CN114720957A - Radar target detection method and system and storable medium - Google Patents
Radar target detection method and system and storable medium Download PDFInfo
- Publication number
- CN114720957A CN114720957A CN202210638434.9A CN202210638434A CN114720957A CN 114720957 A CN114720957 A CN 114720957A CN 202210638434 A CN202210638434 A CN 202210638434A CN 114720957 A CN114720957 A CN 114720957A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- radar
- detection method
- 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
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/414—Discriminating targets with respect to background clutter
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a radar target detection method, a system and a storable medium, relating to the technical field of radar target detection, wherein the method comprises the following steps: the method comprises the steps of obtaining radar receiving data, preprocessing the radar receiving data to form a data set, and dividing the data set into a training set and a verification set; constructing an enhanced convolutional neural network, training the enhanced convolutional neural network by using the training set and the data labels corresponding to the training set, verifying by using the verification set, and acquiring a network corresponding to the verification set with the minimum loss as an optimal detection weight network; the invention can improve the learning representation capability of the convolutional neural network.
Description
Technical Field
The invention relates to the technical field of radar target detection, in particular to a radar target detection method, a radar target detection system and a storage medium.
Background
At present, with the miniaturization and stealth of sea surface targets, the slow and small sea surface targets become key objects for radar warning. Detection of such small targets has long been a problem in target detection in the context of sea clutter. Typically, the radar scattering cross-sectional area of sea surface small targets is weak, and these targets have very low signal-to-noise ratios in conventional radars. Since the target moves at a slow speed and the sea clutter has a wide doppler bandwidth, the target and the sea clutter are difficult to distinguish in doppler, and the conventional detection method is difficult to work in such a situation.
However, for the detection of small targets floating on the sea surface, high doppler and high range resolution regimes ("dual high" regimes) are typically employed to address this problem. Under the 'double high system', target echoes received by the radar provide more available information, and considering that the high-resolution radar needs to face extremely complex clutter and target echo characteristics, joint detection, namely a characteristic-based detection technology, can be realized based on one or more different characteristics of the clutter and the target echoes. How to select the difference characteristics from the complex clutter and target characteristics is a difficult problem, and the existing method mainly relates to the characteristics of a signal layer, such as fractal characteristics, chaotic characteristics, time domain characteristics, frequency domain characteristics and time-frequency domain characteristics. Most of the existing intelligent detection methods are based on traditional machine learning such as K nearest neighbor algorithm, support vector machine, decision tree and the like, and are matched with multi-domain multi-feature designed manually to detect small targets, and the convolutional neural network detection method based on controllable false alarm still needs to be explored
Therefore, how to provide a radar target detection method capable of solving the above problems is a problem that needs to be solved by those skilled in the art.
Disclosure of Invention
In view of this, the invention provides a radar target detection method, a radar target detection system and a storage medium, wherein the enhanced convolutional neural network provided by the invention takes a multilayer residual error network as a backbone structure, introduces asymmetric convolution processing, and enhances the extraction and characterization capability of the network on fine features and edge features.
In order to achieve the purpose, the invention adopts the following technical scheme:
a radar target detection method comprises the following steps:
the method comprises the steps of obtaining radar receiving data, preprocessing the radar receiving data to form a data set, and dividing the data set into a training set and a verification set;
constructing an enhanced convolutional neural network, training the enhanced convolutional neural network by using the training set and the data labels corresponding to the training set, verifying by using the verification set, and acquiring a network corresponding to the verification set with the minimum loss as an optimal detection weight network;
preprocessing data to be detected, inputting the preprocessed data to be detected into the optimal detection weight network for detection to obtain an output result, and comparing the output result with a preset threshold value to judge whether a target exists in the data to be detected.
Preferably, the specific process of determining whether the target exists in the data to be detected includes:
and if the output result is less than or equal to the preset threshold value, judging that the data to be detected contains the target, otherwise, judging that the data to be detected does not contain the target.
Preferably, the enhanced convolutional neural network comprises a large convolutional layer, 4 residual error modules and a dense connection layer which are connected in sequence, wherein the residual error modules comprise 2 basic modules which are connected in sequence.
Preferably, the base module includes a plurality of sub-networks and a two-dimensional pooling layer, which are connected in sequence, and the sub-networks include a first convolution layer, a plurality of asymmetric convolution layers, a first normalization layer, a first active layer, a second convolution layer, a second normalization layer, and a second active layer, which are connected in sequence.
Preferably, the specific process of the pretreatment comprises:
converting the radar receiving data into a two-dimensional time-frequency diagram;
marking the two-dimensional time-frequency graph by using prior information to obtain a corresponding data label, wherein the data label comprises data only containing clutter but not containing a target and data containing the target;
and adjusting the size of the two-dimensional time-frequency graph, and taking the two-dimensional time-frequency graph with the adjusted size as the data set.
Preferably, the specific process of determining the preset threshold includes:
inputting data only containing clutter and not containing targets in the training set into the optimal detection weight network, storing a network output result into a variable C, and sequencing the variable C from small to large to obtain a 1 x N-dimensional real vector C';
calculating a sequence number i according to a preset false alarm probability, and further determining a detection threshold value T, wherein a specific expression of i is as follows:
in the formula (I), the compound is shown in the specification,PFAto preset the false alarm probability, T is the ith element in C'.
Preferably, the Adam method is specifically adopted for training and optimizing the enhanced convolutional neural network.
Further, the present invention also provides a detection system using the radar target detection method described in any one of the above, including:
the data acquisition module is used for acquiring radar receiving data, preprocessing the radar receiving data to form a data set, and dividing the data set into a training set and a verification set;
the network construction and training module is used for constructing an enhanced convolutional neural network, training the enhanced convolutional neural network by using the training set and the data labels corresponding to the training set, verifying by using the verification set, and acquiring a network corresponding to the verification set with the minimum loss as an optimal detection weight network;
the detection module is used for preprocessing data to be detected, inputting the preprocessed data to be detected into the optimal detection weight network for detection to obtain an output result, and comparing the output result with a preset threshold value to judge whether a target exists in the data to be detected.
Further, the present invention also provides a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the radar target detection method according to any one of the above.
Compared with the prior art, the invention discloses a radar target detection method, a system and a storage medium, and has the following advantages:
(1) the invention constructs an enhanced convolution neural network taking a residual error network as a backbone structure, introduces an asymmetric convolution module, enhances the network characteristic extraction and learning expression capability and realizes the steady false alarm control performance;
(2) the detection method designed by the invention realizes intelligent detection of the target controllable false alarm, avoids complex and possibly redundant manual feature design, and has one-time detection time of about 12 milliseconds on a GPU parallel platform;
(3) compared with the existing intelligent detection method, the detection method designed by the invention has higher detection efficiency and detection probability.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is an overall flowchart of a radar target detection method according to the present invention;
FIG. 2 is a schematic block diagram of a radar target detection system according to the present invention;
fig. 3 is a diagram of actually measured detection probability on an IPIX data set by the enhanced convolutional neural network-based radar target detection method provided in the embodiment of the present invention;
fig. 4 is a diagram of a control result of actually measured false alarm probability on an IPIX data set by the radar target detection method based on the enhanced convolutional neural network according to the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
Referring to the attached drawing 1, the embodiment of the invention discloses a radar target detection method, which comprises the following steps:
the method comprises the steps of obtaining radar receiving data, preprocessing the radar receiving data to form a data set, and dividing the data set into a training set and a verification set;
wherein, the data set is divided into a training set and a verification set according to the proportion of 7: 3;
constructing an enhanced convolutional neural network, training the enhanced convolutional neural network by using a training set and data labels corresponding to the training set, verifying by using a verification set, and acquiring a network corresponding to the verification set with the minimum loss as an optimal detection weight network;
preprocessing the data to be detected, inputting the preprocessed data to be detected into the optimal detection weight network for detection to obtain an output result, and comparing the output result with a preset threshold value to judge whether the target exists in the data to be detected.
In a specific embodiment, the specific process of the pre-treatment comprises:
converting radar receiving data into a two-dimensional time-frequency graph by using a time-frequency transformation method;
the time-frequency transformation method selects short-time Fourier transformation, and the specific expression is as follows:
in the formula (I), the compound is shown in the specification,f STFT (. cndot.) represents a short-time Fourier transform, M represents the number of radar data,T i a time-frequency diagram representing the ith data, the dimension of which is L x L,X i represents the ith signal sample with dimension 1 x L, L being the signal length;
the method comprises the steps of marking a two-dimensional time-frequency graph by utilizing expert knowledge and prior information to obtain a corresponding data label, wherein the data label comprises data only containing clutter but not containing targets and data containing targets, namely marking a true value on each time-frequency pattern in a data set, setting a true value of a clutter sample to be 0 and a true value of a target sample to be 1, and using a cross entropy loss function for network training, wherein the definition is as follows:
where n is the number of training samples, the summation is over all the training inputs x, y is the true label of the input samples, ln (·) represents the natural logarithm, and O is the output of the dense connection layer.
Adjusting the pixels of the two-dimensional time-frequency image to be 256 multiplied by 256 and the size to be 224 multiplied by 3, namely, each image is three channels of red, green and blue, the length and the width of each channel are 224 pixels, randomly overturning the channels in the horizontal direction with the probability of 0.5, then respectively normalizing the three channels of red, green and blue of the obtained image, setting the three-channel normalized mean values to be 0.485, 0.456 and 0.406 respectively, setting the normalized variances to be 0.229, 0.224 and 0.225 respectively, and taking the two-dimensional time-frequency image after the adjustment as a data set.
In a specific embodiment, the enhanced convolutional neural network comprises a large convolutional layer, 4 residual modules and a dense connection layer which are connected in sequence, wherein the residual modules comprise 2 basic modules which are connected in sequence.
Specifically, the backbone structure of the enhanced convolutional neural network can be a residual network with 18 layers, and asymmetric convolutional blocks are embedded in conventional convolutional layers in the backbone network, so that the extraction and characterization capabilities of the backbone network on fine features and edge features are enhanced.
The last layer of the enhanced convolutional neural network is a dense connection layer using a SoftMax function and used for outputting clutter probability and target probability corresponding to each sample, wherein the specific expression of the SoftMax function is as follows:
in the formula (I), the compound is shown in the specification,O i representing the output of the ith neuron in the densely connected layer,e (·)is an exponential function and j represents the number of neurons in this densely connected layer.
Specifically, the residual error network may be formed by cascading 1 large convolutional layer, 8 basic modules, and 1 dense connection layer, where the number of input and output channels of the large convolutional layer is 3 and 64, respectively, and 7 × 7 convolutional kernels are used, the convolutional step is 2, and the number of zero-padding turns around is 3, that is, the input of the large convolutional layer is,H p AndW p for the number of pixels of the height and width of the input image, the output of the large convolution layer isWherein:
where K is the convolution kernel size, P is the number of zero-filled turns around, and S is the convolution step.
The output of the large convolution layer is firstly normalized in batch, and is defined as follows:
wherein z is the output of the large convolution layer,μAndσ 2respectively, mean and variance of the batch samples,Representing the product by element,γAndβrespectively, the parameters of zooming and panning.
The output of the batch normalization is activated by a ReLU function, which is defined as
In the formula, x is input of a function, max (0, x) represents the maximum value selected from 0 and x, the neuron activity value output by the activation function is subjected to two-dimensional maximum value pooling operation to reduce sampling, the size of a pooling kernel is 3 multiplied by 3, the pooling step length is 2, the number of zero padding circles around is 1, and if the input characteristic of the pooling layer is M X N D dimensional data Y, Y is mapped to each M X N dimensional characteristicdIt can be divided into a plurality of regionsAnd selecting the maximum activity value of all neurons in the region as the representation of the region by maximum pooling, wherein the specific expression is as follows:
in the formula (I), the compound is shown in the specification,x i is the activity value, sign, of each neuron in the regionIndicating the belonging.
In a specific embodiment, the base module comprises a plurality of sub-networks and a two-dimensional pooling layer which are connected in sequence, wherein the sub-networks comprise a first convolution layer, a plurality of asymmetric convolution layers, a first normalization layer, a first activation layer, a second convolution layer, a second normalization layer and a second activation layer which are connected in sequence.
Specifically, 8 basic modules can be divided into 4 groups of sub-networks, each group of sub-networks is formed by cascading 2 basic modules with the same structure, batch normalization operation is performed after layers are coiled in the modules and activated by using a ReLU function, the head and the tail of each basic module are connected in an identical manner to form short circuits, and the input of the first-level sub-network is the output of the pooling operation.
Each basic module in the first-level sub-network comprises 2 conventional convolutional layers and 2 asymmetric convolutional layers; each conventional convolution layer uses a 3 multiplied by 3 convolution kernel, the convolution step length is 1, and the number of zero padding turns around is 1; the 2 asymmetric convolution layers respectively use convolution kernels of 1 x 3 and 3 x 1, other parameters are consistent with those of the conventional convolution layers, the output of the 2 asymmetric convolution layers is added with the output of the first conventional convolution layer, then the obtained product is taken as the input of the second conventional convolution layer after being subjected to batch normalization operation and ReLU function activation, the batch normalization output of the second conventional convolution layer is added with the input of the first-level sub-network, and then the obtained product is taken as the output of the first-level sub-network after being subjected to ReLU function activation;
the input of the second-level sub-network is the output of the first-level sub-network, and each basic module in the second-level sub-network comprises 2 conventional convolutional layers and 2 asymmetric convolutional layers; the number of input channels and output channels of the 4 convolutional layers is 128, the convolution step length of the first conventional convolutional layer and the convolution step length of the asymmetric convolutional layer in the first basic module are 2, and other parameters and the structure of the sub-network are consistent with that of the first-level sub-network; similarly, the input of the third-level sub-network is the output of the second-level sub-network, and each basic module in the third-level sub-network comprises 2 conventional convolutional layers and 2 asymmetric convolutional layers; the number of input and output channels of the 4 convolutional layers is 256, and other parameters and the structure of the sub-network are consistent with that of the second-level sub-network; the input of the fourth-level sub-network is the output of the third-level sub-network, and each basic module in the fourth-level sub-network comprises 2 conventional convolutional layers and 2 asymmetric convolutional layers; the number of input and output channels of the 4 convolutional layers is 512, and other parameters and the structure of the sub-network are consistent with that of the third-level sub-network;
the output of the fourth sub-network is subjected to a two-dimensional mean pooling operation with a pooling kernel size of 7 x 7 and a pooling step size of 1, also for regionsAnd taking the average value of all neuron activity values in the region as the representation of the region by using the average value pooling, wherein the specific expression is as follows:
the output after pooling is used as the input of a dense connection layer, the number of input channels and the number of output channels are respectively 512 and 2, the output of the dense connection layer is mapped and output through a SoftMax function to formP={P 1, P 2Therein ofP 1AndP 2respectively represent clutter probability and target probability, anP 1+P 2=1。
In a specific embodiment, the Adam method is specifically adopted for training the optimized enhanced convolutional neural network.
Specifically, the Adam algorithm is based on a small-batch gradient descent method, if orderf(x;θ) A neural network is represented that is a network of nerves,θwhen a small batch gradient descent method is used for network parameters, K training samples are selected each timeS K The partial derivative about the loss function at the number of t iterations is:
in the formula (I), the compound is shown in the specification,Lwith K being the batch size, then updating the network parameters t,For learning rate, the parameter updates the difference at each iterationThe Adam algorithm calculates, on the one hand, an exponentially weighted average of the squares of the gradientsOn the other hand, an exponentially weighted average of the gradients is calculated, namely:
in the formula (I), the compound is shown in the specification,β 1andβ 2the attenuation rates of the two moving averages are respectively taken asβ 1=0.9、β 2=0.99,M t AndG t the mean (first moment) and the variance (second moment) of the non-subtracted mean of the gradient, respectively, the parameter update difference of the Adam algorithm is:
in the formula (I), the compound is shown in the specification,εis a very small constant set for maintaining numerical stability, and takes a value of 10-7To 10-10The network training batch size is 256, and the learning rate is 10-3The iteration loop is 50, where the loop-to-iteration relationship is updated for each small batch to one iteration,and updating the samples of all the training sets into a round, wherein the verification set does not participate in the training and optimization of the network and is only used for verifying the performance of the current weight of the network, so that the weight with the minimum loss of the verification set is selected as the optimal detection model.
In a specific embodiment, the specific process of determining the preset threshold includes:
inputting data only containing clutter and not containing targets in a training set into an optimal detection weight network, storing a network output result into a variable C, and sequencing the variable C from small to large to obtain a 1 x N-dimensional real vector C';
calculating a sequence number i according to a preset false alarm probability, and further determining a detection threshold value T, wherein a specific expression of i is as follows:
in the formula (I), the compound is shown in the specification,PFAto preset false alarm probability, the value is generally 10^ -6 magnitude, and T is the ith element in C'.
In a specific embodiment, the specific process of determining whether the target exists in the data to be detected includes:
and if the output result P is less than or equal to the preset threshold value T, judging that the data to be detected contains the target, otherwise, judging that the data to be detected does not contain the target.
Referring to fig. 2, an embodiment of the present invention further provides a detection system using the radar target detection method according to any one of the above embodiments, including:
the data acquisition module is used for acquiring radar receiving data, preprocessing the radar receiving data to form a data set, and dividing the data set into a training set and a verification set;
the network construction and training module is used for constructing an enhanced convolutional neural network, training the enhanced convolutional neural network by using a training set and data labels corresponding to the training set, verifying by using a verification set, and acquiring a corresponding network as an optimal detection weight network when the loss of the verification set is minimum;
the detection module is used for preprocessing the data to be detected, inputting the preprocessed data to be detected into the optimal detection weight network for detection to obtain an output result, and comparing the output result with a preset threshold value to judge whether the target exists in the data to be detected.
An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when being executed by a processor, the computer program implements the radar target detection method according to any one of the above embodiments.
As shown in fig. 3, the target processing flow provided by the embodiment of the present invention obtains a probability curve of detection of a radar target based on an enhanced convolutional neural network under different false alarm probabilities, where the detection probability is obtained by detecting for more than ten thousand times under each false alarm probability. It can be seen that the detection method provided by the embodiment of the invention has a false alarm probability of 10-3The detection probability is about 0.864, and when the false alarm probability is higher than 0.003, the detection probability can reach more than 0.9.
The detection probability comparison results of the radar target detection method based on the enhanced convolutional neural network and several classical sea surface small target detection methods are shown in table 1, wherein fractal characteristics refer to a detection method based on signal fractal characteristics, three characteristics refer to a detection method based on signal time domain and frequency domain, time-frequency characteristics refer to a detection method based on time-frequency characteristics, and HH, HV, VH and VV are four polarization channels of the radar in HH, HV, VH and VV respectively. It can be seen that the method provided by the embodiment of the invention has higher detection probability on four polarized channels.
TABLE 1 comparison of detection probability between the detection method of the present invention and the classical detection method
The actual false alarm probability can be obtained by calculating the number of clutter samples determined as targets by the detection method, and the false alarm control performance of the detection method can be analyzed by comparing the actual false alarm probability with the preset false alarm probability, as shown in fig. 4. The dotted line represents the ideal false alarm probability, i.e. the preset false alarm probability is equal to the actual false alarm probability, and the solid line represents the actual false alarm of the detection method. It can be seen that the two lines are approximately coincident, proving the robustness of false alarm control of the detection method.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (9)
1. A radar target detection method is characterized by comprising the following steps:
the method comprises the steps of obtaining radar receiving data, preprocessing the radar receiving data to form a data set, and dividing the data set into a training set and a verification set;
constructing an enhanced convolutional neural network, training the enhanced convolutional neural network by using the training set and the data labels corresponding to the training set, verifying by using the verification set, and acquiring a network corresponding to the verification set with the minimum loss as an optimal detection weight network;
preprocessing data to be detected, inputting the preprocessed data to be detected into the optimal detection weight network for detection to obtain an output result, and comparing the output result with a preset threshold value to judge whether a target exists in the data to be detected.
2. The radar target detection method according to claim 1, wherein the specific process of determining whether the target exists in the data to be detected includes:
and if the output result is less than or equal to the preset threshold value, judging that the data to be detected contains the target, otherwise, judging that the data to be detected does not contain the target.
3. The radar target detection method of claim 1, wherein the enhanced convolutional neural network comprises a large convolutional layer, 4 residual error modules and a dense connection layer which are connected in sequence, wherein the residual error modules comprise 2 basic modules which are connected in sequence.
4. The radar target detection method of claim 3, wherein the base module comprises a plurality of sub-networks and a two-dimensional pooling layer which are connected in sequence, and the sub-networks comprise a first convolution layer, a plurality of asymmetric convolution layers, a first normalization layer, a first active layer, a second convolution layer, a second normalization layer and a second active layer which are connected in sequence.
5. The radar target detection method according to claim 1, wherein the preprocessing comprises:
converting the radar receiving data into a two-dimensional time-frequency diagram;
marking the two-dimensional time-frequency graph by using prior information to obtain a corresponding data label, wherein the data label comprises data only containing clutter but not containing a target and data containing the target;
and adjusting the size of the two-dimensional time-frequency graph, and taking the two-dimensional time-frequency graph with the adjusted size as the data set.
6. The radar target detection method according to claim 5, wherein the specific process of determining the preset threshold includes:
inputting data only containing clutter and not containing targets in the training set into the optimal detection weight network, storing a network output result into a variable C, and sequencing the variable C from small to large to obtain a 1 x N-dimensional real vector C';
calculating a sequence number i according to a preset false alarm probability, and further determining a detection threshold value T, wherein a specific expression of i is as follows:
in the formula (I), the compound is shown in the specification,PFAto preset the false alarm probability, T is the ith element in C'.
7. The radar target detection method of claim 5, wherein the Adam method is specifically adopted for training and optimizing the enhanced convolutional neural network.
8. A detection system using the radar target detection method according to any one of claims 1 to 7, characterized by comprising:
the data acquisition module is used for acquiring radar receiving data, preprocessing the radar receiving data to form a data set, and dividing the data set into a training set and a verification set;
the network construction and training module is used for constructing an enhanced convolutional neural network, training the enhanced convolutional neural network by using the training set and the data labels corresponding to the training set, verifying by using the verification set, and acquiring a network corresponding to the verification set with the minimum loss as an optimal detection weight network;
the detection module is used for preprocessing data to be detected, inputting the preprocessed data to be detected into the optimal detection weight network for detection to obtain an output result, and comparing the output result with a preset threshold value to judge whether a target exists in the data to be detected.
9. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the radar target detection method according to any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210638434.9A CN114720957A (en) | 2022-06-08 | 2022-06-08 | Radar target detection method and system and storable medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210638434.9A CN114720957A (en) | 2022-06-08 | 2022-06-08 | Radar target detection method and system and storable medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114720957A true CN114720957A (en) | 2022-07-08 |
Family
ID=82232822
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210638434.9A Pending CN114720957A (en) | 2022-06-08 | 2022-06-08 | Radar target detection method and system and storable medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114720957A (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110363151A (en) * | 2019-07-16 | 2019-10-22 | 中国人民解放军海军航空大学 | Based on the controllable radar target detection method of binary channels convolutional neural networks false-alarm |
CN110796009A (en) * | 2019-09-29 | 2020-02-14 | 航天恒星科技有限公司 | Method and system for detecting marine vessel based on multi-scale convolution neural network model |
CN111123257A (en) * | 2019-12-30 | 2020-05-08 | 西安电子科技大学 | Radar moving target multi-frame joint detection method based on graph space-time network |
CN111722199A (en) * | 2020-08-10 | 2020-09-29 | 上海航天电子通讯设备研究所 | Radar signal detection method based on convolutional neural network |
-
2022
- 2022-06-08 CN CN202210638434.9A patent/CN114720957A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110363151A (en) * | 2019-07-16 | 2019-10-22 | 中国人民解放军海军航空大学 | Based on the controllable radar target detection method of binary channels convolutional neural networks false-alarm |
CN110796009A (en) * | 2019-09-29 | 2020-02-14 | 航天恒星科技有限公司 | Method and system for detecting marine vessel based on multi-scale convolution neural network model |
CN111123257A (en) * | 2019-12-30 | 2020-05-08 | 西安电子科技大学 | Radar moving target multi-frame joint detection method based on graph space-time network |
CN111722199A (en) * | 2020-08-10 | 2020-09-29 | 上海航天电子通讯设备研究所 | Radar signal detection method based on convolutional neural network |
Non-Patent Citations (1)
Title |
---|
瞿奇哲: "基于深度学习的雷达信号识别", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110927706B (en) | Convolutional neural network-based radar interference detection and identification method | |
CN108229404B (en) | Radar echo signal target identification method based on deep learning | |
CN109597043B (en) | Radar signal identification method based on quantum particle swarm convolutional neural network | |
CN107728142B (en) | Radar high-resolution range profile target identification method based on two-dimensional convolutional network | |
CN112001270B (en) | Ground radar automatic target classification and identification method based on one-dimensional convolutional neural network | |
CN111160176B (en) | Fusion feature-based ground radar target classification method for one-dimensional convolutional neural network | |
Awad | An Unsupervised Artificial Neural Network Method for Satellite Image Segmentation. | |
CN112561796B (en) | Laser point cloud super-resolution reconstruction method based on self-attention generation countermeasure network | |
CN114595732B (en) | Radar radiation source sorting method based on depth clustering | |
CN114429156A (en) | Radar interference multi-domain feature countermeasure learning and detection identification method | |
CN110690930B (en) | Information source number detection method and device | |
Lin et al. | DL-CFAR: A novel CFAR target detection method based on deep learning | |
CN113759323B (en) | Signal sorting method and device based on improved K-Means joint convolution self-encoder | |
CN111707998B (en) | Sea surface floating small target detection method based on connected region characteristics | |
CN110933633A (en) | Onboard environment indoor positioning method based on CSI fingerprint feature migration | |
Nuhoglu et al. | Image segmentation for radar signal deinterleaving using deep learning | |
CN111190158A (en) | Complex radar radiation source identification method and device, electronic equipment and storage medium thereof | |
CN114895263A (en) | Radar active interference signal identification method based on deep migration learning | |
CN113673312A (en) | Radar signal intra-pulse modulation identification method based on deep learning | |
CN114117912A (en) | Sea clutter modeling and inhibiting method under data model dual drive | |
CN115097396A (en) | Radar active interference identification method based on CNN and LSTM series model | |
CN112835008B (en) | High-resolution range profile target identification method based on attitude self-adaptive convolutional network | |
Wen et al. | Modeling of correlated complex sea clutter using unsupervised phase retrieval | |
CN116482618B (en) | Radar active interference identification method based on multi-loss characteristic self-calibration network | |
Zhang et al. | Estimation of complex high-resolution range profiles of ships by sparse recovery iterative minimization method |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220708 |