A method for suppressing chaff jamming based on two-stage UNet

Abstract

The present application belongs to the technical field of electronic countermeasures, jamming suppression and signal recognition, and particularly relates to a shielding chaff jamming suppression method based on a two-stage UNet. The method comprises the following steps: acquiring signals using a wideband polarization radar to obtain measured data; performing pulse compression and normalization to obtain target, chaff jamming and mixed range image sequence gray maps; constructing pseudo-measured data to obtain target, chaff jamming, pseudo-measured target and chaff jamming mixed range image sequence gray maps and segmentation true value maps; constructing a chaff jamming suppression network based on a two-stage UNet and a loss function thereof; training using the pseudo-measured data to obtain an optimal chaff jamming suppression network based on a two-stage UNet; and testing to achieve the suppression of shielding chaff jamming. The method has good segmentation effect on the target foreground under shielding chaff jamming, good suppression effect on chaff jamming, small environmental influence and the ability to cope with various types and stages of chaff jamming.

Description

Technical Field

[0001] The present invention belongs to the technical fields of electronic countermeasures, interference suppression and signal recognition, and in particular relates to a shielded chaff interference suppression method based on a dual-stage UNet. Background Art

[0002] Radar active homing guidance has been widely used due to its advantages, including long range, high detection accuracy, and all-day, all-weather operation. However, it also suffers from the disadvantage of being susceptible to the battlefield electromagnetic environment. Chaff jamming, one of the oldest passive jammers, involves projecting a large number of chaff fibers to scatter radar electromagnetic waves, generating strong echoes with clutter characteristics in the radar receiver. This achieves electromagnetic deception or suppression of the radar, effectively protecting friendly targets. It plays a vital role in electronic countermeasures and is a major threat to radar seekers. In particular, shielded chaff jamming not only produces chaff jamming echoes that are several times stronger than the ship's echo, thereby suppressing or drowning the ship's echo within the airspace, but also attenuates radar electromagnetic waves, creating an electromagnetic shield for the ship, rendering it undetectable by the radar and having a similar effect to "stealth." Therefore, to increase the probability of radar target detection, improving the radar seeker's ability to counter shielded chaff jamming remains a hot topic and a difficult problem both domestically and internationally.

[0003] Currently, chaff suppression methods primarily include: designing filters in the wavelet transform domain to suppress chaff interference; designing Doppler filters in the frequency domain to suppress chaff interference; suppressing chaff interference in the polarization domain through polarization cancellation; and using the principle of optimal space-time processing to derive an optimal space-time wave velocity former to suppress chaff cloud interference. These methods can suppress chaff interference to a certain extent, but due to the complex diffusion process and variable characteristics of chaff clouds, existing countermeasures struggle to achieve both effective suppression of chaff interference and universal applicability to chaff interference at any diffusion stage. Furthermore, their suppression of shielded chaff interference is insufficient to meet practical requirements. Summary of the Invention

[0004] The purpose of the present invention is to address the problems of complex echo data and low target detection probability of targets under shielded chaff interference. A shielded chaff interference suppression method based on a two-stage UNet is proposed. The method can not only accurately identify the range resolution unit of the target under shielded chaff interference, but also improve the signal-to-interference ratio of the target.

[0005] In order to achieve the above objectives, the following technical solutions are adopted:

[0006] The shielded chaff interference suppression method comprises the following steps:

[0007] S1. Using a broadband polarimetric radar, collecting target echoes under diluted chaff interference and target echoes under shielded chaff interference, and recording the position, angle, and velocity information of the target during data collection, to obtain target echoes under diluted chaff interference and target echoes under shielded chaff interference with real information;

[0008] S2, pulse compressing and normalizing the target echo under the diluted chaff interference with real information obtained in S1 to obtain a target range image sequence grayscale image and a chaff interference range image sequence grayscale image; pulse compressing and normalizing the target echo under the masked chaff interference with real information obtained in S1 to obtain a mixed range image sequence grayscale image of the target and the chaff interference;

[0009] The S2 target range image sequence grayscale image, the chaff interference range image sequence grayscale image and the target and chaff interference mixed range image sequence grayscale image all include X range resolution units, where X is the number of sampling points.

[0010] S3, using the target range image sequence grayscale image and the chaff interference range image sequence grayscale image obtained in S2 to construct pseudo measured data, and obtaining a target range image sequence true value image, a chaff interference range image sequence true value image, a pseudo measured target and chaff interference mixed range image sequence grayscale image, and a segmentation true value image;

[0011] S4. Construct a chaff interference suppression network based on a two-stage UNet;

[0012] The chaff interference suppression network based on the dual-stage UNet is composed of two UNet cascades, namely the first-stage UNet and the second-stage UNet;

[0013] During training or verification, the input image of the chaff interference suppression network based on the dual-stage UNet is a grayscale image of a range image sequence of a mixture of a pseudo-measured target and chaff interference; during testing, the input image of the chaff interference suppression network based on the dual-stage UNet is a grayscale image of a range image sequence of a mixture of a target and chaff interference;

[0014] In the first stage, UNet segments the input image into target foreground and background to obtain the segmentation result;

[0015] The label of the segmentation result, i.e. the segmentation label, is the segmentation truth map;

[0016] The second-stage UNet uses the segmentation result of the first-stage UNet and the result of the dot product of the input image of the first-stage UNet as the input image of the second-stage UNet and processes it to obtain a suppression result, where the suppression result is the result of restoring the target amplitude information in the input image of the second-stage UNet;

[0017] The label of the suppression result, i.e., the suppression label, is a true value image of the target range image sequence;

[0018] The input image, segmentation result, segmentation label, suppression result, and suppression label of the two-stage UNet-based chaff interference suppression network are of the same size;

[0019] S5. Construct the loss function of the chaff interference suppression network based on the two-stage UNet;

[0020] The calculation formula of the loss function described in S5 is:

[0021] (1)

[0022] in, They are the cross entropy loss and Dice loss used by UNet in the first stage and the smooth L1 loss used by UNet in the second stage; a is the weight of the cross entropy loss; b is the weight of the Dice loss; c is the weight of the smooth L1 loss;

[0023] The cross entropy loss used by UNet in the first stage is defined as:

[0024] (2)

[0025] N is the number of pixels in a batch of input images, is the pixel number; M represents the number of pixel categories. In the chaff interference suppression scenario, there are only two categories: target foreground and background, so M=2. is the category number, Indicates that the category is target foreground, Indicates that the category is background; is the pixel of the input image The category label, Represents the pixel points of the input image Not in category , Represents the pixel points of the input image Belong to category ; Represents the pixel point of the predicted input image Belong to category probability;

[0026] The calculation formula of the Dice loss is:

[0027] (3)

[0028] Where P represents the segmentation result, G represents the segmentation label, |G∩P| is the sum of the results of the segmentation result and the pixel product of the segmentation label, |G| and |P| are the sum of all elements of the segmentation label and the segmentation result respectively; the Dice loss range is between [0,1], which is the probability of the non-overlapping part between the segmentation result area and the segmentation label area;

[0029] The second stage UNet uses smooth L1 loss The mean absolute error of pixels;

[0030] S6. Model tuning: Grid search is used to obtain the optimal hyperparameters for the two-stage UNet-based chaff suppression network.

[0031] S7, network training: using the RMSprop optimization algorithm to update the parameters of the dual-stage UNet-based chaff interference suppression network with the optimal hyperparameters obtained in S6, so that the loss function of the dual-stage UNet-based chaff interference suppression network constructed in S5 is minimized, thereby obtaining the optimal dual-stage UNet-based chaff interference suppression network;

[0032] S8, network test: Input the grayscale image of the mixed range image sequence of the target and the chaff interference obtained in S2 into the optimal chaff interference suppression network based on the two-stage UNet obtained in S7 for testing to obtain the segmentation result and suppression result.

[0033] The S3 specifically includes the following sub-steps:

[0034] S31, randomly selecting P slow time series of range resolution units from the target range image sequence grayscale image obtained in S2, dividing them by the maximum value of the series for normalization, and reconstructing the target range image sequence ground truth image according to the spatial distribution and scattering intensity of the target scattering points;

[0035] S32, randomly selecting Q slow time series of range resolution units from the grayscale image of the chaff interference range image sequence obtained in S2, dividing the value in the slow time series by the maximum value of the series to perform normalization, and then multiplying it by the weight coefficient of the normal distribution curve to obtain a true value image of the chaff interference range image sequence;

[0036] S33, performing weighted summation on the target range image sequence truth image obtained in S31 and the chaff interference range image sequence truth image obtained in S32 to obtain a grayscale image of a range image sequence of a mixture of pseudo-measured target and chaff interference;

[0037] S34 , subjecting the target range image sequence truth map obtained in S31 to CA-CFAR detection to obtain a segmentation truth map.

[0038] P in S31 is less than X; Q in S32 is less than X; when performing weighted summation in S33, the ratio of weights is the signal-to-interference ratio; the value of a pixel point in the segmentation truth map in S34 is 1, indicating that the value of the corresponding pixel point in the target range image sequence truth map is greater than the CA-CFAR detector threshold; and the value of 0 indicates that the value of the corresponding pixel point in the target range image sequence truth map is less than or equal to the CA-CFAR detector threshold.

[0039] S4 Both the first-stage UNet and the second-stage UNet include one input layer, three feature extraction layers, three feature enhancement layers and one output layer; the input layer includes two convolutional layers with a kernel size of 3×3.

[0040] The feature extraction layer includes a pooling layer with a kernel size of 2×2 and two convolution layers with a kernel size of 3×3; the feature enhancement layer includes a deconvolution layer with a kernel size of 2×2 and two convolution layers with a kernel size of 3×3; the output layer is a convolution layer with a kernel size of 1×1 and an output channel number of 2 or 1; the number of channels of the output layer of the first-stage UNet is 2, and the number of output channels of the second-stage UNet is 1.

[0041] The absolute error of the pixel point S5 is Indicates that, and

[0042]

[0043] and are the pixels of the suppression results and suppression labels respectively, where The value range is [1,N].

[0044] During model tuning and network training in S6 and S7, the input image, segmentation label and suppression label of the chaff interference suppression network based on the two-stage UNet are respectively the grayscale image of the range image sequence of the pseudo-measured target and chaff interference mixture obtained in S3, the segmentation true value image and the target range image sequence true value image.

[0045] Beneficial effects

[0046] The shielded chaff interference suppression method based on a dual-stage UNet described in the present invention has the following beneficial effects compared with existing chaff interference suppression methods:

[0047] 1. The method benefits from S3's construction of pseudo-measured data, eliminating the tedious process of manual labeling in previous supervised learning, saving manpower and time;

[0048] 2. The shielded chaff interference suppression method benefits from the excellent performance of the first-stage UNet of the two-stage UNet-based chaff interference suppression network constructed by S4 in the task of segmenting the target foreground and background. The shielded chaff interference suppression method has good recognition effect and high segmentation accuracy for the range resolution unit of the target under shielded chaff interference;

[0049] 3. This method benefits from the dual-stage UNet-based chaff interference suppression network constructed by S4 and the training of the dual-stage UNet-based chaff interference suppression network by S5. It avoids the shortcomings of traditional methods that rely too much on prior knowledge, can effectively suppress chaff interference, and improves the signal-to-interference ratio more than existing methods.

[0050] 4. The method benefits from the high generalization and robustness of the chaff interference suppression network based on the two-stage UNet constructed by S4. Compared with existing methods, it is less affected by complex environments and can cope with chaff interference at all stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of a shielded chaff interference suppression method based on a two-stage UNet of the present invention;

[0052] Figure 2 The present invention provides a method for suppressing shielded chaff interference based on a dual-stage UNet, comprising a target range image sequence grayscale image, a chaff interference range image sequence grayscale image, and a target and chaff interference mixed range image sequence grayscale image;

[0053] Figure 3 The present invention is a shielded chaff interference suppression method based on a two-stage UNet, which is constructed according to measured data, including a target range image sequence truth map, a chaff interference range image sequence truth map, a pseudo-measured target and chaff interference mixed range image sequence grayscale map, and a segmentation truth map;

[0054] Figure 4 It is a diagram of a network structure of chaff interference suppression based on a dual-stage UNet of a shielded chaff interference suppression method of the present invention;

[0055] Figure 5 The present invention is a method for suppressing shielded chaff interference based on a two-stage UNet. The method comprises grayscale images of a mixed range image sequence of pseudo-measured targets and chaff interference at different interference-to-signal ratios, segmentation results, suppression results, segmentation truth images, and target range image sequence truth images.

[0056] Figure 6 The invention discloses a method for suppressing shielded chaff interference based on a two-stage UNet, which includes grayscale images of a mixed range image sequence of a target and chaff interference, segmentation results, and suppression results. DETAILED DESCRIPTION

[0057] The following is a further explanation and detailed description of the shielded chaff interference suppression method based on the dual-stage UNet of the present invention in conjunction with the accompanying drawings and embodiments.

[0058] Example 1

[0059] This example illustrates the process of using the method described in this invention to suppress target echo data received by a wideband polarimetric radar against a chaff jamming background. The method proposed in this invention is theoretically applicable to target recognition and jamming, clutter, or noise countermeasures in any jamming, clutter, or noise environment. It should be noted that the dataset required is a set of images or sequences containing distinguishable target features to be suppressed. Figure 1 The present invention is a process of a shielded chaff interference suppression method based on a two-stage UNet, comprising the following steps:

[0060] S1, data acquisition; in specific implementation, a broadband polarization radar is used to collect target echoes under diluted chaff interference and target echoes under shielded chaff interference, and the position, angle and velocity information of the target are recorded when collecting data to obtain target echoes under diluted chaff interference and target echoes under shielded chaff interference with real information;

[0061] The broadband polarimetric radar is a Ku-band high-resolution polarimetric radar system, and its transmission signal is a periodic linear frequency modulation signal consisting of 90 pulsed horizontal polarization waves and 70 vertical polarization waves;

[0062] The broadband polarimetric radar is placed on a platform 15 meters above the sea surface when collecting target echoes under diluted chaff interference and target echoes under shielded chaff interference, with the background being a sea with moderate wind and waves of 0.1 to 0.5 meters;

[0063] The target echo under diluted chaff interference collected by the broadband polarimetric radar is 8000 meters away from the radar, the distance between the chaff interference and the radar is between 7600 meters and 7900 meters, the target and the chaff interference are not aliased at the range resolution unit, and contains 8000 pulses;

[0064] The target echo under shielded chaff interference collected by the broadband polarimetric radar, the distance between the chaff interference and the radar ranges from 7900 meters to 8200 meters, the target is a ship 8000 meters away from the radar, the target is within the coverage of the chaff interference, the target and the chaff interference are aliased at the range resolution unit, and contains 8000 pulses;

[0065] S2, pulse compression, specifically: pulse compressing and normalizing the target echo under the diluted chaff interference with real information obtained in S1 to obtain a target range image sequence grayscale image and a chaff interference range image sequence grayscale image; pulse compressing and normalizing the target echo under the shielded chaff interference with real information obtained in S1 to obtain a mixed range image sequence grayscale image of the target and the chaff interference;

[0066] The target range image sequence grayscale image, the chaff interference range image sequence grayscale image and the target and chaff interference mixed range image sequence grayscale image each include X range resolution units;

[0067] Figure 2 The present invention provides a target range image sequence grayscale image (2a), a chaff interference range image sequence grayscale image (2b), and a target and chaff interference mixed range image sequence grayscale image (2c) in a shielded chaff interference suppression method based on a dual-stage UNet.

[0068] S3, constructing pseudo measured data, that is, using the target range image sequence grayscale image and the chaff interference range image sequence grayscale image obtained in S2 to construct pseudo measured data, obtaining the target range image sequence true value image, the chaff interference range image sequence true value image, the pseudo measured target and chaff interference mixed range image sequence grayscale image and the segmentation true value image, specifically including the following sub-steps:

[0069] S31, randomly selecting N slow time series of range resolution units from the target range image sequence grayscale image obtained in S2, dividing them by the maximum value of the series for normalization, and reconstructing the target range image sequence true value image according to the spatial distribution and scattering intensity of the target scattering points;

[0070] S32, randomly selecting M slow time series of range resolution units from the grayscale image of the chaff interference range image sequence obtained in S2, dividing the value in the slow time series by the maximum value of the series to perform normalization, and then multiplying it by the weight coefficient of the normal distribution curve to obtain a true value image of the chaff interference range image sequence;

[0071] S33, according to the interference-to-signal ratio, performing weighted summation of the target range image sequence truth image obtained in S31 and the chaff interference range image sequence truth image obtained in S32 to obtain a grayscale image of a range image sequence of a mixture of the pseudo-measured target and the chaff interference;

[0072] S34, subjecting the target range image sequence truth map obtained in S31 to CA-CFAR detection to obtain a CA-CFAR detection result, and using the CA-CFAR detection result as the segmentation truth map (e.g., Figure 3 3d in);

[0073] For a certain pixel point of the segmentation truth map, if the value of the corresponding pixel point in the target range image sequence truth map is greater than the CA-CFAR detector threshold, the value of the pixel point is 1, otherwise the value of the pixel point is 0;

[0074] Figure 3 The present invention provides a shielded chaff interference suppression method based on a dual-stage UNet, comprising a target range image sequence truth map (3a), a chaff interference range image sequence truth map (3b), a pseudo-measured target and chaff interference mixed range image sequence grayscale map (3c), and a segmentation truth map (3d) constructed based on measured data.

[0075] S4. Construct a chaff interference suppression network based on a two-stage UNet;

[0076] The chaff interference suppression network based on the dual-stage UNet is composed of two UNet cascades, namely the first-stage UNet and the second-stage UNet;

[0077] The function of the first-stage UNet is to use the grayscale image of the mixed range image sequence of the target and the chaff interference or the grayscale image of the mixed range image sequence of the pseudo-measured target and the chaff interference as the input image of the first-stage UNet, and process it to obtain a segmentation result, which is the result of segmenting the target foreground and background of the input image of the first-stage UNet;

[0078] The function of the second stage UNet is to take the segmentation result of the first stage UNet and the result of the dot product of the input image of the first stage UNet as the input image of the second stage UNet, and process it to obtain a suppression result, which is the result of restoring the target amplitude information in the input image of the second stage UNet;

[0079] The first-stage UNet and the second-stage UNet both include one input layer, three feature extraction layers, three feature enhancement layers, and one output layer;

[0080] The input layer includes two convolutional layers with a kernel size of 3×3;

[0081] The feature extraction layer includes a pooling layer with a kernel size of 2×2 and two convolutional layers with a kernel size of 3×3;

[0082] The feature enhancement layer includes a deconvolution layer with a kernel size of 2×2 and two convolution layers with a kernel size of 3×3;

[0083] The output layer is a convolutional layer with a kernel size of 1×1 and an output channel number of 2 or 1; the output channel number of the first-stage UNet is 2, and the output channel number of the second-stage UNet is 1;

[0084] Figure 4 It is a diagram of a network structure of chaff interference suppression based on a dual-stage UNet of a shielded chaff interference suppression method of the present invention;

[0085] S5. Construct the loss function of the chaff interference suppression network based on the two-stage UNet;

[0086] The calculation formula of the loss function is:

[0087] (11)

[0088] in, They are the cross entropy loss and Dice loss used by UNet in the first stage and the smooth L1 loss used by UNet in the second stage; a is the weight of the cross entropy loss; b is the weight of the Dice loss; c is the weight of the smooth L1 loss;

[0089] Specifically, in the first stage, UNet uses the sum of cross entropy loss and Dice loss to measure the similarity between the segmentation result of the segmentation network and the segmentation ground truth map;

[0090] The cross entropy loss is defined as:

[0091] (12)

[0092] Where N is the number of pixels in a batch, and M represents the number of pixel categories. In the chaff interference suppression scenario, there are only two categories: target foreground and background, so M = 2. is the category label of pixel n, which is a one-hot encoded vector with only two values ​​of 0 and 1. Represents the probability that the predicted pixel n belongs to category c;

[0093] The Dice loss, also known as the overlap index, is used to calculate the overlap rate between the segmentation result area and the segmentation ground truth area, ranging from [0, 1]. The calculation formula is:

[0094] (13)

[0095] P represents the segmentation result, G represents the segmentation truth map, |G∩P| is the sum of the results of the segmentation result and the corresponding pixel multiplication of the segmentation truth map, |G| and |P| are the sum of all elements of the segmentation truth map and the segmentation result respectively.

[0096] It can be concluded that the calculation formula of Dice loss is:

[0097] (14)

[0098] In order to avoid the defects of L1 and mean square error, the second stage UNet uses smooth L1 loss, and the specific calculation formula is:

[0099] (15)

[0100] in and are the pixels in the network restored image and the segmentation truth image in the recovery stage, is the absolute error of the pixel.

[0101] S5. Using the grayscale image of the mixed range image sequence of the pseudo-measured target and chaff interference obtained in S3 and the ground truth image of the target range image sequence as the input image, segmentation label, and output label of the second stage of the chaff interference suppression network based on the two-stage UNet obtained in S4, respectively, and using the grid search method to obtain the optimal hyperparameters of the chaff interference suppression network based on the two-stage UNet;

[0102] S6. Using the grayscale image of the mixed range image sequence of the pseudo-measured target and chaff interference, the segmentation truth image, and the target range image sequence truth image obtained in S3 as the input image, segmentation label, and output label of the second stage of the chaff interference suppression network based on the two-stage UNet obtained in S4, respectively, and using the grid search method to obtain the optimal hyperparameters of the chaff interference suppression network based on the two-stage UNet;

[0103] S7. Use the grayscale image of the mixed range image sequence of pseudo-measured targets and foil interference, the segmentation true value image and the target range image sequence true value image obtained in S3 as the input image, segmentation label and the output label of the second stage of the foil interference suppression network based on the two-stage UNet with the optimal hyperparameters obtained in S6, and use the RMSprop optimization algorithm to update the parameters of the foil interference suppression network based on the two-stage UNet with the optimal hyperparameters, so that the value of the loss function of the foil interference suppression network based on the two-stage UNet constructed in S5 is minimized, and the optimal foil interference suppression network based on the two-stage UNet is obtained.

[0104] S8, inputting the grayscale image of the mixed range image sequence of the target and the chaff interference obtained in S2 into the optimal chaff interference suppression network based on the two-stage UNet obtained in S7, to obtain a segmentation result and a suppression result;

[0105] So far, from S1 to S8, a shielded chaff interference suppression method based on a two-stage UNet has been completed.

[0106] In the specific implementation, the following three experiments were designed based on the target echoes under diluted chaff interference and the target echoes under shielded chaff interference collected by broadband polarimetric radar, as well as pseudo-measured data constructed based on measured data. Since the shielded chaff interference suppression method based on a two-stage UNet described in the present invention includes two stages: target foreground and background segmentation and chaff interference suppression, in order to comprehensively evaluate the performance of the chaff interference suppression network based on the two-stage UNet, it is necessary to measure both the segmentation effect and the suppression effect. The segmentation part uses the most commonly used indicator for evaluating medical image segmentation algorithms: the Dice coefficient, also known as the overlap index, which is used to calculate the overlap rate between the segmentation result area and the gold standard area, and its value range is between [0,1]. The indicators for evaluating the suppression effect include: PSNR (peak signal-to-noise ratio), SSIM (structural similarity) and IFSJR (average improvement factor of signal-to-interference ratio).

[0107] Experiment 1: To preliminarily verify the effectiveness of the proposed chaff interference suppression network, we constructed three pseudo-measured datasets with interference-to-signal ratios of 0dB, 13.98dB, and 20dB. These datasets were divided into training and test sets in a ratio of 7:3. This training set was used to train a chaff interference suppression network based on a two-stage UNet. The trained network was then tested on the test sets at the corresponding interference-to-signal ratios. The results are shown in Table 1.

[0108] Table 1 Effect of the chaff interference suppression network based on the two-stage UNet on the test set corresponding to the interference-to-signal ratio

[0109]

[0110] Table 1 shows that as the signal-to-interference ratio (SIR) increases, the performance of the dual-stage UNet-based chaff suppression network also declines to a certain extent, with PSNR dropping from 34.64dB to 23.01dB and SSIM dropping from 0.94 to 0.87. This indicates that the gap between the segmentation results and the ground-truth segmentation image, as well as the suppression results and the ground-truth target range image sequence, is widening. In terms of SIR gain, both stages of the dual-stage UNet-based chaff suppression network contribute to improving the SIR, with the average SIR improvement factor exceeding 23dB. This indicates that when the input JSR is 20dB or below, the dual-stage UNet-based chaff suppression network can successfully and reliably suppress shielded chaff interference, achieving good SIR improvement and target information recovery performance after successful suppression.

[0111] Figure 5The present invention is a method for suppressing shielded chaff interference based on a two-stage UNet, which includes grayscale images of a mixed range image sequence of pseudo-measured targets and chaff interference with different interference-to-signal ratios, segmentation results, suppression results, segmentation truth images, and target range image sequence truth images.

[0112] (a) is the 0dB input picture, (b) is the 0dB segmentation result, (c) is the 0dB segmentation label, (d) is the 0dB suppression result, (e) is the 0dB suppression label, (f) is the 13.98dB input picture, (g) is the 13.98dB segmentation result, (h) is the 13.98dB segmentation label, (i) is the 13.98dB suppression result, (j) is the 13.98dB suppression label, (k) is the 20dB input picture, (l) is the 20dB segmentation result, (m) is the 20dB segmentation label, (n) is the 20dB suppression result, and (o) is the 20dB suppression label.

[0113] from Figure 5 As can be seen, the suppression results after processing by the dual-stage UNet-based chaff interference suppression network closely match the labeled image, demonstrating that the proposed dual-stage UNet-based shielded chaff interference suppression method effectively suppresses chaff interference in mixed echoes. After the network processes mixed data containing radar targets and chaff clouds, the overall power level of the chaff interference is significantly suppressed, and the radar target signal is significantly enhanced.

[0114] Experiment 2: Because measured data cannot provide ground truth maps for network training, we constructed a training set of pseudo-measured data containing a large number of samples with a randomly distributed interference-to-signal ratio between 0 and 20 dB. This data was then fed into a two-stage UNet for multiple rounds of training. The measured data was then fed into a chaff interference suppression network based on this trained two-stage UNet.

[0115] Figure 6 The invention discloses a method for suppressing shielded chaff interference based on a dual-stage UNet, comprising a grayscale image sequence of a target and chaff interference mixture (6a), a measured data segmentation result (6b), and a measured data suppression result (6c).

[0116] from Figure 6 It can be seen from the figure that the target image obtained by processing the measured data through the chaff interference suppression network based on the two-stage UNet is very close to the real target information, which shows that the shielded chaff interference suppression method based on the two-stage UNet of the present invention can well suppress the chaff interference in the mixed echo, and the overall power level of the chaff interference is significantly suppressed, and the radar target signal is significantly enhanced.

[0117] The above description is only a preferred embodiment of the present invention, and the present invention should not be limited to the contents disclosed in the embodiment and the accompanying drawings. Any equivalent or modification completed without departing from the spirit disclosed in the present invention shall fall within the scope of protection of the present invention.

Claims

1. A shielded chaff interference suppression method based on a two-stage UNet, characterized in that: The steps include: S1. Using a broadband polarimetric radar, collecting target echoes under diluted chaff interference and target echoes under shielded chaff interference, and recording the position, angle, and velocity information of the target during data collection, to obtain target echoes under diluted chaff interference and target echoes under shielded chaff interference with real information; S2, pulse compressing and normalizing the target echo under the diluted chaff interference with real information obtained in S1 to obtain a target range image sequence grayscale image and a chaff interference range image sequence grayscale image; pulse compressing and normalizing the target echo under the masked chaff interference with real information obtained in S1 to obtain a mixed range image sequence grayscale image of the target and the chaff interference; S3, using the target range image sequence grayscale image and the chaff interference range image sequence grayscale image obtained in S2 to construct pseudo measured data, and obtaining a target range image sequence true value image, a chaff interference range image sequence true value image, a pseudo measured target and chaff interference mixed range image sequence grayscale image, and a segmentation true value image; S4. Construct a chaff interference suppression network based on a two-stage UNet; The chaff interference suppression network based on the dual-stage UNet is composed of two UNet cascades, namely the first-stage UNet and the second-stage UNet; During training or verification, the input image of the chaff interference suppression network based on the dual-stage UNet is a grayscale image of a range image sequence of a mixture of a pseudo-measured target and chaff interference; during testing, the input image of the chaff interference suppression network based on the dual-stage UNet is a grayscale image of a range image sequence of a mixture of a target and chaff interference; In the first stage, UNet segments the input image into target foreground and background to obtain the segmentation result; The label of the segmentation result, i.e. the segmentation label, is the segmentation truth map; The second-stage UNet uses the segmentation result of the first-stage UNet and the result of the dot product of the input image of the first-stage UNet as the input image of the second-stage UNet and processes it to obtain a suppression result, where the suppression result is the result of restoring the target amplitude information in the input image of the second-stage UNet; The label of the suppression result, i.e., the suppression label, is a true value image of the target range image sequence; The input image, segmentation result, segmentation label, suppression result, and suppression label of the two-stage UNet-based chaff interference suppression network are of the same size; S5. Construct the loss function of the chaff interference suppression network based on the two-stage UNet; The calculation formula of the loss function described in S5 is: (1) in, They are the cross entropy loss and Dice loss used by UNet in the first stage and the smooth L1 loss used by UNet in the second stage; is the weight of cross entropy loss; is the weight of Dice loss; is the weight of smooth L1 loss; The cross entropy loss used by UNet in the first stage is defined as: (2) N is the number of pixels in a batch of input images, is the pixel number; M represents the number of pixel categories. In the chaff interference suppression scenario, there are only two categories: target foreground and background, so M=2. is the category number, Indicates that the category is target foreground, Indicates that the category is background; is the pixel of the input image The category label, Represents the pixel points of the input image Not in category , Represents the pixel points of the input image Belong to category ; Represents the pixel point of the predicted input image Belong to category probability; The calculation formula of the Dice loss is: (3) Among them, P represents the segmentation result, G represents the segmentation label, It is the sum of the segmentation result and the pixel multiplication of the segmentation label. and The sum of all elements of the segmentation label and the segmentation result respectively; the range of the Dice loss is Between is the probability of the non-overlapping part between the segmentation result area and the segmentation label area; The second stage UNet uses smooth L1 loss is the mean of the absolute error of the pixels; S6. Model tuning: Grid search is used to obtain the optimal hyperparameters for the two-stage UNet-based chaff suppression network. S7, network training: using the RMSprop optimization algorithm to update the parameters of the dual-stage UNet-based chaff interference suppression network with the optimal hyperparameters obtained in S6, so that the loss function of the dual-stage UNet-based chaff interference suppression network constructed in S5 is minimized, thereby obtaining the optimal dual-stage UNet-based chaff interference suppression network; S8, network test: Input the grayscale image of the mixed range image sequence of the target and the chaff interference obtained in S2 into the optimal chaff interference suppression network based on the two-stage UNet obtained in S7 for testing to obtain the segmentation result and suppression result.

2. The shielded chaff interference suppression method according to claim 1, wherein: S2: The target range image sequence grayscale image, the chaff interference range image sequence grayscale image, and the target and chaff interference mixed range image sequence grayscale image all include X range resolution units, where X is the number of sampling points.

3. The shielded chaff interference suppression method according to claim 1, wherein: S3 specifically includes the following sub-steps: S31, randomly selecting P slow time series of range resolution units from the target range image sequence grayscale image obtained in S2, dividing them by the maximum value of the series for normalization, and reconstructing the target range image sequence ground truth image according to the spatial distribution and scattering intensity of the target scattering points; S32, randomly selecting Q slow time series of range resolution units from the grayscale image of the chaff interference range image sequence obtained in S2, dividing the value in the slow time series by the maximum value of the series to perform normalization, and then multiplying it by the weight coefficient of the normal distribution curve to obtain a true value image of the chaff interference range image sequence; S33, performing weighted summation on the target range image sequence truth image obtained in S31 and the chaff interference range image sequence truth image obtained in S32 to obtain a grayscale image of a range image sequence of a mixture of pseudo-measured target and chaff interference; S34 , subjecting the target range image sequence truth map obtained in S31 to CA-CFAR detection to obtain a segmentation truth map.

4. The shielded chaff interference suppression method according to claim 3, characterized in that: P in S31 is less than X; Q in S32 is less than X; when performing weighted summation in S33, the ratio of weights is the signal-to-interference ratio; the value of a pixel point in the segmentation truth map in S34 is 1, indicating that the value of the corresponding pixel point in the target range image sequence truth map is greater than the CA-CFAR detector threshold; and the value of 0 indicates that the value of the corresponding pixel point in the target range image sequence truth map is less than or equal to the CA-CFAR detector threshold.

5. The shielded chaff interference suppression method according to claim 1, characterized in that: S4 Both the first-stage UNet and the second-stage UNet include one input layer, three feature extraction layers, three feature enhancement layers and one output layer; the input layer includes two convolutional layers with a kernel size of 3×3.

6. The shielded chaff interference suppression method according to claim 5, characterized in that: The feature extraction layer includes a pooling layer with a kernel size of 2×2 and two convolution layers with a kernel size of 3×3; the feature enhancement layer includes a deconvolution layer with a kernel size of 2×2 and two convolution layers with a kernel size of 3×3; the output layer is a convolution layer with a kernel size of 1×1 and an output channel number of 2 or 1; the number of channels of the output layer of the first-stage UNet is 2, and the number of output channels of the second-stage UNet is 1.

7. The shielded chaff interference suppression method according to claim 1, characterized in that: The absolute error of the pixel point S5 is Indicates that, and and are the pixels of the suppression results and suppression labels respectively, where The value range is .

8. The shielded chaff interference suppression method according to claim 1, characterized in that: During model tuning and network training in S6 and S7, the input image, segmentation label and suppression label of the chaff interference suppression network based on the two-stage UNet are respectively the grayscale image of the range image sequence of the pseudo-measured target and chaff interference mixture obtained in S3, the segmentation true value image and the target range image sequence true value image.

Images