Microwave radar sea clutter suppression method and system based on convolutional denoising autoencoder

Abstract

The present invention proposes a microwave radar sea clutter suppression method and system based on convolutional denoising autoencoder, including selecting microwave radar experimental radar target echo pulse train data as reference echo, and performing zero-filling image processing; establishing a time domain decorrelation model of sea clutter; establishing a sequence model of target plus clutter with rich signal-to-clutter ratio according to echo reference data, and performing image processing; using convolutional denoising autoencoder and Inception network, establishing a radar sea clutter suppression neural network model RCSCDAE based on deep convolutional denoising autoencoder; training the center value, variance and weights from hidden layer to output layer of RCSCDAE neural network basis function; using RCSCDAE neural network to suppress sea clutter and restore target signal. The present invention adopts RCSCDAE neural network method to suppress sea clutter and reconstruct target signal, overcomes the problems of low real-time performance and large target reconstruction energy loss of existing traditional methods, and provides new technical means for microwave radar sea clutter suppression.

Description

Technical Field

[0001] The present invention belongs to the technical field of sea clutter suppression, and relates to a microwave radar sea clutter suppression technical solution based on a deep learning network model of convolutional denoising autoencoder. Background Art

[0002] For microwave radar, the traditional clutter suppression methods of moving target indication (MTI) and moving target detection (MTD) for narrowband radar cannot solve the problem of moving targets and clutter during coherent processing and the possible movement of units across distances. Therefore, they are no longer applicable and new methods need to be studied. For microwave target recognition radar, it is necessary not only to filter out clutter at various signal-to-clutter ratios, but also to keep the target signal characteristics intact as much as possible in order to complete subsequent recognition work.

[0003] The incoherent clutter suppression method mainly compares the instantaneous positions of two echo target signals reflected from the same moving target after a certain time interval to distinguish fixed targets from moving signals. However, this method fails in moving sea clutter. When the target speed is unknown, the Hough transform method used to extract line segments in image processing is introduced into the field of radar clutter suppression. Although it can effectively suppress clutter and improve recognition rate under low signal-to-clutter ratio conditions, it has a large amount of calculation and low real-time performance, which makes it difficult to meet the needs of radar engineering practice; and the energy loss of the target signal is large, which brings great difficulties to subsequent recognition problems.

[0004] In summary, there are at least the following technical problems in the prior art: the existing sea clutter suppression effect of microwave radar in strong sea clutter areas is not ideal, the amount of calculation is large and the real-time performance is low, which cannot meet the requirements of radar engineering practice; the energy loss of the target after sea clutter suppression is large, which is difficult to meet the requirements of subsequent identification in engineering practice. Summary of the invention

[0005] The technical problem solved by the present invention is: to overcome the deficiencies of the prior art, to propose a sea clutter suppression scheme based on convolution denoising autoencoder, to improve the sea clutter suppression effect of microwave radar in strong sea clutter areas; to solve the problems of large amount of calculation, low real-time performance and difficulty in meeting the requirements of radar engineering practice in the prior art method for suppressing sea clutter interference of microwave radar, to achieve the technical effect that the sea clutter interference under various signal-to-clutter ratios can be automatically suppressed; and after the sea clutter is suppressed, the target energy loss is less and the requirements of subsequent identification in radar engineering practice can be met.

[0006] In order to solve the above technical problems, the present invention provides a microwave radar sea clutter suppression method based on convolution denoising autoencoder, comprising the following steps:

[0007] 1) Select the radar target echo pulse train data of the microwave radar experiment as the reference echo and perform zero-filling imaging processing;

[0008] 2) Establish a time-domain decorrelation model of sea clutter; establish a sequence model of the target plus clutter with rich signal-to-clutter ratio based on echo reference data, and perform image processing;

[0009] 3) Using convolutional denoising autoencoder and Inception network, establish the radar sea clutter suppression neural network model RCSCDAE based on deep convolutional denoising autoencoder;

[0010] 4) Train the center value, variance and weights from hidden layer to output layer of the RCSCDAE neural network basis function;

[0011] 5) Use the RCSCDAE neural network to suppress sea clutter and restore the target signal.

[0012] Moreover, in the microwave radar experimental radar target echo pulse train data in step 2, the radar bandwidth is defined as B, then the radar range resolution is Δr=c / 2B, where c is the speed of light; along the line of sight of the radar detection target, the detected target is approximately expressed as a certain number of range units, with a bandwidth of Δr, where the echo of the i-th range unit is the superposition of the echoes corresponding to all scattering centers in the range unit, and the model formula is formulated as:

[0013]

[0014] In the formula,

[0015] N i represents the number of target scattering centers located in the i-th range element;

[0016] a i，k is the scattering intensity of the kth scattering center in the ith range element, which is related to the target shape structure;

[0017] τ i，k The arrival time of the kth scattering center in the i-th range element;

[0018] x i The superposition signal of all target scattering center echoes in the i-th range element;

[0019] j is the sign of the imaginary part, I(i) is the real part of the superimposed signal of the echoes from the scattering centers of all targets in the ith range element, and Q(i) is the imaginary part of the superimposed signal of the echoes from the scattering centers of all targets in the ith range element.

[0020] Moreover, the zero-filling image processing in step 1 is implemented by modulo processing of the target data, and the target data is defined as:

[0021]

[0022] In the formula,

[0023] n is the data dimension, |x 1 |,|x 2 |, ..., |x n |x respectively 1 , x 2 , ..., x n The modulus value, x 1 , x 2 , ..., x n are the echo amplitudes of n range elements respectively, I(1), I(2), ..., I(n) are the real part signals of radar echo data respectively, Q(1), Q(2), ..., Q(n) are the imaginary part signals of radar echo data respectively.

[0024] Moreover, the time domain decorrelation model of sea clutter in step 2 is established by using a zero-memory nonlinear transformation method, which includes first transforming the normally distributed Gaussian white noise sequence {x i} as the input of the linear transformation, the output conforms to the normal distribution Gaussian colored noise correlation sequence {y i}, and then use it as the input of nonlinear transformation to obtain the non-Gaussian colored noise correlation sequence {z i}, which is used to construct simulated sea clutter with a specific probability density function PDF and a specific power spectrum density PSD.

[0025] Moreover, in step 3, a denoising autoencoder is used as the training criterion of the neural network, and its output function is defined as:

[0026]

[0027] In the formula,

[0028] y is the denoising autoencoder output;

[0029] is the data where the original data x is contaminated by the corresponding distributed sea clutter;

[0030] f θ () is the encoding network function;

[0031] g θ′ () is the decoding network function;

[0032] The denoising autoencoder is applied to the convolutional neural network to obtain the convolutional denoising autoencoder.

[0033] Moreover, in step 3, each convolutional layer of the convolutional denoising autoencoder uses the Inception network structure.

[0034] Moreover, in step 4, when training the RCSCDAE neural network, the weights of the input layer, each convolution layer, and each deconvolution layer are trained, and f θ () and g θ′ () is the function that needs to be approximated after the neural network is trained.

[0035] Furthermore, the use of the RCSCDAE neural network to suppress sea clutter and restore the target signal in step 5 includes achieving sea clutter suppression and target recovery from the perspective of distinguishing the recognition signal from the time domain image with the clutter signal.

[0036] On the other hand, the present invention also provides a microwave radar sea clutter suppression system based on convolutional denoising autoencoding, which is used to implement the microwave radar sea clutter suppression method based on convolutional denoising autoencoding as described above.

[0037] Furthermore, the following modules are included,

[0038] The first module is used to select the radar target echo pulse train data of the microwave radar experiment as the reference echo and perform zero-filling imaging processing;

[0039] The second module is used to establish a time-domain decorrelation model of sea clutter; establish a sequence model of the target plus clutter with rich signal-to-clutter ratio based on echo reference data, and perform image processing;

[0040] The third module is used to establish a radar sea clutter suppression neural network model RCSCDAE based on deep convolutional denoising autoencoder using convolutional denoising autoencoder and Inception network;

[0041] The fourth module is used to train the center value, variance and weights from the hidden layer to the output layer of the RCSCDAE neural network basis function;

[0042] The fifth module is used to suppress sea clutter and restore target signals using the RCSCDAE neural network.

[0043] The present invention adopts the RCSCDAE neural network method to suppress sea clutter and reconstruct the target signal, overcoming the problems of low real-time performance and large target reconstruction energy loss of the existing traditional methods, and provides a new idea for microwave radar sea clutter suppression. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a principle block diagram of a sea clutter suppression method based on convolution denoising autoencoding according to an embodiment of the present invention;

[0045] Figure 2 is a structural diagram of the RCSCDAE neural network of an embodiment of the present invention;

[0046] Figure 3Schematic diagram of the improvement of signal-to-clutter ratio after sea clutter suppression by RCSCDAE neural network in an embodiment of the present invention;

[0047] Figure 4 It is a schematic diagram comparing the average relative errors before and after the RCSCDAE neural network target reconstruction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0049] The invention proposes a microwave radar sea clutter suppression method based on convolutional denoising autoencoder, which comprises the following steps: selecting microwave radar experimental target echo pulse train data as reference echo, and performing zero-filling image processing; establishing a time domain decorrelation model of sea clutter, and establishing a sequence model of a target plus clutter with a rich signal-to-clutter ratio according to echo reference data, and performing image processing; establishing a radar sea clutter suppression neural network model based on deep convolutional denoising autoencoder (Radar Sea Clutter Suppression on DeepConvolutional Denoising Autoencoder, RCSCDAE) by utilizing the theories and respective advantages of convolutional denoising autoencoder and Inception network; training the central value, variance and weight from hidden layer to output layer of RCSCDAE neural network basis function; and suppressing sea clutter by utilizing the RCSCDAE neural network to restore target signal.

[0050] See also Figure 1 The present invention provides a method for suppressing sea clutter in a microwave radar based on a convolutional denoising autoencoder, comprising the following steps:

[0051] 1) Select the radar target echo pulse train data of the microwave radar experiment as the reference echo and perform zero-filling imaging processing;

[0052] The zero-filling image processing in step 1 belongs to the image processing technology, including the modulo processing of the target data. The target data is defined as:

[0053]

[0054] In the formula,

[0055] n is the data dimension, |x 1 |,|x 2 |, ..., |x n |x respectively 1 , x 2 , ..., x n The modulus value, x 1 , x 2 , ..., x nare the echo amplitudes of n range elements respectively, I(1), I(2), ..., I(n) are the real part signals of radar echo data respectively, Q(1), Q(2), ..., Q(n) are the imaginary part signals of radar echo data respectively.

[0056] For example, every 140 groups of sequences are imaged to obtain an original image with an image block size of 140×360 pixels.

[0057] In the embodiment, data preprocessing includes transforming the received radar signal into the time-frequency domain through time-frequency processing, and then performing data normalization and data imaging. After imaging, an image sequence with a size of 140×360 pixels is obtained, with a total of 2952 original images and 2952 images with noise.

[0058] 2) Establish a time-domain decorrelation model of sea clutter; establish a sequence model of the target plus clutter with rich signal-to-clutter ratio based on echo reference data, and perform image processing;

[0059] The microwave radar sequence signal of the embodiment collects ship-type targets, the target attitude angle changes from 0° to 360°, and a sequence signal is collected every 1°.

[0060] The microwave radar experimental radar target echo pulse train data in step 2 is defined as radar bandwidth B, then the radar range resolution is Δr=c / 2B, where c is the speed of light. Along the line of sight of the radar detection target, the detected target can be approximately expressed as a certain number of range units with a bandwidth of Δr, where the echo of the i-th range unit is the superposition of the echoes corresponding to all scattering centers in the range unit, and the model formula is:

[0061]

[0062] In the formula,

[0063] N i represents the number of target scattering centers located in the i-th range element;

[0064] a i，k is the scattering intensity of the kth scattering center in the ith range element, which is related to the target shape structure;

[0065] τ i，k The arrival time of the kth scattering center in the i-th range element;

[0066] x i The superposition signal of all target scattering center echoes in the i-th range element;

[0067] j is the sign of the imaginary part, I(i) is the real part of the superimposed signal of the echoes from the scattering centers of all targets in the ith range element, and Q(i) is the imaginary part of the superimposed signal of the echoes from the scattering centers of all targets in the ith range element.

[0068] The time domain decorrelation model of sea clutter in step 2 is established by using the zero-memory non-linearily transformation method (ZMNL). The basic idea can be described as follows: first, the normally distributed Gaussian white noise sequence {x i} as the input of the linear transformation, the output conforms to the normal distribution Gaussian colored noise correlation sequence {y i}, and then use it as the input of nonlinear transformation to obtain the non-Gaussian colored noise correlation sequence {z i}, which is used to construct simulated sea clutter with a specific probability density function (PDF) and a specific power spectral density (PSD), and to construct data with rich signal-to-clutter ratios of targets and sea clutter, so that the clutter suppression model has robustness.

[0069] In the embodiment, every 140 groups of sequences in step 2 are imaged to obtain a noise-containing image with an image block size of 140×360 pixels.

[0070] 3) Using the theories and respective advantages of convolutional denoising autoencoders and Inception networks, a radar sea clutter suppression neural network model based on deep convolutional denoising autoencoder (Radar Sea Clutter Suppression on Deep Convolutional Denoising Autoencoder, RCSCDAE) was established;

[0071] In step 3, a denoising autoencoder is used as a training criterion for a neural network, and its output function is defined as:

[0072]

[0073] in,

[0074] y is the denoising autoencoder output;

[0075] is the data where the original data x is contaminated by the corresponding distributed sea clutter;

[0076] f θ () is the encoding network function;

[0077] g θ′ () is the decoding network function;

[0078] The denoising autoencoder is applied to the convolutional neural network to obtain the convolutional denoising autoencoder.

[0079] The use of the convolutional denoising autoencoder and the Inception network in step 3 refers to using the Inception network structure for each convolutional layer of the convolutional denoising autoencoder.

[0080] In the embodiment, the structure of the deep convolutional autoencoder neural network designed using the RCSCDAE algorithm is as follows: Figure 2 As shown, the network consists of two parts: Network 1 and Network 2:

[0081] a) Network 1 can be considered as the feature extraction part, which consists of one convolutional layer (conv) and four Inception structure layers. The Inception layer includes one Inception V1 layer, two Inception V2 layers and one Inception V3 layer, which reduces the network parameters, improves the network performance and increases the network's expression ability. A max pooling layer is added after each Inception V2 and Inception V3 to ensure the translation invariance of the features.

[0082] b) Network 2 is the image reconstruction part, which is used to reconstruct the image and restore the original signal. It consists of seven deconvolution layers, which are denoted as Decon v1, Decon v2, … Decon v7.

[0083] The present invention also proposes that a BN (BatchNormalization) layer and a ReLU activation function layer are added after each convolution operation layer (including a convolution layer conv and four Inception structure layers in network 1) and a deconvolution operation layer (Decon v1, Decon v2, ... Decon v7) to solve the gradient disappearance problem, improve the network accuracy, and make the network easier to optimize.

[0084] The input image passes through network 1 and network 2 in sequence to obtain the output image.

[0085] in,

[0086] The Inception module structure draws on the idea of ​​"network in network", including multiple convolutional layers and a pooling layer, using different convolution kernels to obtain data convolution features under different receptive fields;

[0087] The Inception V1 structure adds a 1×1 convolution operation before the 3×3 and 5×5 convolution operations, reducing the number of channels of the feature maps entering the 3×3 and 5×5 convolution layers, thereby reducing the amount of calculation. At the same time, a 1×1 convolution is added before the output of the pooling layer is merged into the overall output, reducing the number of channels of the pooling layer output. The parameters of the 5×5 convolution kernel of Inception V2 are 25 / 9=2.78 times that of the 3×3 convolution kernel, so two consecutive 3×3 convolutions are used to replace the 5×5 convolution in Inception v1, which not only reduces the number of parameters but also speeds up the calculation. The most important improvement of Inception V3 is the decomposition of convolution, that is, decomposing a 3×3 convolution into two parallel convolutions of 3×1 and 1×3. This decomposition will not reduce the feature expression ability of the network, but will reduce network parameters and improve network performance.

[0088] BN (Batch Normalization) is a batch normalization layer that normalizes the input data of each layer to between 0 and 1, reducing the changes in the distribution of input data of each layer, thereby accelerating the convergence of the network.

[0089] ReLU is a nonlinear activation function that enhances the nonlinear expression ability of the network.

[0090] 4) Train the center value, variance and weights from hidden layer to output layer of the RCSCDAE neural network basis function;

[0091] The original image and the image with noise are used as training sets to train the RCSCDAE neural network, which consists of a feature extraction part and an image reconstruction part, to extract the robust features of the time-frequency image.

[0092] Step 4 is to train the RCSCDAE network, which mainly includes training the weights of the input layer, each convolution layer, and each deconvolution layer. θ () and g θ′ () is the function that needs to be approximated after the neural network is trained.

[0093] During specific implementation, iterative training may be performed until a preset iteration condition (such as the number of iterations) is met, thereby obtaining a final RCSCDAE neural network sea clutter suppression model.

[0094] 5) Use the RCSCDAE neural network to suppress sea clutter and restore the target signal.

[0095] The use of the RCSCDAE neural network to suppress sea clutter and restore the target signal in step 5 includes achieving sea clutter suppression and target recovery from the perspective of distinguishing the recognition signal from the time domain image with the clutter signal.

[0096] The improvement of signal-to-clutter ratio in the quantitative analysis of sea clutter suppression results. Since there is still no unified definition of signal-to-clutter ratio for microwave radar, the signal-to-clutter ratio definition formula commonly used in radar target detection is still used:

[0097] SCR=10log 10 P av / P c

[0098] in,

[0099] P av is the average signal energy;

[0100] P c is the average energy of clutter.

[0101] The specific improvement of signal-to-noise ratio is as follows: Figure 3 shown.

[0102] The quantitative analysis of the target reconstruction performance evaluation results uses the mean square error (MSE) formula:

[0103]

[0104] in,

[0105] x is the original data;

[0106] is the original data x contaminated by the corresponding distributed sea clutter data.

[0107] MSE is used to evaluate the quality of target recovery performance. The specific reconstruction situation is as follows Figure 4 As shown. Figure 4 It can be seen that the mean square error between the restored signal and the original signal approaches 0 at all signal-to-noise ratios, indicating that this method has certain practicability in achieving sea clutter suppression and microwave radar target signal reconstruction.

[0108] In specific implementation, the method proposed in the technical solution of the present invention can be implemented by technical personnel in this field using computer software technology to realize automatic operation process. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of the present invention and computer equipment running the corresponding computer program, should also be within the scope of protection of the present invention.

[0109] In some possible embodiments, a microwave radar sea clutter suppression system based on convolutional denoising autoencoder is provided, comprising the following modules:

[0110] The first module is used to select the radar target echo pulse train data of the microwave radar experiment as the reference echo and perform zero-filling imaging processing;

[0111] The second module is used to establish a time-domain decorrelation model of sea clutter; establish a sequence model of the target plus clutter with rich signal-to-clutter ratio based on echo reference data, and perform image processing;

[0112] The third module is used to establish a radar sea clutter suppression neural network model RCSCDAE based on deep convolutional denoising autoencoder using convolutional denoising autoencoder and Inception network;

[0113] The fourth module is used to train the center value, variance and weights from the hidden layer to the output layer of the RCSCDAE neural network basis function;

[0114] The fifth module is used to suppress sea clutter and restore target signals using the RCSCDAE neural network.

[0115] In some possible embodiments, a microwave radar sea clutter suppression system based on convolutional denoising autoencoder is provided, including a processor and a memory, the memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute the microwave radar sea clutter suppression method based on convolutional denoising autoencoder as described above.

[0116] In some possible embodiments, a microwave radar sea clutter suppression system based on convolutional denoising autoencoder is provided, including a readable storage medium, on which a computer program is stored. When the computer program is executed, a microwave radar sea clutter suppression method based on convolutional denoising autoencoder as described above is implemented.

[0117] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A microwave radar sea clutter suppression method based on convolutional denoising autoencoder, characterized in that: The following steps are involved: 1) Select the radar target echo pulse train data of the microwave radar experiment as the reference echo and perform zero-filling imaging processing; 2) Establish a time-domain decorrelation model of sea clutter; establish a sequence model of the target plus clutter with rich signal-to-clutter ratio based on echo reference data, and perform image processing; 3) Using the convolutional denoising autoencoder and the Inception network, a radar sea clutter suppression neural network model RCSCDAE based on a deep convolutional denoising autoencoder is established. Each convolutional layer of the convolutional denoising autoencoder uses the Inception network structure. 4) Train the center value, variance and weights from hidden layer to output layer of the RCSCDAE neural network basis function; 5) Use the RCSCDAE neural network to suppress sea clutter and restore the target signal.

2. The microwave radar sea clutter suppression method based on convolutional denoising autoencoder according to claim 1 is characterized in that: In the microwave radar experiment in step 2, the radar target echo pulse train data is defined as radar bandwidth B, then the radar range resolution is Δr=c / 2B, where c is the speed of light; along the line of sight of the radar detection target, the detected target is approximately expressed as a certain number of range units with a bandwidth of Δr, where the echo of the i-th range unit is the superposition of the echoes corresponding to all scattering centers in the range unit, and the model is formulated as: In the formula, N i represents the number of target scattering centers located in the i-th range element; a i,k is the scattering intensity of the kth scattering center in the ith range element, which is related to the target shape structure; τ i,k The arrival time of the kth scattering center in the i-th range element; x i The superposition signal of all target scattering center echoes in the i-th range element; j is the sign of the imaginary part, I(i) is the real part of the superimposed signal of the echoes from the scattering centers of all targets in the ith range element, and Q(i) is the imaginary part of the superimposed signal of the echoes from the scattering centers of all targets in the ith range element.

3. The microwave radar sea clutter suppression method based on convolutional denoising autoencoder according to claim 1 is characterized in that: The zero-filling image processing in step 1 is implemented by modulo processing of the target data, and the target data is defined as: In the formula, n is the data dimension, |x1|,|x2|,…,|x n |x1,x2,…,x respectively n The modulus of x1,x2,…,x n are the echo amplitudes of n range elements respectively, I(1), I(2),…, I(n) are the real part signals of radar echo data respectively, Q(1), Q(2),…, Q(n) are the imaginary part signals of radar echo data respectively.

4. The microwave radar sea clutter suppression method based on convolutional denoising autoencoder according to claim 1 is characterized in that: The time domain decorrelation model of sea clutter in step 2 is established by using a zero-memory nonlinear transformation method, which includes first transforming the normally distributed Gaussian white noise sequence {x i } as the input of the linear transformation, the output conforms to the normal distribution Gaussian colored noise correlation sequence {y i }, and then use it as the input of nonlinear transformation to obtain the non-Gaussian colored noise correlation sequence {z i }, which is used to construct simulated sea clutter with probability density function PDF and power spectrum density PSD.

5. The microwave radar sea clutter suppression method based on convolutional denoising autoencoder according to claim 1 is characterized in that: In step 3, a denoising autoencoder is used as the training criterion of the neural network, and its output function is defined as: In the formula, y is the denoising autoencoder output; is the data where the original data x is contaminated by the corresponding distributed sea clutter; f θ () is the encoding network function; g θ′ () is the decoding network function; The denoising autoencoder is applied to the convolutional neural network to obtain the convolutional denoising autoencoder.

6. The microwave radar sea clutter suppression method based on convolutional denoising autoencoder according to claim 5 is characterized in that: In step 4, when training the RCSCDAE neural network, the weights of the input layer, each convolution layer, and each deconvolution layer are trained. θ () and g θ′ () is the function that needs to be approximated after the neural network is trained.

7. The microwave radar sea clutter suppression method based on convolutional denoising autoencoder according to claim 1 or 2 or 3 or 4 or 5 or 6, characterized in that: The use of the RCSCDAE neural network to suppress sea clutter and restore the target signal in step 5 includes achieving sea clutter suppression and target recovery from the perspective of distinguishing the recognition signal from the time domain image with the clutter signal.

8. A microwave radar sea clutter suppression system based on convolutional denoising autoencoder, characterized in that: Used to implement a microwave radar sea clutter suppression method based on convolution denoising autoencoder as described in any one of claims 1-7.

9. The microwave radar sea clutter suppression system based on convolutional denoising autoencoder according to claim 8, characterized in that: Includes the following modules, The first module is used to select the radar target echo pulse train data of the microwave radar experiment as the reference echo and perform zero-filling imaging processing; The second module is used to establish a time-domain decorrelation model of sea clutter; establish a sequence model of the target plus clutter with rich signal-to-clutter ratio based on echo reference data, and perform image processing; The third module is used to establish a radar sea clutter suppression neural network model RCSCDAE based on deep convolutional denoising autoencoder using convolutional denoising autoencoder and Inception network; The fourth module is used to train the center value, variance and weights from the hidden layer to the output layer of the RCSCDAE neural network basis function; The fifth module is used to suppress sea clutter and restore target signals using the RCSCDAE neural network.

Images