Target echo detection method, system, device and medium based on multi-domain reconstructed residual error

Abstract

The application discloses a target echo detection method and system based on multi-domain reconstruction residual, equipment and medium, and the method comprises the steps of collecting sea clutter images and classifying, designing and training a double auto-encoder, training a multi-domain encoder based on the double auto-encoder on a sea clutter image dataset subjected to Fourier transform, ridge wave transform and contour wave transform, calculating the reconstruction residual of the sea clutter image in the multi-domain through the multi-domain encoder, obtaining an optimal expression during target echo detection through SVM supervised learning, and completing target detection based on the reconstruction residual and the optimal expression; the system, equipment and medium are used for realizing a target echo detection method based on multi-domain reconstruction residual; the application learns the current sea clutter structural features through the constructed double auto-encoder, and learns the feature difference between the target and the sea clutter through SVM supervised learning, and has the characteristics of high measurement accuracy and strong applicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection, and particularly relates to a target echo detection method and system based on multi-domain reconstruction residual, a device and a medium. BACKGROUND

[0002] In recent years, unmanned aerial vehicles can perform remote monitoring, communication interception, electronic jamming, precision strikes and other tasks; unmanned aerial vehicles in no-fly zones can pose a great threat to the smoothness of the channel and the safety of ground personnel. Therefore, it is particularly important to counter non-cooperative unmanned aerial vehicles, and the premise of countering non-cooperative unmanned aerial vehicles is to effectively detect them. At present, radar-based unmanned aerial vehicle detection plays an important role in national defense and civil security. On this basis, the problem of maritime unmanned aerial vehicle detection needs to be solved. However, due to the complexity of the sea and the influence of the curvature of the earth, the detection range of low-altitude targets by shore-based and shipborne radars is limited, and maritime low, small and slow target detection faces many difficulties. Existing target detection methods are divided into amplitude information-based detection and echo feature-based detection. Due to the low amplitude of maritime targets, the amplitude-based detection method represented by constant false alarm detection is difficult to detect weak echo targets. The echo feature-based detection method needs a large number of labeled echo samples for learning, and it is difficult to obtain effective learning samples to train the detector in the case of unknown non-cooperative target types and variable sea clutter. Therefore, for maritime low, small and slow target detection, the above two types of target detection methods respectively have the problems of low detection performance and insufficient applicability.

[0003] The invention patent application with publication number CN114384523A proposes a FMCW radar constant false alarm target detection method based on peak value comparison. This method is based on the OS-CFAR ordered constant false alarm detection method, and through the joint FFT transform result and the first detection result, the peak value comparison method is added to eliminate clutter and noise interference, and the shielding problem between targets in a multi-target environment is solved. The disadvantage of this method is that it contains many parameters that are difficult to adjust. When multiple targets exist and noise distribution is variable, it is difficult to obtain suitable parameters and models to achieve optimal detection performance. Moreover, this method estimates the detection threshold through amplitude information, ignoring the information structure features of echo texture, making it difficult to detect targets with weak echo amplitude.

[0004] The patent application with the publication number CN113936205A proposes a sea target detection method based on ensemble learning. Two kinds of base detector network structures based on single-shot multi-frame convolutional neural network and multi-scale feature residual network framework are designed for sea target features to realize detection of targets with different scales. The K-means clustering idea is introduced to optimize the generation of model prediction boxes, and a self-gated non-monotonic function is used as the activation function. Based on the boosting method and weighted voting method, the detection results of the two base detectors are integrated to complete sea target detection. The shortcomings of this method are as follows: first, when the target echo is missing or the sample is scarce, it is difficult to train a neural network with good detection effect. Second, when the sea conditions change, the characteristics of the clutter change, or the target species change, the detection performance of the target detector trained for the new environment and new target will decrease, limiting the applicability of the method. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a target echo detection method, system, device and medium based on multi-domain reconstruction residual, which trains a twin encoder through classified sea clutter images, and trains a sea clutter image dataset subjected to Fourier transform, ridge wave transform and contour wave transform through a double autoencoder, obtains a multi-domain encoder, calculates the reconstruction residual of each collected sea clutter image in the multi-domain sea clutter image dataset through the multi-domain encoder, and finally performs target detection based on the reconstruction residual and the optimal expression obtained through SVM supervised learning during target echo detection, which has the characteristics of high detection accuracy and strong applicability.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is:

[0007] The target echo detection method based on multi-domain reconstruction residual comprises the following steps:

[0008] Step 1, collect sea clutter images, and classify the collected sea clutter images to obtain a sea clutter image dataset;

[0009] Step 2, design a double autoencoder, and train the double autoencoder based on the sea clutter image dataset obtained in step 1;

[0010] Step 3, perform Fourier transform, ridge wave transform and contour wave transform on the sea clutter image dataset used in step 2, and train a double autoencoder in the multi-domain through the double autoencoder designed in step 2. After the training is completed, a multi-domain encoder is obtained;

[0011] Step 4, input the sea clutter image dataset obtained in step 1 into the multi-domain encoder obtained in step 3, and calculate the reconstruction residual of each collected sea clutter image in the multi-domain sea clutter image dataset;

[0012] Step 5, based on the reconstruction residual calculated in step 4, the optimal expression in the target echo detection is obtained through SVM supervised learning;

[0013] Step 6, the reconstruction residual of the sea clutter image sample of the to-be-detected region in the multi-domain is calculated through step 4, and whether the target exists is judged based on the optimal expression obtained in step 5, and the target detection is completed.

[0014] The step 1 specifically comprises the following steps:

[0015] Step 1.1, acquiring sea clutter images through a networked radar;

[0016] Step 1.2, dividing the sea clutter images acquired in step 1.1 into two categories according to whether the target exists: general sea clutter images without targets and sea clutter images with cooperative targets.

[0017] The step 2 specifically comprises the following steps:

[0018] Step 2.1, designing a double autoencoder:

[0019] First, the first encoder E11 extracts features from the original sea clutter image p 11 , to obtain the image features f 11 after dimensionality reduction and compression of the original sea clutter image p 11 , and then the decoder D1 decodes the image features f 11 to obtain the reconstructed image p 12 , and then the second encoder E12 extracts features from the reconstructed image p 12 , to obtain the reconstruction features f 12 of the reconstructed image p 12 ;

[0020] Step 2.2, inputting the sea clutter image dataset into the double autoencoder designed in step 2.1 to perform unsupervised learning training of the double autoencoder:

[0021] 30-50% of the general sea clutter images without targets are used as sea clutter image data for training the double autoencoder, and the unsupervised learning training principle of the double autoencoder can be represented by the following formula:

[0022]

[0023] Wherein, p 11 is the input original sea clutter image, p 12 is the reconstructed image, f 11 is the image feature extracted by the first encoder E11 from the original sea clutter image p 11 , and f 12 is the reconstruction feature of the reconstructed image p 12The extracted reconstruction features, alpha and beta are self-set parameters.

[0024] The step 3 specifically includes the following steps:

[0025] The 30-50% of the original sea clutter images p in the general sea clutter images without targets used in step 2.2 11 Respectively, Fourier transform, ridgelet transform, contourlet transform, wherein the original sea clutter image p 11 The image sample obtained by Fourier transform is p 21 , p 21 As a data sample set for training the multi-domain encoder, train the double auto-encoder in the corresponding transform domain:

[0026]

[0027] The original sea clutter image p 11 The image sample obtained by ridgelet transform is p 31 , p 31 As a data sample set for training the multi-domain encoder, train the double auto-encoder in the corresponding transform domain:

[0028]

[0029] The original sea clutter image p 11 The image sample obtained by contourlet transform is p 41 , p 41 As a data sample set for training the multi-domain encoder, train the double auto-encoder in the corresponding transform domain:

[0030]

[0031] After training, the multi-domain encoder is obtained:

[0032]

[0033]

[0034]

[0035]

[0036] The step 4 specifically includes the following steps:

[0037] The sea clutter image containing the cooperative target and the general sea clutter image without the target remaining in step 2.2 are input into the multi-domain encoder obtained in step 3, the reconstruction residual of the sea clutter image in the selected data set is calculated, and the calculation formula is as follows:

[0038]

[0039]

[0040]

[0041]

[0042] The reconstruction residual in the multi-domain is obtained, wherein d1 is the reconstruction residual of the original sea clutter image p in the selected sea clutter image data set, d2 is the reconstruction residual of the image p obtained after Fourier transform of the original sea clutter image, d3 is the reconstruction residual of the image p obtained after ridgelet transform of the original sea clutter image, and d4 is the reconstruction residual of the image p obtained after contourlet transform of the original sea clutter image. 11 21 31 41

[0043] The step 5 specifically comprises the following steps:

[0044] Step 5.1, taking the reconstruction residual in the multi-domain of the sea clutter image in the training data set as the learning feature of the SVM supervised learning;

[0045] Step 5.2, taking the sea clutter image with the cooperative target in the training data set as the positive sample and marking the label y=1, and taking the general sea clutter image without the target as the negative sample and marking the label y=-1, as the sample label of the SVM supervised learning;

[0046] Step 5.3, obtaining the optimal expression in the target echo detection based on the learning feature of step 5.1 and the sample label of step 5.2 through the SVM supervised learning:

[0047] min(y×(γ1d1+γ2d2+γ3d3+γ4d4),

[0048] The maximum discrimination expression is obtained through the SVM supervised learning:

[0049] ε=γ1d1+γ2d2+γ3d3+γ4d4+η,

[0050] The parameter values of γ1, γ2, γ3, γ4 and η in the optimal expression of the target echo detection are obtained through the training.

[0051] The step 6 specifically comprises the following steps:

[0052] ​​​​The multi-domain reconstruction is performed on the to-be-detected region to obtain a reconstructed residual {d1, d2, d3, d4} of the to-be-detected region, and a maximum discriminant expression formula ε=γ1d1+γ2d2+γ3d3+γ4d4+η is used for judgment: when ε>0, it is considered that the target exists, at this time, the position and speed information of all target tracks detected are extracted, the target detection calculation based on the signal echo feature and the reconstructed residual is completed, and a target detection result is obtained; when ε<0, it is considered that the to-be-detected region is a sea clutter without target.

[0053] The target echo detection system based on multi-domain reconstructed residual includes:

[0054] The sea clutter image acquisition module acquires the sea clutter image and classifies the acquired sea clutter image.

[0055] The double auto-encoder design and training module designs and trains the double auto-encoder based on the sea clutter image acquired by the sea clutter image acquisition module.

[0056] The multi-domain encoder assembly module respectively performs Fourier transform, ridgelet transform and contourlet transform on the sea clutter image dataset used by the double auto-encoder design and training module, and trains the double auto-encoder in the multi-domain through the double auto-encoder designed by the double auto-encoder design and training module to obtain the multi-domain encoder.

[0057] The reconstructed residual calculation module calculates the reconstructed residual of each sample picture in the multi-domain through the multi-domain encoder obtained by the multi-domain encoder assembly module.

[0058] The SVM supervised learning module obtains the optimal expression in the target echo detection through SVM supervised learning.

[0059] The target judgment module judges whether there is a target based on the reconstructed residual calculated by the reconstructed residual calculation module and the optimal expression obtained by the SVM supervised learning module.

[0060] The target echo detection device based on multi-domain reconstructed residual includes:

[0061] The memory is used to store the computer program for implementing the target echo detection method based on multi-domain reconstructed residual.

[0062] The processor is used to execute the computer program to implement the target echo detection method based on multi-domain reconstructed residual.

[0063] A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the target echo detection method based on multi-domain reconstructed residual.

[0064] The beneficial effects of the present application are that, compared with the prior art:

[0065] 1、The present application learns the current sea clutter structural features through the constructed double auto-encoder, reconstructs the sea clutter features while reconstructing the image, and distinguishes the non-cooperative target with unknown features from the sea clutter based on the target detection method of the reconstruction residual error.

[0066] 2、The present application takes the cooperative target as the positive sample in the SVM supervised learning, and is not the direct learning of the target features, but the learning of the difference amount of the target and sea clutter features, so that the requirement for the positive sample is low.

[0067] In summary, the present application learns the current sea clutter structural features through the constructed double auto-encoder, and learns the difference amount of the target and sea clutter features through the SVM supervised learning, and has the characteristics of high measurement accuracy and strong applicability. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 The method flowchart of the present application.

[0069] Figure 2 The multi-domain double auto-encoder structure diagram of the present application.

[0070] Figure 3 (a) is the no-target reconstruction residual error calculation result diagram of the present application.

[0071] Figure 3 (b) is the target reconstruction residual error calculation result diagram of the present application.

[0072] Figure 4 The target echo detection algorithm principle diagram based on the multi-domain reconstruction residual error of the present application.

[0073] Figure 5 The sea clutter target detection result diagram of the simulation experiment of the present application in different intensity noise environments. DETAILED DESCRIPTION

[0074] The present application will be described in detail below with reference to the drawings.

[0075] Referring to Figure 1 , the target echo detection method based on the multi-domain reconstruction residual error comprises the following steps:

[0076] Step 1, collect sea clutter images, and classify the collected sea clutter images to obtain a sea clutter image dataset;

[0077] Step 2, design a double autoencoder, and train the double autoencoder based on the sea clutter image dataset obtained in step 1;

[0078] Step 3, Fourier transform, ridgelet transform and contourlet transform are performed on the sea clutter image dataset used in step 2, and the double autoencoder in multiple domains is trained through the double autoencoder designed in step 2, and the multi-domain encoder is obtained after training;

[0079] Step 4, input the sea clutter image dataset obtained in step 1 into the multi-domain encoder obtained in step 3, and calculate the reconstruction residual of each collected sea clutter image in the multi-domain in the sea clutter image dataset;

[0080] Step 5, based on the reconstruction residual calculated in step 4, the optimal expression of target echo detection is obtained through SVM supervised learning;

[0081] Step 6, calculate the reconstruction residual of the sea clutter image sample of the detection area in the multi-domain through step 4, and judge whether there is a target based on the optimal expression obtained in step 5 to complete target detection.

[0082] The step 1 specifically comprises the following steps:

[0083] Step 1.1, collect sea clutter images through a networked radar;

[0084] Step 1.2, the sea clutter images collected in step 1.1 are divided into two categories according to whether there is a target: general sea clutter images without target and sea clutter images with cooperative target.

[0085] The step 2 specifically comprises the following steps:

[0086] Step 2.1, design a double autoencoder:

[0087] The double autoencoder is a kind of artificial neural network mainly used in unsupervised learning, which has two structures of encoder and decoder, and is composed of convolutional layer, pooling layer and the like. Its function is to perform representation learning on input information by taking the input information as a learning target, and output a reconstructed image of the original image. The double autoencoder model is designed based on the traditional autoencoder in the application, and the structure comprises two encoders and a decoder, and the function is to obtain the reconstructed image and the reconstructed feature of the input image;

[0088] The double autoencoder designed in the application first performs feature extraction on the original sea clutter image p 11 by the first encoder E11 to obtain the image feature f 11 after dimensionality reduction and compression of the original sea clutter image p11 Then the decoder D1 processes the image features f 11 Decoding yields the reconstructed image p 12 After obtaining the reconstructed image, the second encoder E12 processes the reconstructed image p. 12 Feature extraction is performed to obtain the reconstructed image p. 12 Reconstruction features f 12 ;

[0089] Step 2.2: Input the sea clutter image dataset into the dual autoencoder designed in Step 2.1 to perform unsupervised learning training on the dual autoencoder:

[0090] Using 30-50% of general sea clutter images without targets as sea clutter image data for training the dual autoencoder, the unsupervised learning training principle of the dual autoencoder can be expressed by the following formula:

[0091]

[0092] Where, p 11 For the input raw sea clutter image, p 12 To reconstruct the image, f 11 For the first encoder E11, the raw sea clutter image p 11 Extracted image features, f 12 For the second encoder E12 to reconstruct the image p 12 The extracted reconstructed features are α and β, which are user-defined parameters.

[0093] See Figure 2 Step 3 specifically includes the following steps:

[0094] The purpose of establishing a multi-domain encoder is to simultaneously establish multiple dual autoencoders and train the sea clutter encoder using sea clutter image data in multiple transform domains. The aim is to integrate the detection results in multiple transform domains and improve the detection accuracy of the target to be detected. This invention proposes to use three domain transformations—Fourier transform, ridge transform, and contour wave transform—to process the original image and train the corresponding dual autoencoders to fully learn the sea clutter echo features.

[0095] The original sea clutter image p from the 30-50% target-free general sea clutter image used in step 2.2 11 Fourier transform, ridge transform, and contour wave transform were performed respectively, with the original sea clutter image p 11 The image sample obtained by Fourier transform is p 21 , will p 21 The data sample set is used as the training data for the multi-domain encoder to train the corresponding dual autoencoder in the transform domain:

[0096]

[0097] The original sea clutter image p 11 The image sample obtained after the ridgelet transform is p 31 p is input into the multi-domain encoder trained in step 1.1 31 The data sample set trained as the multi-domain encoder is used to train the double auto-encoder in the corresponding transform domain:

[0098]

[0099] The original sea clutter image p 11 The image sample obtained after the contourlet transform is p 41 p is input into the multi-domain encoder trained in step 1.1 41 The data sample set trained as the multi-domain encoder is used to train the double auto-encoder in the corresponding transform domain:

[0100]

[0101] The multi-domain encoder obtained after the training is:

[0102]

[0103]

[0104]

[0105]

[0106] The sea clutter image and its features can be reconstructed in the multi-domain.

[0107] The step 4 specifically comprises the following steps:

[0108] The trained double auto-encoder is used as the reconstructor of the sea clutter image, and the detection image can be encoded and decoded to obtain the reconstructed image of the approximate general sea clutter image. The reconstruction residual represents the difference between the to-be-detected image and the reconstructed image itself and its features, and can be used to judge whether there is a target echo in the to-be-detected region.

[0109] The sea clutter image containing the cooperative target and the general sea clutter image without the target remaining in step 2.2 are input into the multi-domain encoder obtained in step 3, and the reconstruction residual of the sea clutter image in the selected data set is calculated, and the calculation formula is as follows:

[0110]

[0111]

[0112]

[0113]

[0114] The reconstruction residual in the multi-domain is obtained, wherein d1 is the reconstruction residual of the original sea clutter image p in the selected sea clutter image data set, d2 is the reconstruction residual of the image p obtained after Fourier transform of the original sea clutter image, d3 is the reconstruction residual of the image p obtained after ridgelet transform of the original sea clutter image, and d4 is the reconstruction residual of the image p obtained after contourlet transform of the original sea clutter image. 11 21 31 41

[0115] Referring to Figure 3 (a), if the reconstruction residual of the to-be-inspected region is small, it indicates that the feature of the to-be-inspected region conforms to the feature of the general sea clutter, and it is considered that no target exists. Conversely, referring to Figure 3 (b), when the to-be-inspected region contains a target, the reconstruction residual of the reconstructed image will be significantly increased, and the target can be obviously distinguished from the adjacent region.

[0116] Referring to Figure 4 , the step 5 specifically comprises the following steps:

[0117] Step 5.1, taking the reconstruction residuals {d1, d2, d3, d4} of the sea clutter images in the multi-domain in the training data set as the learning features of the SVM supervised learning;

[0118] Step 5.2, taking the sea clutter images containing cooperative targets in the training data set as positive samples and marking the labels y=1, and taking the general sea clutter images without targets as negative samples and marking the labels y=-1, as the sample labels of the SVM supervised learning;

[0119] Step 5.3, based on the learning features of step 5.1 and the sample labels of step 5.2, the optimal expression for target echo detection is obtained through SVM supervised learning:

[0120] min(y×(γ1d1+γ2d2+γ3d3+γ4d4),

[0121] The maximum discrimination expression is obtained through SVM supervised learning:

[0122] ε=γ1d1+γ2d2+γ3d3+γ4d4+η,

[0123] The parameter values of γ1, γ2, γ3, γ4 and η in the optimal expression for target echo detection are obtained through training. Taking the cooperative target as a positive sample is to learn the difference between the target and the sea clutter, and thus the requirement for the positive sample is low. Even if the features of the cooperative target and the non-cooperative target are different, the detector can also obtain good detection effect through the reconstruction residual.

[0124] The step 6 specifically comprises the following steps:​​​​

[0125] The multi-domain reconstruction is performed on the to-be-detected region to obtain reconstructed residual errors {d1, d2, d3, d4} of the to-be-detected region, and a maximum discrimination expression ε = γ1d1 + γ2d2 + γ3d3 + γ4d4 + η is used for judgment: when ε > 0, it is considered that the target exists, at this time, position and speed information of all target point trails detected are extracted, target detection calculation based on signal echo characteristics and reconstructed residual errors is completed, and a target detection result is obtained; when ε < 0, it is considered that the to-be-detected region is sea clutter without a target.

[0126] Simulation experiment

[0127] Simulation experiment 1

[0128] In order to detect the noise suppression ability of the target echo detection method based on the multi-domain reconstruction residual error of the application, that is, the detection effect of the sea clutter target in a high-noise environment, in this simulation experiment, a sea clutter image data set containing a cooperative target in a sea area is collected in different noise environments, and the target echo detection system based on the multi-domain reconstruction residual error trained in the application is used for target detection, the reconstructed image of the sea clutter image containing the target is obtained, and the position and speed information of the detected target point trail are extracted.

[0129] As Figure 5 Fig. 1 shows the detection effect of the target echo detection method based on the multi-domain reconstruction residual error of the application on the target in the sea clutter image under different noise conditions. In the figure, groups 1-6 are set as different noise intensity environments from small to large, the horizontal coordinate is the azimuth angle of the target in the polar coordinate, the vertical coordinate is the distance of the target in the polar coordinate, three different color point trails represent the motion trajectories of three different targets detected, and the right image is the reconstructed image of the sea clutter image containing the target in different noise environments. Through the restoration of the target motion trajectory and the reconstruction result of the sea clutter image containing the target, it can be found that complete target point trail information can be detected in a high-noise environment, and the position of the detected target point trail is basically consistent with that in a low-noise environment, which shows that the target can be detected with high precision and the target point trail information can be obtained in a high-noise environment, and it can be verified that the noise suppression effect of the target echo detection method based on the multi-domain reconstruction residual error of the application is good, and the problem that the target echo is weakly affected by noise and difficult to detect is effectively solved.

[0130] Simulation experiment 2

[0131] To verify the target echo detection method based on multi-domain reconstruction residual of the present application compared with the existing method, the target detection precision in sea clutter is improved, the present method and a variety of CFAR (i.e. constant false alarm rate) detection methods mainly used for sea clutter target detection are compared. Based on the sea clutter image data set collected in simulation experiment 1 under different noise environments, the target echo detection method based on multi-domain reconstruction residual of the present application and the variety of CFAR detection methods are used to detect the target in sea clutter, and the target track information detected is extracted, the position error of the target track detected and the real motion track of the target is calculated, and the detection results are shown in Table 1:

[0132] The OSPA distance of each detection method for detecting sea clutter target

[0133]

[0134] Table 1

[0135] Among them, the 1-6 experimental groups are the same as simulation experiment 1, indicating 6 different noise intensity environments. The data in the table is the OSPA distance of the selected detection method for detecting sea clutter target in the 6 experimental groups, which reflects the average error of the detected target track and its real motion track. The lower the value, the higher the precision of the target position information detection. It can be seen that in each group of experiments, the OSPA distance value of the target echo detection method based on multi-domain reconstruction residual of the present application for detecting sea clutter target is obviously lower than that of the traditional CFAR algorithm, and the position detection of the target in sea clutter has higher precision.

[0136] The target echo detection system based on multi-domain reconstruction residual comprises:

[0137] The sea clutter image acquisition module acquires sea clutter images and classifies the acquired sea clutter images, and is used to realize step 1 of the target echo detection method based on multi-domain reconstruction residual of the present application;

[0138] The double autoencoder design and training module designs and trains a double autoencoder based on the sea clutter images acquired by the sea clutter image acquisition module, and is used to realize step 2 of the target echo detection method based on multi-domain reconstruction residual of the present application;

[0139] The multi-domain encoder assembly module respectively performs Fourier transform, ridgelet transform and contourlet transform on the sea clutter image data set used by the double autoencoder design and training module, and trains the double autoencoder in the multi-domain through the double autoencoder designed by the double autoencoder design and training module to obtain a multi-domain encoder, which is used to realize step 3 of the target echo detection method based on multi-domain reconstruction residual of the present application;

[0140] reconstruction residual calculation module: calculating the reconstruction residual of each sample picture in the multi-domain through the multi-domain encoder calculation data obtained by the multi-domain encoder assembly module, used to realize step 4 of the target echo detection method based on multi-domain reconstruction residual;

[0141] SVM supervised learning module: obtaining the optimal expression in target echo detection through SVM supervised learning, used to realize step 5 of the target echo detection method based on multi-domain reconstruction residual;

[0142] target judgment module: judging whether there is a target through the reconstruction residual calculated by the reconstruction residual calculation module and the optimal expression obtained by the SVM supervised learning module, used to realize step 6 of the target echo detection method based on multi-domain reconstruction residual.

[0143] The target echo detection device based on multi-domain reconstruction residual comprises:

[0144] memory: used to store the computer program for realizing the target echo detection method based on multi-domain reconstruction residual;

[0145] processor: used to realize the target echo detection method based on multi-domain reconstruction residual when executing the computer program.

[0146] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the target echo detection device based on multi-domain reconstruction residual, and connects each part of the target echo detection device based on multi-domain reconstruction residual through various interfaces and lines.

[0147] The processor implements the steps of the above-mentioned target echo detection method based on multi-domain reconstruction residual when executing the computer program, for example: collecting sea clutter images and classifying; designing and training a double autoencoder; training the double autoencoder to obtain a multi-domain encoder; calculating reconstruction residual through the multi-domain encoder; obtaining the optimal expression in target echo detection based on the reconstruction residual and SVM supervised learning; judging whether there is a target through the reconstruction residual and the optimal expression; and implementing the target echo detection method based on multi-domain reconstruction residual.

[0148] Alternatively, the processor implements the functions of the modules in the above-mentioned system when executing the computer program, for example: a sea clutter image acquisition module: acquiring sea clutter images and classifying the acquired sea clutter images; a double autoencoder design and training module: designing and training a double autoencoder; a multi-domain encoder assembly module: respectively performing Fourier transform, ridgelet transform, and contourlet transform on a sea clutter image dataset; a reconstruction residual calculation module: calculating the reconstruction residual of each sample picture in the multi-domain through the multi-domain encoder; and a SVM supervised learning module: obtaining the optimal expression in target echo detection through SVM supervised learning. The result of the target echo detection method based on multi-domain reconstruction residual is output.

[0149] Illustratively, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a preset function, which are used to describe the execution process of the computer program in the device of the target echo detection method based on multi-domain reconstruction residual. For example, the computer program can be divided into a sea clutter image acquisition module; a double autoencoder design and training module; a multi-domain encoder assembly module; a reconstruction residual calculation module; and a SVM supervised learning module. The specific functions of each module are as follows: the sea clutter image acquisition module: acquiring sea clutter images and classifying the acquired sea clutter images; the double autoencoder design and training module: designing and training a double autoencoder; the multi-domain encoder assembly module: respectively performing Fourier transform, ridgelet transform, and contourlet transform on a sea clutter image dataset; the reconstruction residual calculation module: calculating the reconstruction residual of each sample picture in the multi-domain through the multi-domain encoder; and the SVM supervised learning module: obtaining the optimal expression in target echo detection through SVM supervised learning, and outputting the result of the target echo detection method based on multi-domain reconstruction residual.

[0150] The device based on the target echo detection of multi-domain reconstruction residual can be a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The device based on the target echo detection of multi-domain reconstruction residual can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above is an example of the device based on the target echo detection of multi-domain reconstruction residual, and does not constitute a limitation on the device based on the target echo detection of multi-domain reconstruction residual, and can include more components, or combine certain components, or different components, for example, the device based on the target echo detection of multi-domain reconstruction residual can also include an input and output device, a network access device, a bus and the like.

[0151] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the device based on the target echo detection of multi-domain reconstruction residual by running or executing the computer program and / or modules stored in the memory, and calling the data stored in the memory.

[0152] The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playing function, an image playing function and the like) and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book and the like) and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0153] The application further provides a computer readable storage medium, characterized by storing a computer program, wherein the computer program is executed by a processor to realize the steps of the method for target echo detection based on multi-domain reconstruction residual.

[0154] The modules / units of the system based on the target echo detection of multi-domain reconstruction residual are stored in a computer readable storage medium if they are realized in the form of software function units and sold or used as independent products.

[0155] The application realizes all or part of the processes in the target echo detection method based on multi-domain residual reconstruction, and can also be completed by a computer program instructing related hardware. The computer program can be stored in a computer readable storage medium, and the computer program can realize the steps of the target echo detection method based on multi-domain residual reconstruction when executed by a processor. The computer program includes computer program codes, which can be in the form of source code, object code, executable files or preset intermediate forms.

[0156] The computer readable storage medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. that can carry the computer program codes.

[0157] It should be noted that the contents contained in the computer readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer readable storage medium does not include electric carrier signals and telecommunication signals.

[0158] It should be noted that the embodiments of the application can be realized by hardware, software or a combination of software and hardware. The hardware part can be realized by special logic; the software part can be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or a special designed hardware.

[0159] Those skilled in the art can understand that the above-mentioned devices and methods can be realized by computer executable instructions and / or included in processor control codes, for example, such codes are provided on a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware) or a data carrier such as an optical or electronic signal carrier. The devices of the application and their modules can be realized by hardware circuits such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be realized by software executed by various types of processors, and can also be realized by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0160] The above is only a specific embodiment of the application, but the protection scope of the application is not limited thereto, and any modification, equivalent replacement and improvement made by those skilled in the art within the technical range disclosed by the application, as long as it is within the spirit and principle of the application, should be covered within the protection scope of the application.

Claims

1. A method for target echo detection based on multi-domain reconstructed residual, characterized in that, The method comprises the following steps: Step 1, collecting sea clutter images, and classifying the collected sea clutter images to obtain a sea clutter image dataset; Step 2, designing a double autoencoder, and training the double autoencoder based on the sea clutter image dataset obtained in step 1, the specific steps being as follows: Step 2.1, designing a double autoencoder: First, the first encoder E11 processes the raw sea clutter image. Feature extraction is performed to obtain the original sea clutter image. Image features after dimensionality reduction and compression Then the decoder D1 processes the image features. Decoding yields the reconstructed image. After obtaining the reconstructed image, the second encoder E12 processes the reconstructed image. Feature extraction is performed to obtain the reconstructed image. Reconstruction features ; Step 2.2, inputting the sea clutter image dataset into the double autoencoder designed in step 2.1 to perform unsupervised learning training on the double autoencoder: 30-50% of the general sea clutter images without targets are used as the sea clutter image data for training the double autoencoder, and the principle of unsupervised learning training of the double autoencoder can be represented by the following formula: ， wherein, is the input raw sea clutter image, is the reconstructed image, is the raw sea clutter image as input to the first encoder E11 extracted image features, is the reconstructed image as input to the second encoder E12 extracted reconstructed features, α and β is a self-set parameter; Step 3, performing Fourier transform, ridgelet transform and contourlet transform on the sea clutter image dataset used in step 2, and training the double autoencoder in multiple domains through the double autoencoder designed in step 2 to obtain a multi-domain encoder after the training is completed, the specific steps being as follows: Original sea clutter images from the set of 30-50% of the non-targeted general sea clutter images used in step 2.2 Fourier transform, ridgelet transform, contourlet transform are performed respectively, wherein the original sea clutter image The image samples obtained by Fourier transform are , and The data sample set trained as a multi-domain encoder is used to train the bi-auto-encoder in the corresponding transform domain: Original sea clutter image The image sample obtained after the ridgelet transform is , and As a data sample set for training the multi-domain encoder, the double auto-encoder in the corresponding transform domain is trained: Original sea clutter image The image sample obtained after the contourlet transform is , and the image sample obtained after the contourlet transform is As a data sample set for training the multi-domain encoder, the double auto-encoder in the corresponding transform domain is trained: After the training is completed, a multi-domain encoder is obtained: Step 4, inputting the sea clutter image dataset obtained in step 1 into the multi-domain encoder obtained in step 3 to calculate the reconstruction residual of each collected sea clutter image in the sea clutter image dataset in the multiple domains; Step 5, obtaining the optimal expression for target echo detection through SVM supervised learning based on the reconstruction residual calculated in step 4; Step 6, calculating the reconstruction residual of the sea clutter image sample of the to-be-detected region in the multiple domains through step 4, and judging whether there is a target based on the optimal expression obtained in step 5 to complete target detection.

2. The multi-domain reconstruction residual based target echo detection method of claim 1, wherein, The step 1 specifically comprises the following steps: Step 1.1, collecting sea clutter images through a networked radar; Step 1.2, dividing the sea clutter images collected in step 1.1 into two categories according to whether there is a target: general sea clutter images without targets and sea clutter images with cooperative targets.

3. The multi-domain reconstruction residual based target echo detection method of claim 1, wherein, The step 4 specifically comprises the following steps: The sea clutter images with cooperative targets and the remaining general sea clutter images without targets in step 2.2 are input into the multi-domain encoder obtained in step 3 to calculate the reconstruction residual of the sea clutter images in the selected dataset, and the calculation formula is as follows: reconstruction residuals in multiple domains , , , } wherein is the reconstruction residual of the original sea clutter image in the selected sea clutter image data set, is the reconstruction residual of the image after Fourier transformation of the original sea clutter image, is the reconstruction residual of the image after ridgelet transformation of the original sea clutter image, is the reconstruction residual of the image after contourlet transformation of the original sea clutter image.

4. The multi-domain reconstruction residual based target echo detection method of claim 1, wherein, The step 5 specifically comprises the following steps: Step 5.1, Reconstruction residuals of sea clutter images in the training dataset in multiple domains as learning features of SVM supervised learning , , , } Step 5.2, Labeling the sea clutter images with cooperative targets in the training dataset as positive samples y = 1, general sea clutter images without targets are labeled as negative samples y = -1, sample labels for SVM supervised learning; Step 5.3, obtaining the optimal expression for target echo detection through SVM supervised learning based on the learning features of step 5.1 and the sample labels of step 5.2: ， The maximum discrimination expression can be obtained through SVM supervised learning: ， The target echo detection optimal expression is obtained by training parameter values.

5. The multi-domain reconstruction residual based target echo detection method of claim 4, wherein, The step 6 specifically comprises the following steps: The multi-domain reconstruction is performed on the to-be-detected region to obtain a reconstructed residual error of the to-be-detected region , , , } based on a maximum discrimination expression is performed: when ε>0, it is considered that the target exists, at this time, position and speed information of all target tracks detected are extracted, target detection calculation based on signal echo characteristics and reconstructed residual error is completed, and a target detection result is obtained; when ε<0, it is considered that the to-be-detected region is sea clutter without a target.

6. A target echo detection system based on multi-domain reconstructed residual error, employing the method according to any one of claims 1 to 5, characterized in that It comprises: a sea clutter image collection module for collecting sea clutter images and classifying the collected sea clutter images; a double autoencoder design and training module for designing and training a double autoencoder based on the sea clutter images collected by the sea clutter image collection module; a multi-domain encoder assembly module for performing Fourier transform, ridgelet transform and contourlet transform on the sea clutter image dataset used by the double autoencoder design and training module, and training the double autoencoder in multiple domains through the double autoencoder designed by the double autoencoder design and training module to obtain a multi-domain encoder; a reconstruction residual calculation module for calculating the reconstruction residual of each sample picture in the dataset in the multiple domains through the multi-domain encoder obtained by the multi-domain encoder assembly module; a reconstruction residual calculation module for calculating the reconstruction residual of each sample picture in the dataset in the multiple domains through the multi-domain encoder obtained by the multi-domain encoder assembly module; SVM supervised learning module: obtaining optimal expression in target echo detection through SVM supervised learning; Target judgment module: judging whether there is a target through the reconstructed residual calculated by the reconstructed residual calculation module and the optimal expression obtained by the SVM supervised learning module.

7. A target echo detection device based on multi-domain reconstructed residual error, characterized in that, It comprises: Memory: used for storing a computer program for implementing the target echo detection method based on multi-domain reconstructed residual as claimed in any one of claims 1-5; Processor: used for implementing the target echo detection method based on multi-domain reconstructed residual as claimed in any one of claims 1-5 when the computer program is executed.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the target echo detection method based on multi-domain reconstructed residual as claimed in any one of claims 1-5.

Images