in the SAR image, and each group of scattering centers generates an echo signal correspondingly.

[0049] As an optional implementation, the orthogonal matching pursuit (OMP) method is used to extract the scattering center of the sample model in the SAR image, and several groups of scattering center parameters can be obtained depending on the scattering characteristics of the sample model. The corresponding echo signal can be generated by each group of scattering center parameters.

[0050] S5. Superimpose and stack each group of echo signals respectively to obtain first echo data and second echo data.

[0051] S5.1. Obtain the first echo data by superimposing the echo signals generated by each scattering center. The first echo data is a two-dimensional amplitude matrix, represented by M ASC .

[0052] S5.2. Obtain second echo data by stacking the echo signals generated by each scattering center on the image channel. The second echo data is a three-dimensional amplitude matrix, represented by {M j},j∈[1,n],M j represents the echo data of the echo signal generated by the jth scattering center, and n represents the number of scattering centers. Figure 3 As shown, Figure 3 Each rectangle in (a) represents an M j , Figure 3 (b) is the second echo data after stacking.

[0053] S6. Calculate the RCS of the second echo data based on the RCS calibration result.

[0054] In some feasible implementations, the method for calculating the RCS of the second echo data includes:

[0055] S6.1. Obtain the highest amplitude value of the echo signal corresponding to each scattering center respectively.

[0056] S6.2. Normalize the echo data of the echo signal corresponding to each scattering center based on the highest amplitude value corresponding to the scattering center.

[0057] Assume that the highest amplitude value of the echo data of the jth scattering center is A j , for M j , based on A j The normalization process is expressed as:

[0058]

[0059] In the formula, Indicates M j The normalized result of .

[0060] The echo data corresponding to each scattering center are normalized, and the normalized second echo data is expressed as

[0061] S6.3. Calculate the RCS corresponding to each highest amplitude value based on the RCS calibration result.

[0062] Through the RCS calibration results, we can obtain A j The corresponding RCS value is denoted as R j .

[0063] S6.4. Use the normalized echo data of each scattering center to weight the RCS corresponding to the highest amplitude value of the scattering center to obtain the RCS corresponding to the echo signal of each scattering center.

[0064] For the jth scattering center, the normalized echo data is The RCS value corresponding to the highest amplitude value is R j , then the RCS matrix S corresponding to the echo signal of the scattering center j Expressed as: S j =M j ·R j .

[0065] S6.5. The RCS corresponding to the echo signals of each scattering center is stacked to form the RCS of the second echo data, expressed as {S j}.

[0066] S7. With the goal of minimizing the total loss of the embedded physical information neural network, the total RCS and the first echo data are used as true values, and the RCS of the second echo data is used as input to train the embedded physical information neural network and obtain a prediction model.

[0067] The embedded physical information neural network is one of the key points to improve the prediction method of this application. In some optional implementations, such as Figure 4 As shown, the embedded physical information neural network includes a convolutional neural network CNN and a deep neural network DNN;

[0068] CNN includes seven convolutional layers, namely the first convolutional layer Conv0 to the seventh convolutional layer Conv6, wherein the first convolutional layer Conv0, the second convolutional layer Conv1, the third convolutional layer Conv2 and the sixth convolutional layer Conv5 are all connected to the batch normalization layer BatchNormalization; the first convolutional layer Conv0, the third convolutional layer Conv2, the fourth convolutional layer Conv3, the fifth convolutional layer Conv4 and the sixth convolutional layer Conv5 are all connected to the random inactivation layer Dropout; the first convolutional layer Conv0, the third convolutional layer Conv2 and the fourth convolutional layer Conv3 are all connected to the residual block Res (respectively Res0, Res1 and Res2); the third convolutional layer Conv2, the fifth convolutional layer Conv4 and the sixth convolutional layer Conv5 are all connected to the attention layer Se (respectively Se0, Se1 and Se2); the seventh convolutional layer Conv6 is connected to the PINN layer;

[0069] The DNN includes four DNN subnets, namely the first DNN subnet Dnn0 to the fourth DNN subnet Dnn3, wherein the first DNN subnet Dnn0, the second DNN subnet Dnn1 and the fourth DNN subnet Dnn3 are all connected with a batch normalization layer BatchNormalization; the second DNN subnet Dnn1 and the fourth DNN subnet Dnn3 are all connected with a random dropout layer Dropout;

[0070] CNN is connected to DNN through the flatten layer. The output layer of DNN is the output layer of the embedded physical information neural network, which is the output layer of the prediction model after training.

[0071] The total loss of the embedded physical information neural network includes physical information loss PINN Loss and data information loss Data Loss. Figure 1 As shown, the total RCS is used as the truth value to optimize the data information loss Data Loss, and the first echo data is used as the truth value to optimize the physical information loss PINNLoss.

[0072] During the training process, get the matrix I output by the seventh convolutional layer Conv6 pred , the number of filters in this layer is 1 (filters = 1), so I pred It is single channel data. pred is regarded as the scattering center image reconstructed by the neural network, I pred The shape is similar to the first echo data M ASC The target shape is similar, but its amplitude is inverted through a neural network to predict the target RCS information. Based on this, the physical information loss PINN Loss is designed. pred With MASC The structural similarity of the target in the dataset can be obtained by using the structural similarity loss function (SSIM) to obtain PINNLoss. DataLoss can be obtained using the Huber function.

[0073] When calculating PINN Loss, normalization processing is required. In some optional implementations, for I pred The normalization process is:

[0074]

[0075] In the formula, I norm represents the normalized result, Indicates I pred The average value, σ I represents the standard deviation of the image, and ε1 is a small constant term used to prevent division by 0. ASC The normalization process is similar.

[0076] S8. Using the prediction model, predict the total RCS of the target to be measured based on the SAR image of the target to be measured.

[0077] For SAR images of the target to be measured, see Figure 1 , the corresponding second echo data is obtained by the orthogonal matching pursuit method, and then the RCS of the second echo data is obtained by the RCS calibration result, which is input into the prediction model to obtain the omnidirectional RCS of the target to be measured.

[0078] The present application example also verifies the performance of the above prediction method. Figure 5 As shown, the embodiment of the present application provides four different types of ship targets as experimental samples, which are marked as samples 1 to 4 (represented as shiptarget1, shiptarget2, shiptarget3 and shiptarget4). Sample 4 does not participate in training and is used for generalization ability verification.

[0079] 7879 sets of data from different azimuths of samples 1, 2 and 3 (random sampling with a minimum interval of 0.1°, the theoretical maximum number of samples is 10800) M ASC and total RCS) were prepared by the above-mentioned method. Among them, 4076 sets of data were randomly assigned to participate in training, and compared with the maximum sampling number, the proportion of data participating in training was about 37.74%.

[0080] In the training data, 80% is used as training set and 20% is used as validation set. As the input layer data of the neural network with embedded physical information; M ASCParticipate in the calculation of PINNLoss; the total RCS is used as the true value data of the Output layer of the embedded physical information neural network and participates in the calculation of DataLoss. The loss value of Data Loss is obtained using the Huber function. The total loss TotalLoss can be expressed as:

[0081] Total Loss=W1×Data Loss+W2×PINN Loss,

[0082] As an example, in the formula, W1=W2=0.5 means that the loss values ​​calculated by DataLoss and PINNLoss have the same weight during training.

[0083] After training, the 3803 sets of data that did not participate in the training were used as comparison samples. The total RCS of samples 1, 2, and 3 was predicted. The results are as follows: Figure 6 In addition, to further evaluate the prediction performance, comparative samples with azimuth intervals of approximately 1°, 5°, and 10° were selected for further display. The results are shown in Figure 2. Figure 6 (b) Figure 6 (c) and Figure 6 (d) The total RCS calculated by the GO-PO hybrid method is represented by the blue line as the true value; Figure 6 (a) and Figure 6 In (b), the predicted values ​​of the prediction model are represented by scattered points. Figure 6 (c) and Figure 6 In (d), the predicted values ​​of the prediction model are connected by red lines. Since the test set is obtained by random sampling of the total samples, it is difficult to find all the comparison samples at strict angle intervals. Since the RCS data used in the training process of the embedded physical information neural network is expressed in decibels (dB), Figure 6 The mean value shown in is the dB average result of the directly calculated RCS value. This method can intuitively reflect the prediction performance of the model. From the comparison results, it can be seen that the prediction model shows excellent prediction performance under different sampling intervals.

[0084] Depend on Figure 6 It can be seen that the prediction model predicts better for samples with strong total RCS median values, and the predicted points with better predictions under continuous azimuths are consistent with the trend of the true value. From the perspective of the mean, the prediction error of all samples is less than 1dBsm, and there is no significant difference in the prediction performance under different azimuth sampling intervals, which verifies that the prediction model has good prediction performance under the training condition of limited target detection samples.

[0085] The generalization performance of the prediction model is tested using sample No. 4, which has never participated in training. The prediction results are as follows: Figure 7 As shown. It can be seen that when the sampling interval is about 5° and 10° (5° corresponds to Figure 7 (a), 10° corresponds to Figure 7 (b)), the mean error of the prediction model is higher than that of samples 1, 2, and 3 that participated in the training, but it can still basically match the trend of the true value in continuous azimuths, especially for samples with a strong total RCS median. This shows that the prediction model has good generalization ability.

[0086] Based on the same purpose, an embodiment of the present application further provides a storage medium, which stores a computer program. When the computer program is run, the target RCS intelligent prediction method based on SAR images of the above embodiment is executed.

[0087] In addition, an embodiment of the present application further provides a device, including a storage medium and a processor, wherein a computer program is stored in the storage medium, and when the processor runs the computer program, the target RCS intelligent prediction method based on SAR images of the above embodiment is executed.

[0088] The present invention is not limited to the above-mentioned specific embodiments, but extends to any new features or any new combination disclosed in this specification, as well as any new method or process steps or any new combination disclosed.

Claims

1. A target RCS intelligent prediction method based on SAR images, characterized in that: include: Get the facet RCS and total RCS of the sample model at different azimuths; Acquire a SAR image based on the bin RCS; Performing RCS calibration on the SAR image; Extracting each group of scattering centers of the sample model in the SAR image, each group of scattering centers correspondingly generates an echo signal; Superimposing and stacking each group of echo signals respectively to obtain first echo data and second echo data; Calculate the RCS of the second echo data based on the RCS calibration result; With the goal of minimizing the total loss of the embedded physical information neural network, taking the total RCS and the first echo data as true values ​​and the RCS of the second echo data as input, training the embedded physical information neural network to obtain a prediction model; The prediction model is used to predict the total RCS of the target to be measured based on the SAR image of the target to be measured.

2. The target RCS intelligent prediction method based on SAR image according to claim 1, characterized in that: The SAR image contains at least two calibration points; Performing RCS calibration based on the SAR image includes: Determining the amplitude value of each of the calibration points in the SAR image; Based on the amplitude value of each calibration point and the RCS of each calibration point, a mapping relationship between the RCS and the amplitude value of the SAR image is calibrated.

3. The target RCS intelligent prediction method based on SAR image according to claim 2, characterized in that: The step of calibrating a mapping relationship between the RCS and the amplitude value of the SAR image based on the amplitude value of each calibration point and the RCS of each calibration point includes: Based on the amplitude value of each calibration point and the RCS of each calibration point, a mapping relationship between the RCS and the amplitude value of the SAR image is fitted.

4. The target RCS intelligent prediction method based on SAR image according to claim 3, characterized in that: The mapping relationship between the RCS and the amplitude value of the SAR image is fitted by fitting the amplitude value of each calibration point and the RCS of each calibration point using a linear regression method.

5. The target RCS intelligent prediction method based on SAR image according to any one of claims 2 to 4, characterized in that: Each of the calibration points is located at a corner position relative to the sample model when it is located in the middle of the SAR image.

6. The target RCS intelligent prediction method based on SAR image according to claim 1, characterized in that: The total loss of the embedded physical information neural network includes physical information loss and data information loss, the total RCS is used to optimize the data information loss, and the first echo data is used to optimize the physical information loss.

7. The target RCS intelligent prediction method based on SAR image according to claim 1, characterized in that: Calculating the RCS of the second echo data based on the RCS calibration result includes: Obtain the highest amplitude value of the echo signal corresponding to each scattering center respectively; Normalizing the echo data of the echo signal corresponding to each scattering center based on the highest amplitude value corresponding to the scattering center; Based on the RCS calibration results, the RCS corresponding to each highest amplitude value is calculated respectively; The normalized echo data of each scattering center is used to weight the RCS corresponding to the highest amplitude value of the scattering center to obtain the RCS corresponding to the echo signal of each scattering center; The RCS corresponding to the echo signals of each scattering center is stacked to form the RCS of the second echo data.

8. The target RCS intelligent prediction method based on SAR images as claimed in claim 1, characterized in that: The embedded physical information neural network includes a convolutional neural network CNN and a deep neural network DNN; The CNN includes seven convolutional layers, wherein the first convolutional layer, the second convolutional layer, the third convolutional layer and the sixth convolutional layer are all connected to a batch normalization layer; the first convolutional layer, the third convolutional layer, the fourth convolutional layer, the fifth convolutional layer and the sixth convolutional layer are all connected to a random dropout layer; the first convolutional layer, the third convolutional layer and the fourth convolutional layer are all connected to a residual block; the third convolutional layer, the fifth convolutional layer and the sixth convolutional layer are all connected to an attention layer; the seventh convolutional layer is connected to a PINN layer; The DNN includes four DNN subnets, wherein the first DNN subnet, the second DNN subnet and the fourth DNN subnet are all connected with a batch normalization layer; the second DNN subnet and the fourth DNN subnet are all connected with a random dropout layer; The CNN is connected to the DNN via a flattening layer.

9. A storage medium storing a computer program, characterized in that: When the computer program is run, the target RCS intelligent prediction method based on SAR images as described in any one of claims 1 to 8 is executed.

10. A device comprising a storage medium and a processor, wherein the storage medium stores a computer program, characterized in that: When the processor runs the computer program, it executes the target RCS intelligent prediction method based on SAR images as described in any one of claims 1 to 8.

Images