CN109709536A - A kind of SAR moving target detection method based on convolutional neural networks - Google Patents
A kind of SAR moving target detection method based on convolutional neural networks Download PDFInfo
- Publication number
- CN109709536A CN109709536A CN201910065920.4A CN201910065920A CN109709536A CN 109709536 A CN109709536 A CN 109709536A CN 201910065920 A CN201910065920 A CN 201910065920A CN 109709536 A CN109709536 A CN 109709536A
- Authority
- CN
- China
- Prior art keywords
- layer
- neural networks
- target
- convolutional neural
- sar
- 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
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
The present invention discloses a kind of SAR moving target detection method based on convolutional neural networks, applied to SAR moving object detection field, target auxiliary data range gate number to be detected can be used to require the defects of high for of the existing technology, method of the invention passes through building convolutional neural networks, and neural network is trained, the angle of the opposite synthetic aperture of total Doppler frequency, target detected according to neural network, is calculated target velocity, to realize that SAR moving-target detects;Has the advantages of detection is realized in the case where target auxiliary data range gate number to be detected is few.
Description
Technical field
The invention belongs to Radar Technology field, in particular to a kind of synthetic aperture radar (Synthetic Aperture
Radar is abbreviated as SAR) Detection for Moving Target.
Background technique
It is the one of synthetic aperture radar that moving-target, which detects (Moving Target Indication, be abbreviated as MTI) technology,
Item key technology.Moving target detection technique is able to detect the moving target within the scope of beam, and transports to its movement velocity etc.
Moving-target parameter is estimated.
Traditional MTI method mainly includes that (Space time adaptive processing, writes a Chinese character in simplified form for space-time adaptive processing
For STAP) (bibliography 1:L.E.Brennan, L.S.Reed.Theory of adaptive radar [J] .IEEE
Transactions on Aerospace and Electronic Systems, 1973,9 (2): 237-252) and STAP
Various innovatory algorithms (bibliography 2:H.Wang, L.Cai.On adaptive spatial-temporal processing
for airborne surveillance radar systems[J].IEEE Transactions on aerospace and
Electronic systems, 1994,30 (3): 660-670, bibliography 3:J.S.Goldstein,
I.S.Reed.Reduced-rank adaptive filtering[J].IEEE Transactions on Signal
Processing,1997,45(2):492-496).In recent years, researcher proposes to utilize the linear classifier in machine learning
Method (the bibliography 4:A.E.Khatib, K.Assaleh, H.Mir.Learning- of MTI are realized with multinomial classifier
based space-time adaptive processing[C]//International Conference on
Communications.IEEE, 2013:1-4, bibliography 5:A.E.Khatib, K.Assaleh, H.Mir.Space-time
adaptive processing using pattern classification[J].IEEE Transactions on
Signal Processing,2015,63(3):766-779).However, these methods exist to target supplementary number to be detected can be used
It requires high according to range gate number or is only capable of realizing that detection etc. is lacked in the case where SAR moving target signal-clutter power is relatively high
It falls into, thus the application of existing MTI method is limited, it is difficult to meet the needs of practical application.
Summary of the invention
In order to solve the above technical problems, the present invention proposes a kind of SAR moving target detection method based on convolutional neural networks,
Accurate SAR moving-target detection can be achieved in the case where available auxiliary data range gate deficiency and low signal to noise ratio.
A kind of the technical solution adopted by the present invention are as follows: SAR moving target detection method based on convolutional neural networks, comprising:
S1, building convolutional neural networks;It include: 4 convolutional layers, 2 pond layers, final point of 2 full articulamentums and 1
Total 9 layers of class layer, be denoted as: level 1 volume lamination, level 2 volume lamination, the 3rd layer of pond layer, the 4th layer of convolutional layer, the 5th layer of convolutional layer,
6th layer of pond layer, the 7th layer of full articulamentum, the 8th layer of full articulamentum and the 9th layer of final classification layer;The convolution of level 1 volume lamination
Core size is 5 × 1, channel number 64, and the convolution kernel size of level 2 volume lamination is 5 × 1, channel number 96, the 3rd layer of pond
The filter size for changing layer is 3 × 3, and step-length 1, the convolution kernel size of the 4th layer of convolutional layer is 5 × 1, channel number 128, the
The convolution kernel size of 5 layers of convolutional layer is 5 × 1, channel number 128, and the filter size of the 6th layer of pond layer is 3 × 3, step-length
The output node for being the 2, the 7th layer of full articulamentum is 1000;The output node of 8th layer of full articulamentum is 192;9th layer final
The output node of classification layer is K;
K=A × B, A are the possibility number of total Doppler frequency to be detected, and B is target to be detected with respect to synthetic aperture
Angle possibility number
S2, building SAR moving-target detect training dataset;SAR moving-target detection training dataset includes K × Q × H instruction
Practice data matrix X(a,b,q,h);
Wherein, a indicates the serial number of total Doppler frequency;B indicates serial number of the target with respect to the angle of synthetic aperture;Q is indicated
The serial number of auxiliary data, q=1,2 ..., Q, Q indicate the sum of auxiliary data range gate;H indicates the training of target amplitude building
The serial number of data, h=1,2 ..., H, H indicate the amplitude sum of construction target.
S3, training dataset is detected according to the SAR moving-target of step S2 building, to the convolutional neural networks of step S1 building
It is trained;The weight and offset parameter of convolutional neural networks are specifically obtained using formula Θ=argminJ (Θ), wherein Θ
It is all parameters comprising convolutional neural networks, J (Θ) is cross entropy loss function.
Wherein, p (k | X(k,q,h);It Θ) is the defeated of convolutional neural networks the last layer
Data out.
S4, SAR echo data to be detected is detected according to the convolutional neural networks after step S3 training, obtains mesh
Mark speed.Specifically, a, b is calculated according to the following formula respectively;
B=mod ((k+1), B)
The speed of corresponding SAR moving target is calculated according to above-mentioned a, b.
Beneficial effects of the present invention: a kind of SAR moving target detection method based on convolutional neural networks of the invention, only
The detection of SAR moving-target can be achieved in the case where there are 8 auxiliary data range gates, and under the application environment of low signal to noise ratio, move
Target echo is submerged in clutter and noise, still can get be more than 90% accuracy;Compared with the existing methods, side of the present invention
Method requires less auxiliary range gate number, and can show robustness in isomerous environment;The method of the present invention has following excellent
Point:
1, useful feature is successively extracted using deep learning neural network, using pond layer, reduces network parameter, reduced
The over-fitting degree of model;
2, a large amount of training datas are utilized, each class generates the training data of identical quantity, avoids class imbalance and asks
Topic;
3, auxiliary data is only used as jamming pattern, thus greatly reduce auxiliary data range gate number requirement and its
The influence of isomerism.
Detailed description of the invention
Fig. 1 is the program flow chart that present invention implementation provides;
Fig. 2 is the convolutional neural networks structure chart that present invention implementation provides.
Specific embodiment
In order to facilitate the description contents of the present invention, make following term definition first:
Observing matrix when defining 1, sky
Observing matrix X is a two-dimensional matrix when empty, wherein storing after SAR echo signal Range compress as a result, should
Each row of two-dimensional matrix represents each pulse, and each column represent each antenna channels, the n-th column element of m row of the two-dimensional matrix
Mathematic(al) representation is
First item indicates the echo of SAR moving-target signal, Section 2 χ in formula (1)m,nIndicate clutter and noise jamming, m=
1 ..., M, n=1 ..., N, M are the transmitting umber of pulses in a coherent processing inteval, and N is antenna channels number, m, n difference
It is the index of pulse and antenna, α is the amplitude of moving-target,It is random phase, ln=Ln/ λ, LnIt is n-th of antenna with respect to the 1st
The distance of a antenna, λ are the signal wavelengths of radar system work, and θ is angle of the target with respect to synthetic aperture, ftIt is total Doppler
Frequency, including radar platform move caused Doppler frequency fdDoppler frequency f caused by being moved with targetv, can be expressed as
ft=fd sin(θ)+fv (2)
Wherein, φ is angle of the orientation speed of moving-target relative to target velocity, vtIt is the speed of moving-target, vpIt is
The speed of radar platform, frIt is the pulse recurrence frequency of radar system.
Define 2, convolutional neural networks
Convolutional neural networks are a classifiers end to end, include convolutional layer, pond layer, full articulamentum, softmax layers
With classification layer.
Define 3, convolutional layer
Convolutional layer completes linear convolution and Nonlinear Processing two operations, can be expressed as
Wherein,It is the output data and input data of this layer, γ respectivelyc() is non-linear
Operator is activated,It is corresponding weighting parameter and offset parameter respectively,It is to output and input
Channel index, Fc×Fc, ξ, η=1 ..., FcIt is filter size, ScIt is the step-length of convolutional layer, indicates two adjacent inputs
Interval between subregion.
Define 4, pond layer
The operation that pond layer is completed can be expressed as
Wherein,It is the output data and input data of this layer, F respectivelyp×FpIt is the filter for inputting subregion
Size, SpIt is the step-length of pond layer, indicates the interval between two adjacent input subregions.
Define 5, full articulamentum
The operation that full articulamentum carries out can be expressed as
Wherein,It is the output data and input data of this layer, γ respectivelyf() is the nonlinear function of full articulamentum
Operator,It is weight vector and the biasing for connecting m-th of element of this layer of input vector and output vector,<>indicates
The inner product of two vectors.
Define 6, final classification layer
Final classification layer carries out following calculate
Wherein, yk,yfIt is the output data and input data of this layer, w respectivelyk,bkIt is this layer of input vector of connection and output
The weight vector of k-th of element of vector and biasing.
For convenient for those skilled in the art understand that technology contents of the invention, with reference to the accompanying drawing to the content of present invention into one
Step is illustrated.
A kind of process of the SAR moving target detection method based on convolutional neural networks used as shown in Figure 1 for the present invention
Figure, the specific implementation process is as follows:
Step 1: building convolutional neural networks
Construct convolutional neural networks, the network structure as shown in Fig. 2, the network include altogether 4 convolutional layers, 2 pond layers,
2 full articulamentums and 1 final classification layer.The convolution kernel size of level 1 volume lamination is 5 × 1, and channel number 64 uses full 0
Supplement, step-length 1, the input matrix of this layer is each training data, i.e. input size is 438 × 3 × 2;Level 2 volume lamination
Convolution kernel size is 5 × 1, and channel number 96 is supplemented using full 0, step-length 1, and the input matrix of this layer is upper one layer defeated
Matrix out, i.e. input size is 438 × 3 × 64;The filter size of 3rd layer of pond layer be 3 × 3, step-length 1, this layer it is defeated
Entering matrix is upper one layer of output matrix, i.e. input size is 438 × 3 × 96;The convolution kernel size of 4th layer of convolutional layer be 5 ×
1, channel number 128 is supplemented using full 0, step-length 1, and the input matrix of this layer is upper one layer of output matrix, i.e. input is big
Small is 438 × 3 × 96;The convolution kernel size of 5th layer of convolutional layer is 5 × 1, and channel number 128 is supplemented using full 0, and step-length is
1, the input matrix of this layer is upper one layer of output matrix, i.e. input size is 438 × 3 × 128;The filtering of 6th layer of pond layer
Device size is 3 × 3, and step-length 2, the input matrix of this layer is upper one layer of output matrix, i.e., input size be 438 × 3 ×
128;The input node number of 7th layer of full articulamentum is 219 × 2 × 128, and output node is 1000;8th layer of full articulamentum
Input node number be 1000, output node be 192;The input node number of 9th layer of final classification layer is 192,
Output node is 15.
Step 2: initialization radar system parameters
Initialize radar system parameters: initialization radar system parameters: the signal wavelength of radar system work: λ;One phase
Transmitting umber of pulse in dry-cure interval: M;Antenna channels number: N;Radar system pulse recurrence frequency: fr;Radar platform
Speed: vp.The signal wavelength lambda value that radar system works in the present embodiment is 0.0312m, other parameters for details see attached table 1.
The system parameter list of 1 measured data of table
Parameter | Measured data |
Antenna channels number N | 3 |
Transmitting umber of pulse M in one coherent processing inteval | 438 |
Pulse recurrence frequency fr(Hz) | 2.1716e+03 |
Radar platform speed vp(m/s) | 104.2616 |
Step 3: building SAR moving-target detects training dataset
It is combined with target with respect to the possibility of the angle of synthetic aperture using 15 total Doppler frequencies, limited 8 auxiliary
Data, 201 target amplitudes come enhanced SAR moving-target detection training dataset, one of training dataIt is one
The two-dimensional data matrix of 438 rows 3 column, two-dimensional matrixThe n-th column element of m row are as follows:
Wherein, subscript a, b, q, h respectively indicate different total Doppler frequencies, target with respect to the angle of synthetic aperture, auxiliary
Help the amplitude of data range gate and arteface target, and to represent different total Doppler frequencies opposite with target by K=A × B, k
K, the corresponding relationship of a, b is given below in the combination of the angle of synthetic aperture
B=mod ((k+1), B) (11)
Wherein,It indicates to be rounded downwards, its remainder is sought in mod () expression, it should be noted that as b ≠ 0, b=mod
[(k+1), B], as b=0, b=B.
A=5, B=3, a=1,2 in the present embodiment, 3,4,5, b=1,2,3, k=0,1,2 ..., 14, q=1,2 ...,
8, h=1,2 ..., 201.
In formula (9), random phaseObey [0,2 π) on be uniformly distributed, and with a, b, q, h change, α(a,b,h)It is
Kth class (according to (10) (11) formula, h-th of target amplitude of k) training data is obtained by a, b, andWhereinIt is the upper limit and lower bound of the amplitude of kth class training data respectively.
Step 4: training convolutional neural networks
Using the training dataset of step 3 come training convolutional neural networks, obtained using formula Θ=argminJ (Θ)
The weight and offset parameter of convolutional neural networks, wherein Θ is all parameters comprising convolutional neural networks, and J (Θ) is cross entropy
Loss function, i.e.,In the formula, and p (k | X(k,q,h);It Θ) is convolutional neural networks the last layer
Output.
The optimization problem is solved using the method for gradient decline and backpropagation, can be obtainedWherein i=1,2 ..., 400 indicates the number of iterations of gradient decline, the initial value of parameter vector
Θ (0) is randomly provided, and η (i) is the learning rate of i-th iteration,It is that gradient calculates.
Step 5: detection SAR moving target
Using SAR echo data matrix to be detected as the input of network, test processes are then carried out, obtain corresponding class
Not Shuo kth class, whereinWorking as k=0,1,2 ..., K-1 can first according to (10) (11) formula
The angle of corresponding a-th of total Doppler frequency synthetic aperture opposite with b-th of target is obtained, it is last according to (2), (3), (4)
The speed of corresponding SAR moving target can be obtained in formula.
Those of ordinary skill in the art will understand that the embodiments described herein, which is to help reader, understands this hair
Bright principle, it should be understood that protection scope of the present invention is not limited to such specific embodiments and embodiments.For ability
For the technical staff in domain, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made
Any modification, equivalent substitution, improvement and etc. should be included within scope of the presently claimed invention.
Claims (8)
1. a kind of SAR moving target detection method based on convolutional neural networks characterized by comprising
S1, building convolutional neural networks;It include: 4 convolutional layers, 2 pond layers, 2 full articulamentums and 1 final classification layer
It is 9 layers total, it is denoted as: level 1 volume lamination, level 2 volume lamination, the 3rd layer of pond layer, the 4th layer of convolutional layer, the 5th layer of convolutional layer, the 6th
Layer pond layer, the 7th layer of full articulamentum, the 8th layer of full articulamentum and the 9th layer of final classification layer;The convolution kernel of level 1 volume lamination is big
Small is 5 × 1, channel number 64, and the convolution kernel size of level 2 volume lamination is 5 × 1, channel number 96, the 3rd layer of pond layer
Filter size be 3 × 3, step-length 1, the convolution kernel size of the 4th layer of convolutional layer is 5 × 1, channel number 128, the 5th layer
The convolution kernel size of convolutional layer is 5 × 1, channel number 128, and the filter size of the 6th layer of pond layer is 3 × 3, step-length 2,
The output node of 7th layer of full articulamentum is 1000;The output node of 8th layer of full articulamentum is 192;9th layer of final classification
The output node of layer is K;
S2, building SAR moving-target detect training dataset;
S3, training dataset is detected according to the SAR moving-target of step S2 building, the convolutional neural networks of step S1 building is carried out
Training;
S4, SAR echo data to be detected is detected according to the convolutional neural networks after step S3 training, obtains target speed
Degree.
2. a kind of SAR moving target detection method based on convolutional neural networks according to claim 1, which is characterized in that
K=A × B in step S1, A are the possibility number of total Doppler frequency to be detected, and B is target to be detected with respect to synthetic aperture
Angle possibility number.
3. a kind of SAR moving target detection method based on convolutional neural networks according to claim 2, which is characterized in that
It includes K × Q × H training data matrix X that SAR moving-target described in step S2, which detects training dataset,(a,b,q,h);
Wherein, a indicates the serial number of total Doppler frequency;B indicates serial number of the target with respect to the angle of synthetic aperture;Q indicates auxiliary
The serial number of data, q=1,2 ..., Q, Q indicate the sum of auxiliary data range gate;H indicates the training data of target amplitude building
Serial number, h=1,2 ..., H, H indicate construction target amplitude sum.
4. a kind of SAR moving target detection method based on convolutional neural networks according to claim 3, which is characterized in that
Step S3 uses formula Θ=arg min J (Θ) specifically to obtain the weight and offset parameter of convolutional neural networks, wherein Θ
It is all parameters comprising convolutional neural networks, J (Θ) is cross entropy loss function.
5. a kind of SAR moving target detection method based on convolutional neural networks according to claim 4, which is characterized in thatWherein, k represents the angle of different total Doppler frequencies synthetic aperture opposite with target
Combination serial number;K=0,1 ..., K-1, q indicate the serial number of auxiliary data, q=1,2 ..., Q;H indicates target amplitude building
Training data serial number, h=1,2 ..., H;p(k|X(k,q,h);Θ) be convolutional neural networks the last layer output data.
6. a kind of SAR moving target detection method based on convolutional neural networks according to claim 5, which is characterized in that
Step S4 further include: according to the angle of the corresponding total Doppler frequency of the SAR to be detected being calculated, the opposite synthetic aperture of target
Degree, obtains target velocity.
7. a kind of SAR moving target detection method based on convolutional neural networks according to claim 6, which is characterized in that
The serial number a calculating formula of total Doppler frequency are as follows:
Wherein,It indicates to be rounded downwards.
8. a kind of SAR moving target detection method based on convolutional neural networks according to claim 6, which is characterized in that
Serial number b calculating formula of the target with respect to the angle of synthetic aperture are as follows:
B=mod ((k+1), B)
Wherein, mod () indicates the operation that rems.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910065920.4A CN109709536A (en) | 2019-01-24 | 2019-01-24 | A kind of SAR moving target detection method based on convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910065920.4A CN109709536A (en) | 2019-01-24 | 2019-01-24 | A kind of SAR moving target detection method based on convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109709536A true CN109709536A (en) | 2019-05-03 |
Family
ID=66261740
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910065920.4A Pending CN109709536A (en) | 2019-01-24 | 2019-01-24 | A kind of SAR moving target detection method based on convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109709536A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN112183534A (en) * | 2020-10-07 | 2021-01-05 | 西安电子科技大学 | Moving target intelligent combined detection method based on video synthetic aperture radar |
CN112180338A (en) * | 2020-06-10 | 2021-01-05 | 四川九洲电器集团有限责任公司 | Holographic digital array radar target quantity estimation method and system |
CN112748432A (en) * | 2020-12-25 | 2021-05-04 | 中国科学院空天信息创新研究院 | Method and device for alternately executing strip mode and wide area MTI mode by airborne SAR |
CN113240047A (en) * | 2021-06-02 | 2021-08-10 | 西安电子科技大学 | SAR target recognition method based on component analysis multi-scale convolutional neural network |
CN113253272A (en) * | 2021-07-15 | 2021-08-13 | 中国人民解放军国防科技大学 | Target detection method and device based on SAR distance compressed domain image |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6870501B2 (en) * | 2001-06-26 | 2005-03-22 | Raytheon Company | Digital radio frequency tag |
DE502004001548D1 (en) * | 2003-06-23 | 2006-11-02 | Eads Deutschland Gmbh | SIGNAL EVALUATION METHOD IN A SAR / MTI PULSE RADAR SYSTEM |
CN106228124A (en) * | 2016-07-17 | 2016-12-14 | 西安电子科技大学 | SAR image object detection method based on convolutional neural networks |
CN106443596A (en) * | 2016-09-12 | 2017-02-22 | 电子科技大学 | SVM (support vector machine) space-time adaptive processing method |
CN106597425A (en) * | 2016-11-18 | 2017-04-26 | 中国空间技术研究院 | Radar object positioning method based on machine learning |
CN107015214A (en) * | 2017-06-06 | 2017-08-04 | 电子科技大学 | A kind of space-time adaptive processing method based on management loading |
CN107256396A (en) * | 2017-06-12 | 2017-10-17 | 电子科技大学 | Ship target ISAR characteristics of image learning methods based on convolutional neural networks |
CN107784320A (en) * | 2017-09-27 | 2018-03-09 | 电子科技大学 | Radar range profile's target identification method based on convolution SVMs |
CN108280412A (en) * | 2018-01-12 | 2018-07-13 | 西安电子科技大学 | High Resolution SAR image object detection method based on structure changes CNN |
CN108872988A (en) * | 2018-07-12 | 2018-11-23 | 南京航空航天大学 | A kind of inverse synthetic aperture radar imaging method based on convolutional neural networks |
CN108960190A (en) * | 2018-07-23 | 2018-12-07 | 西安电子科技大学 | SAR video object detection method based on FCN Image Sequence Model |
CN109086799A (en) * | 2018-07-04 | 2018-12-25 | 江苏大学 | A kind of crop leaf disease recognition method based on improvement convolutional neural networks model AlexNet |
-
2019
- 2019-01-24 CN CN201910065920.4A patent/CN109709536A/en active Pending
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6870501B2 (en) * | 2001-06-26 | 2005-03-22 | Raytheon Company | Digital radio frequency tag |
DE502004001548D1 (en) * | 2003-06-23 | 2006-11-02 | Eads Deutschland Gmbh | SIGNAL EVALUATION METHOD IN A SAR / MTI PULSE RADAR SYSTEM |
CN106228124A (en) * | 2016-07-17 | 2016-12-14 | 西安电子科技大学 | SAR image object detection method based on convolutional neural networks |
CN106443596A (en) * | 2016-09-12 | 2017-02-22 | 电子科技大学 | SVM (support vector machine) space-time adaptive processing method |
CN106597425A (en) * | 2016-11-18 | 2017-04-26 | 中国空间技术研究院 | Radar object positioning method based on machine learning |
CN107015214A (en) * | 2017-06-06 | 2017-08-04 | 电子科技大学 | A kind of space-time adaptive processing method based on management loading |
CN107256396A (en) * | 2017-06-12 | 2017-10-17 | 电子科技大学 | Ship target ISAR characteristics of image learning methods based on convolutional neural networks |
CN107784320A (en) * | 2017-09-27 | 2018-03-09 | 电子科技大学 | Radar range profile's target identification method based on convolution SVMs |
CN108280412A (en) * | 2018-01-12 | 2018-07-13 | 西安电子科技大学 | High Resolution SAR image object detection method based on structure changes CNN |
CN109086799A (en) * | 2018-07-04 | 2018-12-25 | 江苏大学 | A kind of crop leaf disease recognition method based on improvement convolutional neural networks model AlexNet |
CN108872988A (en) * | 2018-07-12 | 2018-11-23 | 南京航空航天大学 | A kind of inverse synthetic aperture radar imaging method based on convolutional neural networks |
CN108960190A (en) * | 2018-07-23 | 2018-12-07 | 西安电子科技大学 | SAR video object detection method based on FCN Image Sequence Model |
Non-Patent Citations (6)
Title |
---|
SHAO, C. QU AND J. LI: "A performance analysis of convolutional neural network models in SAR target recognition", 《2017 SAR IN BIG DATA ERA: MODELS, METHODS AND APPLICATIONS (BIGSARDATA), BEIJING》 * |
Z. LIN, K. JI, M. KANG, X. LENG AND H. ZOU: "Deep Convolutional Highway Unit Network for SAR Target Classification With Limited Labeled Training Data", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
Z. LIU, D. K. C. HO, X. XU AND J. YANG: "Moving Target Indication Using Deep Convolutional Neural Network", 《IEEE ACCESS》 * |
李健伟,曲长文,彭书娟,邓兵: "基于卷积神经网络的SAR图像舰船目标检测", 《系统工程与电子技术》 * |
肖定坤: "基于深度网络的SAR图像目标检测技术研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
许强,李伟,PIERRE LOUMBI: "深度卷积神经网络在SAR自动目标识别领域的应用综述", 《电讯技术》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN111123257B (en) * | 2019-12-30 | 2023-03-28 | 西安电子科技大学 | Radar moving target multi-frame joint detection method based on graph space-time network |
CN112180338A (en) * | 2020-06-10 | 2021-01-05 | 四川九洲电器集团有限责任公司 | Holographic digital array radar target quantity estimation method and system |
CN112180338B (en) * | 2020-06-10 | 2022-03-01 | 四川九洲电器集团有限责任公司 | Holographic digital array radar target quantity estimation method and system |
CN112183534A (en) * | 2020-10-07 | 2021-01-05 | 西安电子科技大学 | Moving target intelligent combined detection method based on video synthetic aperture radar |
CN112183534B (en) * | 2020-10-07 | 2023-05-23 | 西安电子科技大学 | Moving target intelligent joint detection method based on video synthetic aperture radar |
CN112748432A (en) * | 2020-12-25 | 2021-05-04 | 中国科学院空天信息创新研究院 | Method and device for alternately executing strip mode and wide area MTI mode by airborne SAR |
CN112748432B (en) * | 2020-12-25 | 2023-09-29 | 中国科学院空天信息创新研究院 | Method and device for alternately executing stripe mode and wide area MTI mode by airborne SAR |
CN113240047A (en) * | 2021-06-02 | 2021-08-10 | 西安电子科技大学 | SAR target recognition method based on component analysis multi-scale convolutional neural network |
CN113240047B (en) * | 2021-06-02 | 2022-12-02 | 西安电子科技大学 | SAR target recognition method based on component analysis multi-scale convolutional neural network |
CN113253272A (en) * | 2021-07-15 | 2021-08-13 | 中国人民解放军国防科技大学 | Target detection method and device based on SAR distance compressed domain image |
CN113253272B (en) * | 2021-07-15 | 2021-10-29 | 中国人民解放军国防科技大学 | Target detection method and device based on SAR distance compressed domain image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109709536A (en) | A kind of SAR moving target detection method based on convolutional neural networks | |
CN109086700B (en) | Radar one-dimensional range profile target identification method based on deep convolutional neural network | |
CN103885057B (en) | Adaptive strain sliding window multi-object tracking method | |
CN112184849B (en) | Intelligent processing method and system for complex dynamic multi-target micro-motion signals | |
CN109407067A (en) | Radar moving targets detection and classification integral method based on time-frequency figure convolutional neural networks | |
Huynh-The et al. | Accurate LPI radar waveform recognition with CWD-TFA for deep convolutional network | |
CN103439697B (en) | Target detection method based on dynamic programming | |
CN107832575A (en) | Band feedback maneuvering target Asynchronous Track Fusion based on pseudo-measurement | |
CN104316914B (en) | Radar target self-adaptation detection method depending on shape parameters | |
CN107396322A (en) | Indoor orientation method based on route matching Yu coding and decoding Recognition with Recurrent Neural Network | |
CN106895905B (en) | A kind of ship-radiated noise detection method | |
CN105785338A (en) | Method for optimizing carrier frequency of frequency-agile radar | |
Meng et al. | Prediction of rice yield via stacked LSTM | |
CN107462882A (en) | A kind of multiple maneuver target tracking methods and system suitable for flicker noise | |
CN104950296A (en) | Robustness nonhomogeneity detecting method based on heavily weighted adaptive power residue | |
CN103777189A (en) | Radar weak target detecting method based on information geometry multiple autoregressive model | |
CN109934101A (en) | Radar clutter recognition method based on convolutional neural networks | |
CN107015214A (en) | A kind of space-time adaptive processing method based on management loading | |
Sun et al. | Maneuvering target tracking using IMM Kalman filter aided by Elman neural network | |
CN105116408A (en) | Ship ISAR image structure feature extraction method | |
CN102298141A (en) | Airborne pulse doppler radar iterative solution range ambiguity method | |
CN108387880A (en) | Detection method of small target under a kind of sea clutter background based on multiple dimensioned oriented Lyapunov indexes | |
Liu et al. | An anti‐jamming method in multistatic radar system based on convolutional neural network | |
CN107102293A (en) | The passive co-located method of unknown clutter estimated based on sliding window integral density | |
Fan et al. | Multifractal correlation analysis of autoregressive spectrum-based feature learning for target detection within sea clutter |
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 |
Application publication date: 20190503 |
|
RJ01 | Rejection of invention patent application after publication |