Satellite-borne SAR working mode recognition method based on noise reduction recognition network

Abstract

The invention provides a satellite-borne SAR working mode recognition method based on a noise reduction recognition network. According to the method, reconnaissance stations are arranged in different layout modes, reconnaissance waveforms of a spaceborne SAR in a strip mode, a bunching mode, a sliding bunching mode and a scanning mode are obtained, and reconnaissance station receiving signals corresponding to different working modes are calculated. And selecting the layout mode with the maximum feature difference as the optimal layout mode of the reconnaissance station. And simulating each working mode based on the layout mode, acquiring a noise signal in a simulation environment, and taking a pure signal obtained by calculation as a part of a training set and a test set. And constructing a noise reduction recognition neural network model, training the model through the training set, and performing satellite-borne SAR working mode recognition by using the trained noise reduction recognition neural network model. According to the method, the recognition accuracy of the working mode of the satellite-borne SAR in the noise environment is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of radar communication technology, and in particular to a method for identifying a spaceborne SAR working mode based on a noise reduction recognition network. Background Art

[0002] Spaceborne synthetic aperture radar (SAR), an active microwave imaging sensor, offers all-day, all-weather imaging and observation capabilities and is widely used in military reconnaissance, target identification, and environmental monitoring. However, with the continuous advancement of radar technology, the operating modes of spaceborne SAR have become increasingly complex, including strip, spotlight, and scanning modes, each with significant differences in signal characteristics. Therefore, quickly and accurately identifying the operating mode of spaceborne SAR is crucial for improving reconnaissance efficiency and operational decision-making.

[0003] Currently, methods for identifying the operating modes of spaceborne SAR are primarily divided into traditional signal analysis methods and deep learning-based approaches. Traditional methods typically rely on extracting features such as SAR signal range parameters, Doppler parameters, and received power patterns, and then combining them with orbital parameters for mode discrimination. However, these methods suffer from low recognition accuracy, poor timeliness, and a limited range of identifiable modes.

[0004] In recent years, deep learning technology has made significant progress in target recognition and has also been introduced into spaceborne SAR operating mode recognition. For example, patent application number 201911299741.3 proposes a phased array radar operating mode recognition method based on a multi-layer perceptron (MLP), which achieves pattern recognition through feature extraction and encoding. However, this method is primarily targeted at phased array radars and cannot be directly applied to spaceborne SAR due to significant differences in signal form and operating procedures.

[0005] Existing technologies for spaceborne SAR operating mode recognition still suffer from the following technical deficiencies: Traditional methods rely on manual feature extraction, resulting in low recognition accuracy and timeliness. Deep learning-based methods lack robustness in complex environments, particularly under noisy conditions, with recognition performance significantly degraded. Existing methods fail to fully utilize the multidimensional characteristics of SAR signals, resulting in limited recognition performance. Summary of the Invention

[0006] The present invention aims to solve at least one of the above-mentioned technical problems existing in the prior art.

[0007] To this end, the present invention provides a method for identifying the working mode of a spaceborne SAR based on a noise reduction recognition network.

[0008] The present invention provides a method for identifying a spaceborne SAR working mode based on a noise reduction recognition network, comprising:

[0009] Arrange a number of reconnaissance stations in different layouts, obtain reconnaissance waveforms of different operating modes of the spaceborne SAR in each layout, and calculate the reconnaissance station received signals corresponding to the different operating modes; the operating modes include strip mode, spotlight mode, sliding spotlight mode, and scanning mode;

[0010] The characteristic differences of the reconnaissance waveforms of different working modes of spaceborne SAR obtained under each layout are compared, and the layout with the largest characteristic difference is selected as the optimal layout of the reconnaissance station;

[0011] Based on the determined optimal layout of the reconnaissance stations, each working mode is simulated to obtain the reconnaissance station receiving signal in the simulation environment;

[0012] The calculated received signals of the reconnaissance station corresponding to different working modes are defined as pure signals; the received signals of the reconnaissance station obtained in the simulation environment are defined as noisy signals, and a training set and a test set are constructed; the training set includes the noisy signal, the pure signal and the label sequence, and the test set includes the noisy signal and the label sequence; the label sequence is used to at least mark the working mode corresponding to the signal;

[0013] Constructing a noise reduction and recognition neural network model, wherein the input of the noise reduction and recognition neural network model is a noisy signal, and the output is the spaceborne SAR operating mode corresponding to the noisy signal; the noise reduction and recognition neural network model includes a noise reduction network and a recognition network connected in series, wherein the noise reduction network is used to denoise the input signal, and the recognition network is used to identify the operating mode based on the denoised input signal;

[0014] The constructed training set is used to train the denoising recognition neural network model; the test set is used to evaluate the training results of the denoising recognition neural network model.

[0015] The method for identifying the operating mode of a spaceborne SAR based on a noise reduction recognition network according to the above technical solution of the present invention may also have the following additional technical features:

[0016] In the above technical solution, the layout mode includes at least two of the first layout mode, the second layout mode and the third layout mode;

[0017] In the first layout mode, all reconnaissance stations are arranged within the illumination range of the radar antenna main lobe;

[0018] In the second layout mode, all reconnaissance stations are arranged at the edge of the radar antenna main lobe illumination range;

[0019] In the third layout mode, all reconnaissance stations are arranged between the edge of the radar antenna main lobe illumination range and the antenna side lobe illumination range.

[0020] In the above technical solution, it is assumed that the satellite motion trajectory is the same in each working mode, and all fly along the y-axis; the calculation of the reconnaissance station receiving signal corresponding to different working modes includes: calculating the receiving signal of the reconnaissance station set at (Y, R) on the earth's surface under different working modes when the satellite moves to (0, y); the calculation result of the received signal includes the signal power received by the reconnaissance station and the received echo expression.

[0021] In the above technical solution, in the stripe mode, the calculation method of the signal received by the reconnaissance station includes:

[0022] Calculate antenna patterns at different times and orientations;

[0023] Calculate the echo power density received at the stripe pattern reconnaissance station location based on the antenna pattern;

[0024] The signal power received at the stripe pattern reconnaissance station is calculated based on the echo power density, and its expression is:

[0025]

[0026] Among them, P r1 represents the signal power received at the position of the stripe pattern reconnaissance station; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment;

[0027] Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the stripe mode reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including:

[0028]

[0029] Among them, s1(y,t r ; R) represents the echo expression received at the stripe mode reconnaissance station position; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

[0030] In the above technical solution, in the beamforming mode, the calculation method of the signal received by the reconnaissance station includes:

[0031] Calculate antenna patterns at different times and orientations;

[0032] Calculate the echo power density received at the reconnaissance station in the spotlight mode based on the antenna pattern;

[0033] The signal power received at the spotlight mode reconnaissance station is calculated based on the echo power density, and its expression is:

[0034]

[0035] Among them, P r2 represents the signal power received at the reconnaissance station in the spotlight mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment;

[0036] Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the spotlight mode reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including:

[0037]

[0038] Among them, s2(y,t r ; R) represents the echo expression received by the spotlight mode reconnaissance station; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

[0039] In the above technical solution, in the sliding beamforming mode, the calculation method of the signal received by the reconnaissance station includes:

[0040] Calculate antenna patterns at different times and orientations;

[0041] Calculate the echo power density received at the reconnaissance station position in sliding beam mode based on the antenna pattern;

[0042] The signal power received at the reconnaissance station in the sliding beam mode is calculated based on the echo power density, and its expression is:

[0043]

[0044] Among them, P r3represents the signal power received at the position of the reconnaissance station in sliding beam mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment;

[0045] Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the position of the reconnaissance station in the sliding beam mode based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including:

[0046]

[0047] Among them, s2(y,t r ; R) represents the echo expression received by the reconnaissance station in the spotlight mode; represents the antenna pattern; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiated to the ground; c represents the speed of light; λ represents the wavelength; v b represents the beam movement speed; v represents the reference speed, that is, the satellite flight speed.

[0048] In the above technical solution, in scanning mode, the reconnaissance station antenna only rotates upward in the distance and stays for a set time after each rotation. During this time, the reconnaissance station antenna is fixed in direction, which is equivalent to the strip mode and contains the information of one wave position. This time interval is defined as the sub-strip mode. The number of sub-strips in the scanning mode corresponds to the number of wave position information. Each scan is from the first wave position to the last wave position.

[0049] In scanning mode, the calculation method of the signal received by the reconnaissance station includes:

[0050] Calculate antenna patterns at different times and orientations;

[0051] Calculate the echo power density received at the reconnaissance station position in scanning mode based on the antenna pattern;

[0052] The signal power received at the scanning mode reconnaissance station position is calculated based on the echo power density, and its expression is:

[0053]

[0054] Among them, P r1represents the signal power received at the reconnaissance station position in scanning mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment;

[0055] Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the scanning mode reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including:

[0056]

[0057] Among them, s4(y,t r ; R) represents the echo expression received by the scanning mode reconnaissance station position; s 1m (y,t r ; R) represents the echo expression of the mth sub-strip; M represents the number of sub-strips in the scanning mode; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

[0058] In the above technical solution, the denoising network adopts a BiLSTM module, which includes a cascaded input layer, a bidirectional LSTM layer and a first fully connected layer; the bidirectional LSTM layer includes a hidden layer and an implicit layer;

[0059] The recognition network adopts a RESFFN module, which includes a cascaded second fully connected layer, a first activation function layer, a first batch of normalized layers, a third fully connected layer, a second activation function layer, a second batch of normalized layers, five residual blocks, a fourth fully connected layer, a third activation function layer, a third batch of normalized layers, a fifth fully connected layer, a fourth activation function layer, a fourth batch of normalized layers, a sixth fully connected layer, and a fifth activation function layer;

[0060] The denoising and recognition neural network formed by cascading the denoising network and the recognition network is a BiLSTM-RESFFN network;

[0061] Set the root mean square loss function of the BiLSTM-RESFFN network MSE1 for:

[0062]

[0063] Among them, f1(x i ) represents the output signal of the BiLSTM-RESFFN network after denoising the i-th noise sample, x i represents the clean signal corresponding to the i-th noise sample, i∈[1,n], and n is the total number of samples in the training set; the difference between the denoised signal output by the denoising part of the BiLSTM-RESFFN network and its corresponding clean signal is calculated using the root mean square loss function;

[0064] Set the cross entropy loss function of the BiLSTM-RESFFN network CEL1 for:

[0065]

[0066] Among them, g1(g i ) represents the label of the i-th sample after recognition in the BiLSTM-RESFFN network; g i Represents the true label of the i-th sample; the difference between the recognition label output by the BiLSTM-RESFFN network recognition part and its corresponding true label is calculated through the cross entropy loss function;

[0067] Based on the root mean square loss function and the cross entropy loss function, the parameters of the BiLSTM-RESFFN network are iteratively updated by the gradient descent method to obtain the trained model.

[0068] In the above technical solution, the denoising network adopts a CAE module, which includes a cascaded input layer, a first convolutional layer, a first activation function layer, a first pooling layer, a second convolutional layer, a second activation function layer, a third convolutional layer, a first transposed convolutional layer, a third activation function layer, a fourth convolutional layer, a fourth activation function layer, a fifth convolutional layer and a fifth activation function layer;

[0069] The recognition network adopts a RESFFN module, which includes a cascaded first fully connected layer, a sixth activation function layer, a first batch normalization layer, a second fully connected layer, a seventh activation function layer, a second batch normalization layer, five residual blocks, a third fully connected layer, an eighth activation function layer, a third batch normalization layer, a fourth fully connected layer, a ninth activation function layer, a fourth batch normalization layer, a fifth fully connected layer, and a tenth activation function layer;

[0070] The denoising and recognition neural network formed by cascading the denoising network and the recognition network is the CAE-RESFFN network;

[0071] Set the root mean square loss function of the CAE-RESFFN network MSE2 for:

[0072]

[0073] Among them, f2(x i ) is the output signal of the denoised i-th noise sample of the CAE-RESFFN network, x i is the clean signal corresponding to the i-th noise sample, i∈[1,n], n is the total number of samples in the training set; the difference between the denoised signal output by the denoising part of the CAE-RESFFN network and its corresponding clean signal is calculated using the root mean square loss function;

[0074]

[0075] Among them, g2(g i ) represents the label of the i-th sample after recognition in the CAE-RESFFN network, g i represents the true label of the i-th sample; the difference between the recognition label output by the CAE-RESFFN network recognition part and its corresponding true label is calculated through the cross entropy loss function;

[0076] Based on the root mean square loss function and the cross entropy loss function, the parameters of the CAE-RESFFN network are iteratively updated by the gradient descent method to obtain the trained model.

[0077] In the above technical solution, before using the constructed training set to train the denoising recognition neural network model or using the test set to evaluate the training results of the denoising recognition neural network model, the following steps are also included:

[0078] Perform centralization and standardization operations on samples of the dataset;

[0079] The centering operation includes calculating the mean and standard deviation of each sample sequence, then subtracting the mean from each value of the sequence and dividing it by the standard deviation of the sequence;

[0080] The normalization operation includes calculating the maximum and minimum values of each sample sequence, then subtracting the minimum value from each value in the sequence, and then dividing by the difference between the maximum and minimum values.

[0081] In summary, due to the adoption of the above technical features, the beneficial effects of the present invention are:

[0082] The present invention provides a method for identifying the operating mode of spaceborne SAR (SAR) reconnaissance signals based on a noise reduction recognition network, effectively improving the accuracy of identifying the operating mode of spaceborne SAR reconnaissance signals in noisy environments. Specifically, the present invention analyzes the characteristics of spaceborne SAR signals acquired from different reconnaissance station layouts to determine the optimal reconnaissance station layout. This layout demonstrates significant differences in the waveform characteristics of different operating modes, thereby improving the accuracy of identifying the operating mode of spaceborne SAR reconnaissance signals in noisy environments.

[0083] In view of the fact that existing deep learning algorithms cannot extract data features of spaceborne SAR signals in noisy environments, the present invention builds two noise reduction and recognition neural networks to remove signal noise from the two aspects of long-term and short-term memory and encoding and decoding respectively. This enables the network to have the ability to process spaceborne SAR signals intercepted by reconnaissance aircraft in low signal-to-noise ratio environments, further improving the accuracy of working mode recognition of spaceborne SAR reconnaissance signals in noisy environments.

[0084] Additional aspects and advantages of the invention will become apparent from the description which follows, or may be learned by practice of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0086] Figure 1 This is a flow chart of a method for identifying a spaceborne SAR working mode based on a noise reduction recognition network according to an embodiment of the present invention;

[0087] Figure 2 This is a schematic diagram of the arrangement of a reconnaissance station in one embodiment of the present invention;

[0088] Figure 3 is a schematic diagram of a stripe pattern reconnaissance model according to one embodiment of the present invention;

[0089] Figure 4 is a schematic diagram of a data set constructed in one embodiment of the present invention;

[0090] Figure 5 Schematic diagram of the noise reduction recognition neural network structure constructed in one embodiment of the present invention;

[0091] Figure 6 This is a statistical graph of the recognition results of the test set by the trained BiLSTM-RESFFN network model in one embodiment of the present invention;

[0092] Figure 7 This is a statistical graph of the recognition results of the test set by the trained CAE-RESFFN network model in one embodiment of the present invention. DETAILED DESCRIPTION

[0093] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0094] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0095] Refer to the following Figures 1 to 7 The following describes a method for identifying a spaceborne SAR working mode based on a noise reduction recognition network according to some embodiments of the present invention.

[0096] Some embodiments of the present application provide a method for identifying operating modes of spaceborne SAR based on a noise reduction recognition network.

[0097] like Figure 1 As shown, the first embodiment of the present invention provides a method for identifying the operating mode of a spaceborne SAR based on a noise reduction recognition network, including the following steps S1-S6. The order of steps S1-S6 is merely illustrative, and those skilled in the art may adjust the order of the steps as needed. Steps in different orders may also be executed simultaneously.

[0098] S1. Arrange several reconnaissance stations in different layouts, obtain reconnaissance waveforms of different operating modes of spaceborne SAR in each layout, and calculate the reconnaissance station received signals corresponding to different operating modes; the operating modes include strip mode, spotlight mode, sliding spotlight mode, and scanning mode.

[0099] There are usually multiple reconnaissance stations, and this disclosure uses an example where the number of reconnaissance stations is 5 for illustration.

[0100] In some embodiments, the layout mode includes at least two of a first layout mode, a second layout mode, and a third layout mode; Figure 2 In the illustrated embodiment, the first layout mode is simply indicated as Mode 1 , the second layout mode is simply indicated as Mode 2 , and the third layout mode is simply indicated as Mode 3 .

[0101] In the first layout mode, all reconnaissance stations are arranged within the main lobe illumination range of the radar antenna; in the second layout mode, all reconnaissance stations are arranged at the edge of the main lobe illumination range of the radar antenna; in the third layout mode, all reconnaissance stations are arranged between the edge of the main lobe illumination range of the radar antenna and the side lobe illumination range of the antenna.

[0102] It is assumed that the satellite motion trajectory is the same in each working mode, and all of them fly along the y-axis; the calculation of the reconnaissance station receiving signal corresponding to different working modes includes: calculating the receiving signal of the reconnaissance station set at (Y, R) on the earth's surface under different working modes when the satellite moves to (0, y); the calculation result of the receiving signal includes the signal power received by the reconnaissance station and the received echo expression.

[0103] When the spaceborne SAR operates in stripe mode, the radar always illuminates in the positive side-viewing direction, and the antenna pointing direction remains unchanged, and only one wave position information is included. Based on this, in some embodiments, the calculation method of the signal received by the reconnaissance station in stripe mode includes:

[0104] Calculate the antenna pattern at different times and in different azimuths. The antenna pattern used in this disclosure is a three-dimensional antenna pattern, which can be viewed as the product of the azimuth and elevation patterns, i.e.:

[0105]

[0106] Among them, F θ (θ) represents the two-dimensional pattern function of the azimuth plane, Represents the two-dimensional pattern function of the pitch plane.

[0107] Calculate the echo power density received at the stripe pattern reconnaissance station location based on the antenna pattern; including:

[0108]

[0109] Among them, P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment;

[0110] The signal power received at the stripe pattern reconnaissance station is calculated based on the echo power density, and its expression is:

[0111]

[0112] Among them, P r1 represents the signal power received at the stripe pattern reconnaissance station; γ represents the polarization coefficient, corresponding to the polarization mode of the transmitting antenna and the receiving antenna; It represents the feeder coefficient, which represents the loss when the signal enters the receiving antenna; S represents the echo power density; A r Represents the effective area of the receiving antenna;

[0113] Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the stripe mode reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including:

[0114]

[0115] Among them, s1(y,t r ; R) represents the echo expression received at the stripe mode reconnaissance station position; rect(·) represents the rectangular window function; t rrepresents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

[0116] In a specific embodiment, the method for calculating the instantaneous slant distance includes:

[0117]

[0118] Where H represents the platform height, that is, the satellite flight altitude;

[0119] The calculation method of FM slope includes:

[0120] K r =B / T r

[0121] Where B represents the signal bandwidth, T r Indicates pulse width;

[0122] The calculation method of the width of the beam irradiated to the ground includes:

[0123] L B =λH / D

[0124] Where D represents the antenna size.

[0125] When the spaceborne SAR operates in the spotlight mode, the antenna pointing direction rotates according to a certain pattern. Each time it switches to a different viewing angle in azimuth, there is a short dwell time. The center of the antenna beam generally always points in the direction of the imaging center point. In some embodiments, in the spotlight mode, the calculation method of the signal received by the reconnaissance station includes:

[0126] Calculate antenna patterns at different times and orientations;

[0127] Calculate the echo power density received at the reconnaissance station in the spotlight mode based on the antenna pattern;

[0128] The signal power received at the spotlight mode reconnaissance station is calculated based on the echo power density, and its expression is:

[0129]

[0130] Among them, P r2 represents the signal power received at the reconnaissance station in the spotlight mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; ta Indicates the current moment;

[0131] Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the spotlight mode reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including:

[0132]

[0133] Among them, s2(y,t r ; R) represents the echo expression received by the spotlight mode reconnaissance station; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

[0134] When the spaceborne SAR operates in the sliding spotlight mode, the antenna pointing direction also rotates in azimuth according to a certain pattern. That is, for different viewing angles, the wave position has a short dwell time, which is slightly longer than that in the spotlight mode. The center of the antenna beam is generally always pointed in the direction of the virtual imaging center point. In some embodiments, in the sliding spotlight mode, the calculation method of the reconnaissance station receiving signal includes:

[0135] Calculate antenna patterns at different times and orientations;

[0136] Calculate the echo power density received at the reconnaissance station position in sliding beam mode based on the antenna pattern;

[0137] The signal power received at the reconnaissance station in the sliding beam mode is calculated based on the echo power density, and its expression is:

[0138]

[0139] Among them, P r3 represents the signal power received at the position of the reconnaissance station in sliding beam mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment;

[0140] Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the position of the reconnaissance station in the sliding beam mode based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including:

[0141]

[0142] Among them, s2(y,t r ; R) represents the echo expression received by the reconnaissance station in the spotlight mode; represents the antenna pattern; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiated to the ground; c represents the speed of light; λ represents the wavelength; v b represents the beam movement speed; v represents the reference speed, that is, the satellite flight speed.

[0143] When the spaceborne SAR is in scanning mode, the antenna of the reconnaissance station only rotates in the range-up direction, and there is a dwell time after each rotation. During this time, the antenna's direction is fixed, which can be regarded as a strip mode and also contains the information of a wave position. This time interval is defined as a sub-strip mode. The scanning mode has several sub-strips, which means there are several wave position information. Each scan is from the first wave position to the last wave position. After the scan is completed, it returns to the starting wave position and scans again. In some embodiments, in scanning mode, the calculation method of the signal received by the reconnaissance station includes:

[0144] Calculate antenna patterns at different times and orientations;

[0145] Calculate the echo power density received at the reconnaissance station position in scanning mode based on the antenna pattern;

[0146] The signal power received at the scanning mode reconnaissance station position is calculated based on the echo power density, and its expression is:

[0147]

[0148] Among them, P r1 represents the signal power received at the reconnaissance station position in scanning mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment;

[0149] Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the sub-strip reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground. Add the echo expressions received at each sub-strip reconnaissance station position to obtain the echo expression received at the scanning mode reconnaissance station position, which is:

[0150]

[0151] Among them, s4(y,t r ; R) represents the echo expression received by the scanning mode reconnaissance station position; s 1m (y,t r ; R) represents the echo expression of the mth sub-strip; M represents the number of sub-strips in the scanning mode; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

[0152] Based on the signal power P received by the reconnaissance station in different working modes r and the echo expression s(y,t r ; R), the reconnaissance signals corresponding to the four working modes can be simulated, that is, pure signals.

[0153] It should be noted that the parameters in the above-mentioned spotlight mode, sliding spotlight mode and scanning mode for which specific calculation methods are not specified can be calculated by referring to the specific calculation methods given in the stripe mode.

[0154] S2. Compare the characteristic differences of the reconnaissance waveforms of different working modes of the spaceborne SAR obtained under each layout mode, and select the layout mode with the largest characteristic difference as the optimal layout mode for the reconnaissance station.

[0155] It should be noted that in order to obtain the reconnaissance waveforms of different working modes of spaceborne SAR, it is necessary to build a reconnaissance model of the reconnaissance station for spaceborne SAR signals:

[0156] like Figure 3 As shown, the present disclosure takes the construction of a reconnaissance model of a spaceborne SAR signal operating in strip mode as an example, and the construction method is as follows:

[0157] Set point O as the center of the earth, point E as the imaging center reference point, the satellite's flight speed along the orbit is v, and the radar always illuminates in the positive side view direction;

[0158] The direction passing through point E and perpendicular to the earth's surface is defined as the z-axis direction, the direction of the satellite platform when it is flying directly overhead is defined as the y-axis direction, and the x-axis direction is determined by the right-hand coordinate system principle to establish a spatial rectangular coordinate system.

[0159] R e is the radius of the earth, H is the flying altitude of the satellite, and five reconnaissance stations A, B, C, D, and E are set on the surface of the earth.

[0160] Based on the constructed reconnaissance models, the satellite motion and antenna illumination patterns of different operating modes were simulated to obtain the signals received by the reconnaissance stations under three layouts: Layout 1, Layout 2, and Layout 3. The optimal layout for the reconnaissance stations was selected as the one that resulted in the most distinct signal characteristics. Signal characteristics can be distinguished using existing methods such as similarity analysis, statistical analysis, Fourier transform (FFT), or wavelet transform, which will not be discussed here. The optimal layout selected for this example was Layout 2, which places the five reconnaissance stations at the edges of the antenna main lobe illumination range.

[0161] S3. Simulate each working mode based on the determined optimal layout of the reconnaissance station to obtain the reconnaissance station receiving signal in the simulation environment.

[0162] Specifically, step S3 is a process of generating sample data, and a large amount of data should be generated for different working modes.

[0163] In a specific embodiment, the signal-to-noise ratio of the strip mode and the scanning mode is set within the range of 0dB to 20dB; the signal-to-noise ratio of the beam mode and the sliding beam mode is set within the range of 20dB to 40dB; the optimal layout of the reconnaissance station is used to simulate the received signals of the reconnaissance station in the strip mode, beam mode, sliding beam mode and scanning mode multiple times, and the radar parameters and motion platform parameters are changed in each simulation to generate a data set.

[0164] S4. Define the calculated received signals of the reconnaissance station corresponding to different working modes as pure signals; define the received signals of the reconnaissance station obtained in the simulation environment as noisy signals, and construct training sets and test sets; the training set includes noisy signals, pure signals and label sequences, and the test set includes noisy signals and label sequences; the label sequence is at least used to mark the working mode corresponding to the signal.

[0165] In a specific embodiment, the data set is divided into two parts in a ratio of 5:3 and respectively labeled differently to obtain a training set and a test set. The training set includes a noisy signal, a clean signal and a label sequence; the test set includes a noisy signal and a label sequence.

[0166] like Figure 4 As shown in the figure, the parameters of the noisy signal, clean signal and label sequence of the training set are set as follows:

[0167] The noisy signal contains 375 samples of different types at each signal-to-noise ratio, for a total of 7,500 samples. The clean signal consists of 375 samples corresponding to each operating mode, for a total of 1,500 samples. The label sequence contains two types of labels: the operating mode label and the type label. The operating mode labels include: label 0 for striping mode, label 1 for scanning mode, label 2 for beamforming mode, and label 3 for sliding beamforming mode. The type label includes the type distinction between noisy and clean signals.

[0168] The parameters of the noisy signal and label sequence of the test set are set as follows:

[0169] Each signal-to-noise ratio of the noisy signal contains 375 different types of samples, for a total of 3000 samples. The label sequence only contains labels for the operating mode, which include: striping mode corresponds to label 0, scanning mode corresponds to label 1, beaming mode corresponds to label 2, and sliding beaming mode corresponds to label 3.

[0170] S5. Construct a noise reduction and recognition neural network model, wherein the input of the noise reduction and recognition neural network model is a noisy signal, and the output is the spaceborne SAR working mode corresponding to the noisy signal; the noise reduction and recognition neural network model includes a noise reduction network and a recognition network connected in series, the noise reduction network is used to denoise the input signal, and the recognition network is used to identify the working mode according to the denoised input signal.

[0171] In some embodiments, as Figure 5 As shown, the denoising network adopts a BiLSTM module, which includes a cascaded input layer, a bidirectional LSTM layer and a first fully connected layer; the bidirectional LSTM layer includes a hidden layer and an implicit layer; in a specific embodiment, the input layer input is set to n samples, and the length of each sample is 1×1000; the bidirectional LSTM layer includes a hidden layer and an implicit layer, and the tensor dimension of the hidden layer is set to 64, the input dimension is 1, and the number of hidden layers is 2; the input dimension of the first fully connected layer is set to 128, and the output dimension is 1.

[0172] The recognition network adopts a RESFFN module, which includes a cascaded second fully connected layer, a first activation function layer, a first batch normalization layer, a third fully connected layer, a second activation function layer, a second batch normalization layer, five residual blocks, a fourth fully connected layer, a third activation function layer, a third batch normalization layer, a fifth fully connected layer, a fourth activation function layer, a fourth batch normalization layer, a sixth fully connected layer, and a fifth activation function layer. In a specific embodiment, its parameters are set as follows:

[0173] Set the input dimension of its second fully connected layer to 1000 and the output dimension to 500;

[0174] Set the output dimension of its first batch of normalization layers to 500;

[0175] Set the input dimension of its third fully connected layer to 500 and the output dimension to 200;

[0176] Set the output dimension of its second batch normalization layer to 200;

[0177] Set the input dimension of its fourth fully connected layer to 200 and the output dimension to 200;

[0178] Set the output dimension of its third batch normalization layer to 200;

[0179] Set the input dimension of its fifth fully connected layer to 200 and the output dimension to 100;

[0180] Set the output dimension of its fourth batch normalization layer to 100;

[0181] Set the input dimension of its sixth fully connected layer to 100 and the output dimension to 4;

[0182] Set each activation function layer to have a ReLU activation function, whose formula is as follows:

[0183] ReLU=max(x,0)

[0184] Among them, x is the input of the activation function.

[0185] The denoising and recognition neural network formed by the cascade of the denoising network and the recognition network is a BiLSTM-RESFFN network. Through the BiLSTM-RESFFN network composed of the long short-term memory module, the residual module, and the fully connected layer module in sequence, the accumulation speed of information can be controlled and information can be selectively added and forgotten to identify the working mode of the spaceborne SAR.

[0186] Set the root mean square loss function of the BiLSTM-RESFFN network MSE1 for:

[0187]

[0188] Among them, f1(x i ) represents the output signal of the BiLSTM-RESFFN network after denoising the i-th noise sample, x i represents the clean signal corresponding to the i-th noise sample, i∈[1,n], and n is the total number of samples in the training set; the difference between the denoised signal output by the denoising part of the BiLSTM-RESFFN network and its corresponding clean signal is calculated using the root mean square loss function;

[0189] Set the cross entropy loss function of the BiLSTM-RESFFN networkCEL1 for:

[0190]

[0191] Among them, g1(g i ) represents the label of the i-th sample after recognition in the BiLSTM-RESFFN network; g i Represents the true label of the i-th sample; the difference between the recognition label output by the BiLSTM-RESFFN network recognition part and its corresponding true label is calculated through the cross entropy loss function;

[0192] Based on the root mean square loss function and the cross entropy loss function, the parameters of the BiLSTM-RESFFN network are iteratively updated by the gradient descent method to obtain the trained model. The number of updates can be set to 100 times.

[0193] In other embodiments, the denoising network adopts a CAE module, which includes a cascaded input layer, a first convolutional layer, a first activation function layer, a first pooling layer, a second convolutional layer, a second activation function layer, a third convolutional layer, a first transposed convolutional layer, a third activation function layer, a fourth convolutional layer, a fourth activation function layer, a fifth convolutional layer, and a fifth activation function layer; in a specific embodiment, its parameters are set as follows:

[0194] Set the input of its input layer to n samples, each sample length is 1×1000;

[0195] Set the input dimension of its first convolutional layer to 1 and the output dimension to 64;

[0196] Set the size of its first pooling layer to 2;

[0197] Set the input dimension of its second convolutional layer to 64 and the output dimension to 128;

[0198] Set the input dimension of the first transposed convolutional layer to 128 and the output dimension to 128;

[0199] Set the input dimension of its third convolutional layer to 128 and the output dimension to 64;

[0200] Set the input dimension of its 4th convolutional layer to 64 and the output dimension to 64; set the input dimension of its 5th convolutional layer to 64 and the output dimension to 1.

[0201] The recognition network adopts the RESFFN module, which includes a cascaded first fully connected layer, a sixth activation function layer, a first batch normalization layer, a second fully connected layer, a seventh activation function layer, a second batch normalization layer, five residual blocks, a third fully connected layer, an eighth activation function layer, a third batch normalization layer, a fourth fully connected layer, a ninth activation function layer, a fourth batch normalization layer, a fifth fully connected layer, and a tenth activation function layer. In a specific embodiment, its parameters are set as follows:

[0202] Set the input dimension of its first fully connected layer to 1000 and the output dimension to 500;

[0203] Set the output dimension of its first batch normalization layer to 500;

[0204] Set the input dimension of its second fully connected layer to 500 and the output dimension to 200;

[0205] Set the output dimension of its second batch normalization layer to 200;

[0206] Set the input dimension of its third fully connected layer to 200 and the output dimension to 200;

[0207] Set the output dimension of its third batch normalization layer to 200;

[0208] Set the input dimension of its 4th fully connected layer to 200 and the output dimension to 100;

[0209] Set the output dimension of its 4th batch normalization layer to 100;

[0210] Set the input dimension of its fifth fully connected layer to 100 and the output dimension to 4;

[0211] Set each activation function layer to have a ReLU activation function, whose formula is as follows:

[0212] ReLU=max(x,0)

[0213] Among them, x is the input of the activation function.

[0214] It should be noted that the above layers are numbered with Arabic numerals to distinguish them from the BiLSTM-RESFFN network.

[0215] The denoising and recognition neural network formed by the cascade of the denoising network and the recognition network is the CAE-RESFFN network. The CAE-RESFFN network, which is composed of the encoding module, decoding module, residual module, and fully connected layer module cascaded in sequence, can filter out the high-frequency components or outliers brought by the noise, establish noise-free data for feature reconstruction, and identify the working mode of the spaceborne SAR.

[0216] Set the root mean square loss function of the CAE-RESFFN network MSE2 for:

[0217]

[0218] Among them, f2(x i ) is the output signal of the denoised i-th noise sample of the CAE-RESFFN network, x i is the clean signal corresponding to the i-th noise sample, i∈[1,n], n is the total number of samples in the training set; the difference between the denoised signal output by the denoising part of the CAE-RESFFN network and its corresponding clean signal is calculated using the root mean square loss function;

[0219]

[0220] Among them, g2(g i ) represents the label of the i-th sample after recognition in the CAE-RESFFN network, g i represents the true label of the i-th sample; the difference between the recognition label output by the CAE-RESFFN network recognition part and its corresponding true label is calculated through the cross entropy loss function;

[0221] Based on the root mean square loss function and the cross entropy loss function, the parameters of the CAE-RESFFN network are iteratively updated by the gradient descent method to obtain the trained model.

[0222] S6. Use the constructed training set to train the denoising recognition neural network model; use the test set to evaluate the training results of the denoising recognition neural network model.

[0223] During training, you need to set the learning rate and maximum number of iterations of the denoising recognition neural network model. For example, set the learning rate to 0.001 and the maximum number of iterations to 100.

[0224] In some embodiments, before using the constructed training set to train the denoising recognition neural network model or using the test set to evaluate the training results of the denoising recognition neural network model, the process further includes:

[0225] Perform centralization and standardization operations on samples of the dataset;

[0226] The centering operation includes calculating the mean and standard deviation of each sample sequence, then subtracting the mean from each value of the sequence and dividing it by the standard deviation of the sequence;

[0227] The standardization operation includes calculating the maximum and minimum values of each sample sequence, then subtracting the minimum value from each value in the sequence, and then dividing by the difference between the maximum and minimum values;

[0228] After the above preprocessing, the data set is input into the denoising recognition neural network model for training or testing.

[0229] It should be noted that the two network models, the BiLSTM-RESFFN network and the CAE-RESFFN network, provided in different embodiments of step S5 of this disclosure, can both independently achieve accurate identification of the operating mode of spaceborne SAR. Either one can be used alone, or both can be used simultaneously for identification, mutually verifying each other, and selecting the more reliable result as the final result.

[0230] In a specific embodiment, the BiLSTM-RESFFN network is used to identify the operating mode of the spaceborne SAR. The results are as follows: Figure 6 As shown, Figure 6 The "Predicted label" axis represents the predicted label, and the "True label" axis represents the true label. "Strip" represents the striping mode, "scan" represents the scanning mode, "spot" represents the spotting mode, and "sliding" represents the sliding spotting mode. The values of the diagonal grid from the upper left corner to the lower right corner represent the samples that were correctly identified. The number of samples for each working mode is 750.

[0231] from Figure 6 As can be seen, this embodiment accurately recognized 750 samples in stripe mode, with an accuracy rate of 100%; 720 samples in scanning mode, with an accuracy rate of 96%; 712 samples in spotlight mode, with an accuracy rate of 94.9%; and 686 samples in sliding spotlight mode, with an accuracy rate of 91.5%. Its average accuracy rate for spaceborne SAR operating modes is 95.6%.

[0232] In another specific embodiment, the CAE-RESFFN network is used to identify the operating mode of the spaceborne SAR. Figure 7 As shown. Similarly, Figure 7 The horizontal axis "Predicted label" represents the predicted label, and the vertical axis "True label" represents the true label. "Strip" represents the striping mode, "scan" represents the scanning mode, "spot" represents the spotting mode, and "sliding" represents the sliding spotting mode. The values of the diagonal grid from the upper left corner to the lower right corner represent the samples that were accurately recognized. The number of samples for each working mode is 750.

[0233] from Figure 7It can be seen that the present invention can accurately identify 750 samples in stripe mode with a recognition accuracy of 100%; 719 samples in scanning mode with a recognition accuracy of 95.9%; 694 samples in spotlight mode with a recognition accuracy of 92.5%; and 652 samples in sliding spotlight mode with a recognition accuracy of 86.9%. The recognition accuracy of the overall spaceborne SAR working mode is 93.8%.

[0234] The two methods of the present invention and three existing recognition methods were used to identify the overall spaceborne SAR operating mode, and the recognition accuracy was calculated. The compared algorithms included a traditional recognition network algorithm, a recognition method based on Euclidean distance, and a recognition method based on HOG feature extraction. The results are shown in Table 1.

[0235] Table 1 Comparison of recognition accuracy of different methods (%)

[0236]

[0237] As can be seen from Table 1, the recognition performance of the two denoising recognition networks proposed in the present invention, namely the BiLSTM-RESFFN network and the CAE-RESFFN network, is higher than that of the traditional recognition network, the existing recognition method based on Euclidean distance, and the existing recognition method based on HOG feature extraction. Among them, the recognition accuracy of the BiLSTM-RESFFN network is 8.3% higher than that of the traditional recognition network, 15.02% higher than that of the existing recognition method based on Euclidean distance, and 22.68% higher than that of the existing recognition method based on HOG feature extraction; the recognition accuracy of the CAE-RESFFN network is 6.5% higher than that of the traditional recognition network, 13.22% higher than that of the existing recognition method based on Euclidean distance, and 20.88% higher than that of the existing recognition method based on HOG feature extraction.

[0238] The above simulation results show that the two noise reduction and recognition networks proposed in the present invention have the ability to filter out noise from complex signals and extract pure signals, which can make the recognition of spaceborne SAR working modes more accurate and unaffected by interference.

[0239] In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in any one or more embodiments or examples.

[0240] Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying the working mode of spaceborne SAR based on a noise reduction recognition network, characterized in that: include: Arrange a number of reconnaissance stations in different layouts, obtain reconnaissance waveforms of different operating modes of the spaceborne SAR in each layout, and calculate the reconnaissance station received signals corresponding to the different operating modes; the operating modes include strip mode, spotlight mode, sliding spotlight mode, and scanning mode; The characteristic differences of the reconnaissance waveforms of different working modes of spaceborne SAR obtained under each layout are compared, and the layout with the largest characteristic difference is selected as the optimal layout of the reconnaissance station; Based on the determined optimal layout of the reconnaissance stations, each working mode is simulated to obtain the reconnaissance station receiving signal in the simulation environment; The calculated received signals of the reconnaissance station corresponding to different working modes are defined as pure signals; The received signal of the reconnaissance station obtained in the simulation environment is defined as a noisy signal, and a training set and a test set are constructed; the training set includes the noisy signal, the clean signal and the label sequence, and the test set includes the noisy signal and the label sequence; the label sequence is used to at least mark the working mode corresponding to the signal; Constructing a noise reduction and recognition neural network model, wherein the input of the noise reduction and recognition neural network model is a noisy signal, and the output is the spaceborne SAR operating mode corresponding to the noisy signal; the noise reduction and recognition neural network model includes a noise reduction network and a recognition network connected in series, wherein the noise reduction network is used to denoise the input signal, and the recognition network is used to identify the operating mode based on the denoised input signal; Use the constructed training set to train the denoising recognition neural network model; The test set is used to evaluate the training results of the denoising recognition neural network model.

2. The method for identifying the working mode of spaceborne SAR based on the noise reduction recognition network according to claim 1, characterized in that: The layout mode includes at least two of a first layout mode, a second layout mode and a third layout mode; In the first layout mode, all reconnaissance stations are arranged within the illumination range of the radar antenna main lobe; In the second layout mode, all reconnaissance stations are arranged at the edge of the radar antenna main lobe illumination range; In the third layout mode, all reconnaissance stations are arranged between the edge of the radar antenna main lobe illumination range and the antenna side lobe illumination range.

3. The method for identifying the working mode of spaceborne SAR based on a noise reduction recognition network according to claim 1, characterized in that: It is assumed that the satellite motion trajectory is the same in each working mode, and all of them fly along the y-axis; the calculation of the reconnaissance station receiving signal corresponding to different working modes includes: calculating the receiving signal of the reconnaissance station set at (Y, R) on the earth's surface under different working modes when the satellite moves to (0, y); the calculation result of the receiving signal includes the signal power received by the reconnaissance station and the received echo expression.

4. The method for identifying the working mode of spaceborne SAR based on a noise reduction recognition network according to claim 3, characterized in that: In strip mode, the calculation method of the signal received by the reconnaissance station includes: Calculate antenna patterns at different times and orientations; Calculate the echo power density received at the stripe pattern reconnaissance station location based on the antenna pattern; The signal power received at the stripe pattern reconnaissance station is calculated based on the echo power density, and its expression is: Among them, P r1 represents the signal power received at the stripe pattern reconnaissance station; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment; Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the stripe mode reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including: Among them, s1(y,t r ; R) represents the echo expression received at the stripe mode reconnaissance station position; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

5. The method for identifying the working mode of spaceborne SAR based on a noise reduction recognition network according to claim 3, characterized in that: In the spotlight mode, the calculation method of the signal received by the reconnaissance station includes: Calculate antenna patterns at different times and orientations; Calculate the echo power density received at the reconnaissance station in the spotlight mode based on the antenna pattern; The signal power received at the spotlight mode reconnaissance station is calculated based on the echo power density, and its expression is: Among them, P r2 represents the signal power received at the reconnaissance station in the spotlight mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment; Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the spotlight mode reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including: Among them, s2(y,t r ; R) represents the echo expression received by the spotlight mode reconnaissance station; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

6. The method for identifying the working mode of spaceborne SAR based on a noise reduction recognition network according to claim 3, characterized in that: In the sliding beam mode, the calculation method of the signal received by the reconnaissance station includes: Calculate antenna patterns at different times and orientations; Calculate the echo power density received at the reconnaissance station position in sliding beam mode based on the antenna pattern; The signal power received at the reconnaissance station in sliding beam mode is calculated based on the echo power density, and its expression is: Among them, P r3 represents the signal power received at the position of the reconnaissance station in sliding beam mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment; Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the position of the reconnaissance station in the sliding beam mode based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including: Among them, s2(y,t r ; R) represents the echo expression received by the reconnaissance station in the spotlight mode; represents the antenna pattern; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiated to the ground; c represents the speed of light; λ represents the wavelength; v b represents the beam movement speed; v represents the reference speed, that is, the satellite flight speed.

7. The method for identifying the working mode of spaceborne SAR based on a noise reduction recognition network according to claim 3, characterized in that: In scanning mode, the reconnaissance station antenna only rotates upward in the range and stays for a set time after each rotation. During this time, the reconnaissance station antenna is fixed in a fixed direction, which is equivalent to the strip mode and contains the information of one wave position. This time interval is defined as the sub-strip mode. The number of sub-strips in the scanning mode corresponds to the number of wave position information. Each scan starts from the first wave position to the last wave position. In scanning mode, the calculation method of the signal received by the reconnaissance station includes: Calculate antenna patterns at different times and orientations; Calculate the echo power density received at the reconnaissance station position in scanning mode based on the antenna pattern; The signal power received at the scanning mode reconnaissance station position is calculated based on the echo power density, and its expression is: Among them, P r1 represents the signal power received at the reconnaissance station in scanning mode; γ represents the polarization coefficient; Indicates the feeder coefficient; S indicates the echo power density; A r Represents the effective area of the receiving antenna; P t Indicates the transmitting power of the radar antenna; G t represents the gain of the transmitting antenna; t a Indicates the current moment; Calculate the instantaneous slant range, frequency modulation slope, and beam width to the ground. Calculate the echo expression received at the scanning mode reconnaissance station position based on the instantaneous slant range, frequency modulation slope, and beam width to the ground, including: Among them, s4(y,t r ; R) represents the echo expression received by the scanning mode reconnaissance station position; s 1m (y,t r ; R) represents the echo expression of the mth sub-strip; M represents the number of sub-strips in the scanning mode; rect(·) represents the rectangular window function; t r represents the distance to time; R(y; R) represents the instantaneous slant distance; K r Indicates the frequency modulation slope; L B represents the width of the beam irradiating the ground; c represents the speed of light; λ represents the wavelength.

8. The method for identifying spaceborne SAR working modes based on a noise reduction recognition network according to claim 1, wherein: The denoising network adopts a BiLSTM module, which includes a cascaded input layer, a bidirectional LSTM layer and a first fully connected layer; the bidirectional LSTM layer includes a hidden layer and an implicit layer; The recognition network adopts a RESFFN module, which includes a cascaded second fully connected layer, a first activation function layer, a first batch of normalized layers, a third fully connected layer, a second activation function layer, a second batch of normalized layers, five residual blocks, a fourth fully connected layer, a third activation function layer, a third batch of normalized layers, a fifth fully connected layer, a fourth activation function layer, a fourth batch of normalized layers, a sixth fully connected layer, and a fifth activation function layer; The denoising and recognition neural network formed by cascading the denoising network and the recognition network is a BiLSTM-RESFFN network; Set the root mean square loss function of the BiLSTM-RESFFN network MSE1 for: Among them, f1(x i ) represents the output signal of the BiLSTM-RESFFN network after denoising the i-th noise sample, x i represents the clean signal corresponding to the i-th noise sample, i∈[1,n], and n is the total number of samples in the training set; the difference between the denoised signal output by the denoising part of the BiLSTM-RESFFN network and its corresponding clean signal is calculated using the root mean square loss function; Set the cross entropy loss function of the BiLSTM-RESFFN network CEL1 for: Among them, g1(g i ) represents the label of the i-th sample after recognition in the BiLSTM-RESFFN network; g i Represents the true label of the i-th sample; the difference between the recognition label output by the BiLSTM-RESFFN network recognition part and its corresponding true label is calculated through the cross entropy loss function; Based on the root mean square loss function and the cross entropy loss function, the parameters of the BiLSTM-RESFFN network are iteratively updated by the gradient descent method to obtain the trained model.

9. The method for identifying the working mode of spaceborne SAR based on a noise reduction recognition network according to claim 1, characterized in that: The denoising network adopts a CAE module, which includes a cascaded input layer, a first convolutional layer, a first activation function layer, a first pooling layer, a second convolutional layer, a second activation function layer, a third convolutional layer, a first transposed convolutional layer, a third activation function layer, a fourth convolutional layer, a fourth activation function layer, a fifth convolutional layer and a fifth activation function layer; The recognition network adopts a RESFFN module, which includes a cascaded first fully connected layer, a sixth activation function layer, a first batch normalization layer, a second fully connected layer, a seventh activation function layer, a second batch normalization layer, five residual blocks, a third fully connected layer, an eighth activation function layer, a third batch normalization layer, a fourth fully connected layer, a ninth activation function layer, a fourth batch normalization layer, a fifth fully connected layer, and a tenth activation function layer; The denoising and recognition neural network formed by cascading the denoising network and the recognition network is the CAE-RESFFN network; Set the root mean square loss function of the CAE-RESFFN network MSE2 for: Among them, f2(x i ) is the output signal of the denoised i-th noise sample of the CAE-RESFFN network, x i is the clean signal corresponding to the i-th noise sample, i∈[1,n], n is the total number of samples in the training set; the difference between the denoised signal output by the denoising part of the CAE-RESFFN network and its corresponding clean signal is calculated using the root mean square loss function; Among them, g2(g i ) represents the label of the i-th sample after recognition in the CAE-RESFFN network, g i represents the true label of the i-th sample; the difference between the recognition label output by the CAE-RESFFN network recognition part and its corresponding true label is calculated through the cross entropy loss function; Based on the root mean square loss function and the cross entropy loss function, the parameters of the CAE-RESFFN network are iteratively updated by the gradient descent method to obtain the trained model.

10. The method for identifying spaceborne SAR working modes based on a noise reduction recognition network according to claim 1, characterized in that: Before using the constructed training set to train the denoising recognition neural network model or using the test set to evaluate the training results of the denoising recognition neural network model, the following steps are also included: Perform centralization and standardization operations on samples of the dataset; The centering operation includes calculating the mean and standard deviation of each sample sequence, then subtracting the mean from each value of the sequence and dividing it by the standard deviation of the sequence; The normalization operation includes calculating the maximum and minimum values of each sample sequence, then subtracting the minimum value from each value in the sequence, and then dividing by the difference between the maximum and minimum values.

Images