CN114492744A - Method for generating ground-sea clutter spectrum data sample based on confrontation generation network - Google Patents
Method for generating ground-sea clutter spectrum data sample based on confrontation generation network Download PDFInfo
- Publication number
- CN114492744A CN114492744A CN202210054076.7A CN202210054076A CN114492744A CN 114492744 A CN114492744 A CN 114492744A CN 202210054076 A CN202210054076 A CN 202210054076A CN 114492744 A CN114492744 A CN 114492744A
- Authority
- CN
- China
- Prior art keywords
- data
- ground
- sea clutter
- generator
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a method for generating a sample of ground-sea clutter spectrum data based on an antagonistic generation network, which comprises the following steps: s1, constructing a data set and acquiring random noise according to the distribution characteristics of the sky wave radar ground and sea clutter data; s2, constructing a generative confrontation network based on the distribution characteristics of sky wave radar ground sea clutter data, generating mass sky wave radar ground sea clutter data through the generative confrontation network, a pre-obtained data set and random noise, and forming a preset data set; the generative confrontation network is constructed based on convolution and deconvolution networks, and is trained by a deep learning method and a preset data set. The method solves the problem that the existing sky wave radar ground sea clutter spectrum data labeling and data generating method is low in efficiency.
Description
Technical Field
The invention belongs to the technical field of deep learning, and particularly relates to a method for generating a ground sea clutter spectrum data sample based on a confrontation generating network.
Background
Because the ionosphere environment detection subsystem and the sky wave main radar have different working modes, detection channels and the like, the problems that coordinate registration parameters provided by the ionosphere detection subsystem are inaccurate and inconsistent with main radar target parameters exist, and the like, so that the sky wave radar target positioning error is large.
The sky wave radar sea clutter spectrum data identification technology based on deep learning needs massive training samples, a calibrated sample training model needs to be used in a traditional supervised learning method, and accurate calibration of sky wave radar sea clutter spectrum data directly influences the quality of land-sea boundary identification effect. The sky wave radar ground sea clutter spectrum data are easy to obtain in a large amount, but manual labeling is time-consuming and labor-consuming, and the model requirement cannot be met, so that the method has practical engineering significance on how to deal with the problem of small-scale data. The traditional method for dealing with the small-scale data set training problem is data enhancement, new data is generated from limited data in a synthesis or conversion mode, and the data enhancement technology is always an important means for overcoming the data shortage. The data enhancement method based on the data geometric transformation can relieve the overfitting problem of the neural network to a certain extent and improve the generalization capability of the model. But compared with the original data, the problem of insufficient data is not fundamentally solved by the added data points; meanwhile, the data enhancement mode needs to manually set a conversion function and corresponding parameters, generally, the optimal data enhancement method is difficult to realize by virtue of empirical knowledge, and the generalization performance of the model can only be improved in a limited way. The identification problem of sky wave radar ground sea clutter spectrum data needs massive support with marked spectrum data; however, due to the characteristics of strong interference noise, complex feature transformation form and the like of sky wave radar sea clutter spectrum data, the data is difficult to accurately label by the traditional semi-supervision method, and the time and the labor are consumed for manually calibrating massive sea clutter spectrum data, so that the training difficulty of the model is greatly increased; the sky wave radar ground and sea clutter spectrum data are various in conditions and complex in characteristics, and when the sky wave radar ground and sea clutter spectrum data are marked manually, the ideal ground and sea clutter spectrum data tend to be selected, however, the ideal ground and sea clutter spectrum data are single in type, and the classification model cannot learn abundant characteristics, so that the generalization capability of the model is weak. Therefore, the practical engineering significance is achieved on how to accurately and efficiently finish the labeling of mass geodetic clutter spectrum data and the data generation.
Disclosure of Invention
The invention aims to provide a method for generating a sample of the ground-sea clutter spectrum data based on a countermeasure generation network, which aims to solve the problem of low efficiency of the existing method for labeling and generating the ground-sea clutter spectrum data of a sky wave radar.
The invention adopts the following technical scheme: a method for generating a ground-sea clutter spectrum data sample based on a countermeasure generation network comprises the following steps:
s1, constructing a data set and acquiring random noise according to the distribution characteristics of the sky wave radar ground sea clutter data;
s2, constructing a generative confrontation network based on the distribution characteristics of sky wave radar ground sea clutter data, generating mass sky wave radar ground sea clutter data through the generative confrontation network, a pre-obtained data set and random noise, and forming a preset data set; the generative confrontation network is constructed based on convolution and deconvolution networks, and is trained by a deep learning method and a preset data set.
Further, the generative confrontation network includes a generator and an arbiter:
the generator outputs generated sample data according to the real ground-sea clutter data;
the discriminator obtains the true probability of the generated sample data according to the true ground-sea clutter data and the generated sample data.
Further, the step of performing network training on the generated confrontation network by using the deep learning method and the preset data set specifically comprises:
step S11, constructing a small number of ground sea clutter training samples, wherein the ground sea clutter training samples comprise the same number of ground sea clutter samples with different characteristics;
step S12, training the generator by using a deep learning method and a preset data set, and training a discriminator in the generative confrontation network according to the ground-sea clutter data and the generated data of the generator;
step S13: the steps S11 to S12 are repeatedly performed and the training is stopped when the generated data of the generator satisfies the preset data requirement.
Further, the step of training the generator by using the deep learning method and the preset data set specifically comprises:
step S121, acquiring random noise with the same quantity as the data set;
s122, obtaining generated sample data through the generator by random noise, calculating the true probability of the generated sample by using a discriminator, and carrying out network training on the generator by using the true probability;
and step S123, repeatedly executing the step S121 to the step S122 and stopping training when the generated data of the generator meets the preset data requirement.
Further, the content of training the discriminator in the generative countermeasure network according to the ground-sea clutter data and the generated data of the generator is specifically as follows:
constructing a data set according to the ground-sea clutter data and the generated data; training a discriminator by utilizing a deep learning algorithm according to the ground-sea clutter data and the generated data; wherein the generation data and the ground sea clutter data are equal in number.
Further, the step of generating massive sky wave radar ground sea clutter data through a generating type countermeasure network, a pre-obtained data set and random noise specifically comprises the following steps:
s21, obtaining the number of random noises according to the number of samples required to be generated;
s22, generating ground-sea clutter samples of a target number by using the trained generator and random noise;
and S23, storing the generated ground sea clutter samples.
The invention has the beneficial effects that: the method comprises the steps of modeling the real distribution of the ground-sea clutter aiming at the problems of complex characteristics, single sample, difficult calibration and the like of the ground-sea clutter data of the sky-wave radar, establishing a mapping relation from one-dimensional noise to the real data based on a generated countermeasure network, and constructing a generated countermeasure model based on a convolutional neural network. The method can utilize a small amount of ground-sea clutter data to generate massive data, improves the diversity of the small amount of data, overcomes the problem that the existing deep learning method needs massive data, greatly reduces the labor and time cost in the label calibration process, and has the advantages of easy training, convenient migration and the like.
Drawings
FIG. 1 is a diagram of sky wave radar sea clutter generation data in an embodiment of the present invention;
FIG. 2 is a diagram of sky wave radar sea clutter real data in an embodiment of the present invention;
FIG. 3 is a diagram of sky wave radar ground clutter generation data in an embodiment of the present invention;
FIG. 4 is a diagram of ground clutter real data of a sky-wave radar in an embodiment of the present invention;
FIG. 5 is a flow chart of the generation of the sky wave radar ground sea clutter according to the embodiment of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention provides a method for generating a sample of ground sea clutter spectrum data based on a countermeasure generation network, which comprises the following steps:
s1, constructing a data set and acquiring random noise according to the distribution characteristics of the sky wave radar ground sea clutter data;
s2, constructing a generative confrontation network based on the distribution characteristics of sky wave radar ground sea clutter data, generating mass sky wave radar ground sea clutter data through the generative confrontation network, a pre-obtained data set and random noise, and forming a preset data set; the generative confrontation network is constructed based on convolution and deconvolution networks, and is trained by a deep learning method and the preset data set.
In some embodiments, the generative confrontation network comprises a generator and an arbiter:
the generator outputs generated sample data according to the real ground-sea clutter data;
the discriminator obtains the true probability of generating the sample data according to the true ground-sea clutter data and the generated sample data.
In some embodiments, the step of performing network training on the generated confrontation network by using a deep learning method and a preset data set specifically includes:
step S11, constructing a small number of ground sea clutter training samples, wherein the ground sea clutter training samples comprise the same number of ground sea clutter samples with different characteristics;
step S12, training the generator by using a deep learning method and a preset data set, and training a discriminator in the generative confrontation network according to the ground-sea clutter data and the generated data of the generator;
step S13: and repeatedly executing the steps S11 to S12 and stopping training when the generated data of the generator meet the preset data requirement.
In some embodiments, the step of training the generator by using the deep learning method and the preset data set specifically includes:
step S121, acquiring random noise with the same quantity as the data set;
s122, obtaining generated sample data through a generator by random noise, calculating the true probability of the generated sample by using a discriminator, and carrying out network training on the generator by using the true probability;
and step S123, repeatedly executing the step S121 to the step S122 and stopping training when the generated data of the generator meets the preset data requirement.
In some embodiments, the training of the arbiter in the generative countermeasure network according to the ground-sea clutter data and the generated data of the generator is specifically:
constructing a data set according to the ground-sea clutter data and the generated data; training a discriminator by utilizing the deep learning algorithm according to the ground-sea clutter data and the generated data; wherein the generation data and the ground sea clutter data are equal in number.
In some embodiments, the step of generating the mass sky wave radar ground sea clutter data by the generative countermeasure network, the pre-obtained data set, and the random noise specifically includes:
s21, obtaining the number of random noises according to the number of samples required to be generated;
s22, generating a ground-sea clutter sample with a target number by using the trained generator and random noise;
and S23, storing the generated ground sea clutter samples.
Examples
1. Constructing a generator based on a convolutional neural network, comprising the following steps:
the generator G continuously learns the probability distribution of real data in a training set, the target is to convert input random noise into sky wave radar ground sea clutter spectrum data which can be false or spurious, capture data distribution from the real sky wave radar ground sea clutter spectrum data and map the data distribution to a certain new data space, output the generated data and record the data as G (z), the distribution of the data is recorded as pg (z), and the data looks the same as the sample distribution pr (x) in the training set as much as possible.
The generator is represented by a differentiable function G, the input z is a random variable or a random variable in hidden space, typically a gaussian variable or noise, and the generator G generates a pseudo sample distribution G (z). The generator G network requires input of a constraint condition, and for an input variable, the constraint condition can be input into the first layer or the last layer; noise can also be added to the hidden layer by summing, multiplying or splicing. GAN has no restriction on the dimension of the input variable z, which is a random encoding vector of 100 dimensions.
A deep convolutional neural network is adopted as a basic framework structure of a generator. Eight layers of transposition convolution are used as a basic network structure of the classifier, the multi-stage transposition convolution layer is used for improving the Feature vector dimension and increasing Feature Map, and a LeakyReLU function is used as an activation function in each layer of transposition convolution; and finally, taking a Tanh function as an output layer.
2. The construction of the discriminator based on the convolutional neural network comprises the following substeps:
the input of the discriminator D comprises real data x and generated data G (z), the output is a probability value or a scalar value, which represents the probability that the discriminator D considers that the input is real distribution, the larger the numerical value is, the larger the probability is of the real data, otherwise, the input is considered to be a generated sample. The main goal of discriminator D is to determine if the input is a true sample and provide a feedback mechanism that forms a game of chance with the generation network. This game consists of two scenarios, in the first scenario x is sampled from the real training data as input to the discriminator D, which outputs a number between 0 and 1 indicating the probability that x belongs to a real sample. In the second scenario, the variable z is sampled from an a priori distribution, and G (z) is used as input to arbiter D, which targets output D (G (z)) to approach 0, and generator G targets output D to approach 1.
And a deep convolutional neural network is adopted as a basic framework structure of the discriminator. The method comprises the steps that an input layer, a convolution layer, an … … output layer are used as a basic network structure of a classifier, a multi-stage convolution layer is used for fully extracting multi-level features, a LeakyReLU function is used as an activation function in each convolution layer, a convolution operation with the step length of 2 is used for replacing a pooling layer, the feature vector dimension is reduced, the calculated amount is reduced, overfitting is corrected, and finally the high-level features are extracted; sigmoid function and full connectivity layer are used for classification.
3. The method for realizing the confrontation training of the generator and the discriminator comprises the following steps:
the core of the generative confrontation network is the confrontation training of a generator and a discriminator, wherein the generator continuously generates false samples approaching to the distribution of real data to deceive the discriminator in the training process, and the discriminator distinguishes the generated false samples from the real samples in the learning process.
(3.1) network training
Firstly, normalization processing is carried out on real data, the maximum value max (x) of the real data is 98.58, the minimum value min (x) is-54.12, and the data are mapped between [ -1,1] through the following formula:
let the random noise be a vector z in dimensions 100 x 1, the function of the generator be G, its input be z, and the parameter be θ(G)(ii) a The function of the discriminator is D, its input is x, and the parameter is theta(D). D needs to update theta(D)Minimization of J(D)(θ(D),θ(G)) G needs to be updated by(G)Minimization of J(G)(θ(D),θ(G)) Two networks play games with each other, the loss functions of the two networks depend on each other, and the Nash equilibrium refers to a pair of parameters (theta) through reaching a Nash equilibrium(D),θ(G)) So that theta(D)Is J(D)A minimum value point of (a) and [ theta ] at the same time(G)Is J(G)One minimum value of (a). The optimization of GAN is actually a very small maximization problem, whose objective function is defined as:
from the above formula, we have the basic steps to generate a challenge: firstly, identifying real data by using a discriminator, calculating a loss function of the discriminator to enable a result to be as close to 1 as possible, and performing reverse propagation; then, discriminating false data by using a discriminator, calculating a loss function of the discriminator to enable the result to be as close to 1 as possible, and performing reverse propagation; and finally, generating false data by using a generator, identifying the false data by using a discriminator, calculating a loss function of the generator to enable the result to be as close to 0 as possible, and performing back propagation.
(3.2) evaluation of generated data
We want to generate data to guarantee diversity, by approximating the distribution of real data. Therefore, from the aspect of feature statistics of the original data, the similarity of two groups of data images is measured, and the distance between the feature vectors of the real data and the generated data is calculated.
Firstly, 2048-dimensional feature vectors of generated data and real data are extracted through a pre-trained convolutional neural network model, the 2048-dimensional feature vectors of the real data obey a distribution A, and the feature vectors of the generated data also obey a distribution B. The distance between the two distributions represents the difference between the two data. The most intuitive characteristics of the distributions are mean and variance, so the distance between two distributions is calculated using the mean and covariance matrices, yielding the data evaluation formula as follows:
where x represents true data, g is generated data, μxIs the mean, mu, of the feature vectors of the real datagIs the mean, Σ, of the feature vector of the generated dataxIs the covariance matrix, Σ, of the true data eigenvectorsgIs the covariance matrix that generates the data eigenvectors, Tr represents the trace of the matrix (sum of the principal diagonal elements).
The FID represents the distance between the feature vector of the generated image and the feature vector of the real image, and the closer the distance is, the better the model is generated, i.e. the image has high definition and rich diversity. Training may be stopped when the FID is less than the threshold we set to 20. The final sea clutter generation sample FID score is 14.03 and the ground clutter generation sample FID score is 17.69.
To verify the effectiveness of the present invention, the following tests were performed experimentally. And (3) taking the existing small amount of the geodetic clutter data as the input of the generative countermeasure network according to the steps, and performing iterative training on the network. After the training is finished, 2 100-dimensional random vectors are used as the input of a generator, the generator generates 2 pieces of ground sea clutter data, and the data obtained by the generator and the real ground sea clutter data are drawn to obtain the images 1, 2, 3 and 4. Fig. 1 is a data diagram of sky wave radar sea clutter generation, fig. 2 is a data diagram of sky wave radar sea clutter real data, fig. 3 is a data diagram of sky wave radar ground clutter generation, and fig. 4 is a data diagram of sky wave radar ground clutter real data. The value range of the FID evaluation index is larger than 0, and the lower the value is, the closer the distribution of the generated data is to the real data is, and the generation effect is better. The FID value is calculated for the generated ground clutter and the true sea clutter to yield a result of 14.03, and the FID value is calculated for the generated ground clutter and the true ground clutter to yield a result of 17.69. The method can be used for generation of sky wave radar ground sea clutter data. FIG. 5 is a flowchart of the sky-wave radar ground sea clutter generation, random noise z is used as input of a generator, the generator generates a false sample G (z), calculates the FID of the false sample G (z) and real data x, if the FID value is smaller than a threshold value F, training is stopped, the generator is used for generating data, otherwise, the false sample G (z) and the real data x are simultaneously input to a discriminator, the discriminator judges the probability that the sample is real data, and loss is calculated and fed back to the generator and the discriminator to update parameters until the FID value is smaller than the threshold value F. And inputting the N random noises z into a generator to obtain the required mass sky wave radar ground and sea clutter data.
The invention relates to a method for generating a sample of ground-sea clutter spectrum data based on an antagonistic generation network, which is used for modeling the real distribution of ground-sea clutter aiming at the problems of complex ground-sea clutter data characteristics, single sample, difficult calibration and the like of a sky-wave radar, establishing a mapping relation from one-dimensional noise to real data based on the generation of the antagonistic network, and constructing a generated antagonistic model based on a convolutional neural network. The method can utilize a small amount of ground-sea clutter data to generate massive data, improves the diversity of the small amount of data, overcomes the problem that the existing deep learning method needs massive data, greatly reduces the labor and time cost in the label calibration process, and has the advantages of easy training, convenient migration and the like.
The invention provides a new thought for the generation of the sky wave radar ground sea clutter data, and how to accurately and efficiently finish the marking and data generation of a large amount of sea clutter data; improving the data diversity has practical engineering significance. And a small amount of marked sky wave radar ground sea clutter spectrum data is expanded, and the diversity of the sample is improved.
When massive calibrated ground-sea clutter spectrum data are needed in actual engineering, category characteristic information is added into the real data, the generation network and the discrimination network are utilized to mutually game on the basis of a small amount of precisely calibrated ground-sea clutter spectrum data of the sky wave radar, so that the generation network and the discrimination network achieve Nash balance, iterative training of a generator is completed to realize mapping of prior category information to real distribution, then the generator which completes training is used to massively generate diverse ground-sea clutter sample data with rich characteristics, and finally generation of the two types of data is realized.
Claims (6)
1. A method for generating a sample of ground-sea clutter spectrum data based on a countermeasure generation network is characterized by comprising the following steps:
s1, constructing a data set and acquiring random noise according to the distribution characteristics of the sky wave radar ground sea clutter data;
s2, constructing a generative confrontation network based on the distribution characteristics of sky wave radar ground sea clutter data, generating mass sky wave radar ground sea clutter data through the generative confrontation network, a pre-obtained data set and random noise, and forming a preset data set; the generative confrontation network is constructed based on convolution and deconvolution networks, and is trained by a deep learning method and the preset data set.
2. The method of claim 1, wherein the generative confrontation network comprises a generator and an arbiter:
the generator outputs generated sample data according to the real ground-sea clutter data;
the discriminator obtains the true probability of generating the sample data according to the true ground-sea clutter data and the generated sample data.
3. The method for generating the ground-sea clutter spectrum data sample based on the countermeasure generation network according to claim 1 or 2, wherein the step of performing network training on the generative countermeasure network by using a deep learning method and a preset data set specifically comprises:
step S11, constructing a small number of ground sea clutter training samples, wherein the ground sea clutter training samples comprise the same number of ground sea clutter samples with different characteristics;
step S12, training the generator by using a deep learning method and a preset data set, and training a discriminator in the generative confrontation network according to the ground-sea clutter data and the generated data of the generator;
step S13: and repeatedly executing the steps S11 to S12 and stopping training when the generated data of the generator meet the preset data requirement.
4. The method for generating the data sample of the clutter spectrum based on the countermeasure generation network according to claim 3, wherein the training of the generator using the deep learning method and the preset data set specifically comprises:
step S121, acquiring random noise with the same quantity as the data set;
s122, obtaining generated sample data through a generator by random noise, calculating the true probability of the generated sample by using a discriminator, and carrying out network training on the generator by using the true probability;
and step S123, repeatedly executing the step S121 to the step S122 and stopping training when the generated data of the generator meets the preset data requirement.
5. The method for generating the ground-sea clutter spectrum data sample based on the countermeasure generation network according to claim 4, wherein the training of the discriminator in the generative countermeasure network according to the ground-sea clutter data and the generation data of the generator is specifically:
constructing a data set according to the ground-sea clutter data and the generated data; training a discriminator by utilizing the deep learning algorithm according to the ground-sea clutter data and the generated data; wherein the generation data and the ground sea clutter data are equal in number.
6. The method for generating the ground-sea clutter spectrum data sample based on the countermeasure generation network according to claim 1, wherein the step of generating the mass sky-wave radar ground-sea clutter data by the generative countermeasure network, the pre-obtained data set and the random noise specifically comprises:
s21, obtaining the number of random noises according to the number of samples required to be generated;
s22, generating a ground-sea clutter sample with a target number by using the trained generator and random noise;
and S23, storing the generated ground sea clutter samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210054076.7A CN114492744A (en) | 2022-01-18 | 2022-01-18 | Method for generating ground-sea clutter spectrum data sample based on confrontation generation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210054076.7A CN114492744A (en) | 2022-01-18 | 2022-01-18 | Method for generating ground-sea clutter spectrum data sample based on confrontation generation network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114492744A true CN114492744A (en) | 2022-05-13 |
Family
ID=81512252
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210054076.7A Pending CN114492744A (en) | 2022-01-18 | 2022-01-18 | Method for generating ground-sea clutter spectrum data sample based on confrontation generation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114492744A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115372960A (en) * | 2022-07-11 | 2022-11-22 | 西北工业大学 | Sky wave radar ground and sea clutter data enhancement method for improving generation of countermeasure network |
| CN117217103A (en) * | 2023-11-09 | 2023-12-12 | 南京航空航天大学 | Satellite-borne SAR sea clutter generation method and system based on multi-scale attention mechanism |
-
2022
- 2022-01-18 CN CN202210054076.7A patent/CN114492744A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115372960A (en) * | 2022-07-11 | 2022-11-22 | 西北工业大学 | Sky wave radar ground and sea clutter data enhancement method for improving generation of countermeasure network |
| CN115372960B (en) * | 2022-07-11 | 2024-05-10 | 西北工业大学 | Method for enhancing sky-wave radar land-sea clutter data of improved generation countermeasure network |
| CN117217103A (en) * | 2023-11-09 | 2023-12-12 | 南京航空航天大学 | Satellite-borne SAR sea clutter generation method and system based on multi-scale attention mechanism |
| CN117217103B (en) * | 2023-11-09 | 2024-03-15 | 南京航空航天大学 | Satellite-borne SAR sea clutter generation method and system based on multi-scale attention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106228185B (en) | A kind of general image classifying and identifying system neural network based and method | |
| CN106355151B (en) | A kind of three-dimensional S AR images steganalysis method based on depth confidence network | |
| CN105488528B (en) | Neural network image classification method based on improving expert inquiry method | |
| CN108696331B (en) | Signal reconstruction method based on generation countermeasure network | |
| CN108133232A (en) | A kind of Radar High Range Resolution target identification method based on statistics dictionary learning | |
| CN114492744A (en) | Method for generating ground-sea clutter spectrum data sample based on confrontation generation network | |
| Wang et al. | SSRNet: In-field counting wheat ears using multi-stage convolutional neural network | |
| CN113128355A (en) | Unmanned aerial vehicle image real-time target detection method based on channel pruning | |
| CN113159048A (en) | Weak supervision semantic segmentation method based on deep learning | |
| CN110516525B (en) | SAR image target recognition method based on GAN and SVM | |
| Wang et al. | An unequal deep learning approach for 3-D point cloud segmentation | |
| CN110082738B (en) | Radar target identification method based on Gaussian mixture and tensor recurrent neural network | |
| CN115311502A (en) | Remote sensing image small sample scene classification method based on multi-scale double-flow architecture | |
| CN111222575B (en) | KLXS multi-model fusion method and system based on HRRP target recognition | |
| CN110503148B (en) | Point cloud object identification method with scale invariance | |
| CN114417942B (en) | Clutter recognition method, system, device and medium | |
| CN114879185A (en) | Radar target intelligent identification method based on task experience migration | |
| CN105678790A (en) | High-resolution remote sensing image supervised segmentation method based on variable Gaussian hybrid model | |
| CN116994071A (en) | Multispectral laser radar point cloud classification method based on self-adaptive spectrum residual error | |
| CN116597276A (en) | Target detection method based on improved YOLOv5 model | |
| CN115327504B (en) | Sea clutter amplitude distribution non-typed prediction method based on measurement condition parameters | |
| CN116340846A (en) | Aliasing modulation signal identification method for multi-example multi-label learning under weak supervision | |
| CN113920391B (en) | Target counting method based on scale generation self-adaptive truth diagram | |
| CN115712867A (en) | Multi-component radar signal modulation identification method | |
| CN110555483B (en) | Polarized SAR classification method based on FW-DCGAN feature generation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |