CN106355196A - Method of identifying synthetic aperture radar image targets based on coupled dictionary learning - Google Patents
Method of identifying synthetic aperture radar image targets based on coupled dictionary learning Download PDFInfo
- Publication number
- CN106355196A CN106355196A CN201610707854.2A CN201610707854A CN106355196A CN 106355196 A CN106355196 A CN 106355196A CN 201610707854 A CN201610707854 A CN 201610707854A CN 106355196 A CN106355196 A CN 106355196A
- Authority
- CN
- China
- Prior art keywords
- dictionary
- model
- formula
- learning
- radar image
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Abstract
The invention discloses a method of identifying synthetic aperture radar image targets based on coupled dictionary learning, which combines the shared dictionaries with the integrated- analytic dictionaries to form a coupled dictionary learning model.The application of analytic dictionaries reduces the complexity of the algorithm, making the model more suitable for real-time systems; and the application of shared dictionaries and structured integrated dictionaries solves the problem of classification accuracy. Compared with prior art, the present invention improves the accuracy of multi-class synthetic aperture radar image recognition, and has higher stability in the recognition of SAR image with depression angle changing.
Description
Technical field
The present invention relates to the technical field such as radar image target recognition, especially one kind can effectively reduce algorithm complex,
Improve the diameter radar image target knowledge based on coupling dictionary learning of radar image target recognition accuracy rate and stability
Other method.
Background technology
With the continuous maturation of radar (sar) imaging technique, sar image is more and more wider in military, civil area application
General.It is applied to radar image Motion parameters (sar atr) at present and mainly comprise two masters with conventional Activity recognition technology
Want process, i.e. training process and test process.Training process has three specific processing links, is pretreatment training sample respectively
(filtering clutter etc.), extract the feature (the more commonly used is textural characteristics) of training sample, set up sorter model.Test process
Equally there are three processing links, be pretreatment test sample respectively, extract the feature of test sample, obtained using the training stage
Sorter model carries out classification prediction to test sample.Through development in recent years although dictionary learning miscellaneous is opened
Begin to be widely used in every field, but in sar images steganalysis, the application of dictionary learning is not quite varied.
Jayaraman. propose a kind of method based on rarefaction representation;Dong proposes one kind using sparse coding and single-gene signal
Msrc method, these methods all apply the rarefaction representation thought in dictionary learning, and all obtain in radar image identification field
Certain success, but the accuracy rate of identification and stability need to be improved.
Content of the invention
The present invention is to solve the above-mentioned technical problem existing for prior art, providing one kind can effectively reduce algorithm multiple
The diameter radar image mesh based on coupling dictionary learning of miscellaneous degree, raising radar image target recognition accuracy rate and stability
Mark recognition methodss.
The technical solution of the present invention is: a kind of diameter radar image target recognition based on coupling dictionary learning
Method, including setting up sorter model it is characterised in that setting up sorter model in accordance with the following steps:
Step s1, carries out pretreatment to training set sample, obtains textural characteristics;
Step s2, Optimization Learning obtains analyticity dictionary p and mixed type dictionary h;
If sample, dictionary, sparse coding matrix, then comprehensive dictionary learning model be:
(1)
In formula (1)For constant,It is used to lift the bound term distinguishing accuracy rate;
If mixed type dictionary is,It is shared dictionary,
Selector, jth arranges as follows:
The i-th of mixed type dictionary arranges,,
Then the model of coupling dictionary learning is:
(2)
In formula (2)WithIt is constant;
Introduce an intermediate variable a, then above formula (2) is:
(3)
Update a, h and p respectively,
1) h and p fixes, and updates a, that is, require:
(4)
The closed solutions obtaining a are as follows:
(5)
2) a and h fixes, and updates p, that is, require:
(6)
The closed solutions obtaining dictionary p are as follows:
(7)
3) a and p fixes, and updates dictionary h, that is, require:
(8)
Can be in the hope of the optimal solution of (8) formula using admm algorithm:
(9)
Step s3, determines sorter model;
Represent the sorter model based on mixing dictionary model algorithm with reconstructed error, as the formula:
(10).
Shared dictionary is combined by the present invention with comprehensive-analytical type dictionary, constitutes coupling dictionary learning model.Parsing
The application of type-word allusion quotation reduce algorithm complexity so that model more suitable for real-time system;Shared dictionary and structuring
The application of comprehensive dictionary solve the problems, such as classification accuracy.Compared to existing method, the present invention improves multiclass synthesis
Aperture radar image recognition accuracy rate, and have higher stability in the sar image recognition of depression angle change.
Specific embodiment
The identification method of image target of synthetic aperture radar of the present invention, like the prior art, including i.e. training process and
Test process, with prior art except that setting up sorter model in accordance with the following steps:
Step s1, carries out pretreatment, acquisition textural characteristics to training set sample:
Choose sample class to be trained first, then unify for image cropping to become formed objects (as unification is cut to 80*80
Pixel), note ensureing the integrity of target area.Image after processing is carried out feature extraction process, extracts branching
Reason feature.The textural characteristics that each image is obtained are treated as a column vector, and the row by the textural characteristics of all images
Vector is merged into a set, because obtained column vector dimension is larger, using principal component analysiss (pca) method to texture
Eigenmatrix carries out dimension-reduction treatment;
Step s2, Optimization Learning obtains analyticity dictionary p and mixed type dictionary h;
If sample, dictionary, sparse coding matrix, then comprehensive dictionary learning model be:
(1)
In formula (1)For constant,It is used to lift the bound term distinguishing accuracy rate;
If mixed type dictionary is,It is shared dictionary,
Selector, jth arranges as follows:
The i-th of mixed type dictionary arranges,,
Then the model of coupling dictionary learning is:
(2)
In formula (2)WithIt is constant;
Introduce an intermediate variable a, then above formula (2) is:
(3)
Update a, h and p respectively,
1) h and p fixes, and updates a, that is, require:
(4)
The closed solutions obtaining a are as follows:
(5)
2) a and h fixes, and updates p, that is, require:
(6)
The closed solutions obtaining dictionary p are as follows:
(7)
3) a and p fixes, and updates dictionary h, that is, require:
(8)
Can be in the hope of the optimal solution of (8) formula using admm algorithm:
(9)
Variable a and p all can get closed solutions, and asking h to obtain optimum results using admm method is rapid convergence;
Step s3, determines sorter model:
In coupling dictionary learning algorithm, train the sub- dictionary obtainingOnly numerical value is produced to the sample of kth apoplexy due to endogenous wind larger
Code coefficient, the mapping coefficient very little to the sample beyond kth class, or even close to zero.Meanwhile, train the mixing obtaining
DictionaryCan be according to code coefficientSample in k-th classification is reconstructed, reconstructed error nowLess.Further, sinceWhen,Value very little, thusCan not be used for reconstructing i-th
The sample set of class, so in classification i sample reconstructed errorWeight much larger than sample in classification k
Structure error.
In test phase, if query sample belongs to classification k, by analytical type dictionaryMap the coding vector obtaining
(it is expressed as) ratio is more significant, and byThe sub- dictionary of analytical typeMap the coding vector then very little obtaining.Cause
And, by the sub- dictionary of analytical typeThe reconstructed error producingIt is much smaller thanThe sub- dictionary of analytical type
The reconstructed error producing.Obviously, the reconstructed error of particular category may be used to determine the classification of test sample y
Label, you can to represent the grader based on mixing dictionary model algorithm with reconstructed error, as the formula:
(10)
From formula (10), based on coupling the grader of dictionary learning model after training obtains dictionary h and p of optimum, with regard to energy
Obtain sorter model, such that it is able to Classification and Identification is carried out to test image.
Experimental example:
Method proposed by the present invention is applied to a disclosed radar image data storehouse, i.e. mstar data base.Mstar data
Storehouse is U.S. national defense beforehand research programme division and the common subsidy of Air Force Research Laboratory (darpa/afrl).Wherein comprise substantial amounts of reality
Survey sar ground static target data, including military combat tank, panzer etc..
Mstar data set has very important significance in radar image target identification technology, and it is sar image object
The public database of recognition performance assessment.Target image comprises multiclass vehicle model, and the radar image of every kind of vehicle model comprises
Multiple angles of pitch.Due to the particularity of radar image, multiple pictorial informations are comprised for each angle of pitch, wherein contain thing
The radar image of different visual angles in 0 to 360 degree for the body.For making experiment more convenient, experiment is broadly divided into two parts, divides below
Do not introduce.
Experiment 1: the main checking feasibility in multiclass radar target recognition problem for the present invention of experiment 1.Select ten classes not
It is used for testing with species military vehicle.For same class vehicle, choose depression angle be 17 degree as training dataset, selection is bowed
Visual angle is 15 degree as test data set.Wherein bmp2 and t72 has three kinds of models, different model be configured with nuance,
Select the standard model of this two classes vehicle in training set, in test set, select remaining model.The concrete data such as table 1 of experiment 1
Shown.
Table 1
Obtain sorter model according to embodiment of the present invention training, then test data set is input in sorter model and carries out
The average recognition rate of target recognition, the embodiment of the present invention and its method is to such as table 2.
Table 2
Experiment 2: experiment 2 is mainly used in proving when the depression angle of radar image changes, the present invention has preferable identification steady
Qualitative.The classification that experiment is selected is respectively: 2s1, brdm and zsu234.In mstar, the sample of these three classifications all contains four
Plant the sar image of depression angle, be 15 degree, 17 degree, 30 degree and 45 degree respectively.We select 17 degree as training sample, its excess-three
The sample of individual angle is all respectively as test set.Specific experiment data is shown in Table 3.
Table 3
Specific preprocessing process is consistent with experiment 1, obtains sorter model according to embodiment of the present invention training, then will test
Data set is input in sorter model and carries out target recognition, and the average recognition rate of the embodiment of the present invention and its method is to such as table
2.
Table 4
Radar image affected by depression angle larger, generally after depression angle occurs more significant change, radar image
Obvious change can occur, so the radar target recognition problem under different depression angle is difficult to solve.Real with other from table 4
In the comparing result of proved recipe method it can be seen that
1) test sample collection and training sample set depression angle relatively when (15 degree and 17 degree closely), can be approximate
Think depression angle approximately equal.Now identification species less (only three classes), the discrimination of each method is all higher, experimental result
Relatively;
2) when there is more significant change when the depression angle of test sample collection (30 degree have changed 13 degree compared with 17 degree), respectively
The experimental result of individual method has obvious change, and recognition accuracy has different degrees of decline, compared to additive method, this
Bright recognition accuracy has some superiority;
3) when there is acute variation when the depression angle of test sample collection (45 degree compared with 17 degree acute variation 28 degree), each
The experimental result of method also there occurs significant changes, and recognition accuracy declines substantially.Now compared to additive method, the present invention's
Experimental result has obviously advantage, and accuracy rate has very high lifting.Illustrate the present invention compared to some currently commonly used
There is preferable robustness in depression angle when method changes.
Claims (1)
1. a kind of identification method of image target of synthetic aperture radar based on coupling dictionary learning, including setting up sorter model,
It is characterized in that setting up sorter model in accordance with the following steps:
Step s1, carries out pretreatment to training set sample, obtains textural characteristics;
Step s2, Optimization Learning obtains analyticity dictionary p and mixed type dictionary h;
If sample, dictionary, sparse coding matrix, then comprehensive dictionary learning model be:
(1)
In formula (1)For constant,It is used to lift the bound term distinguishing accuracy rate;
If mixed type dictionary is,It is shared dictionary,
Selector, jth arranges as follows:
The i-th of mixed type dictionary arranges,,
Then the model of coupling dictionary learning is:
(2)
In formula (2)WithIt is constant;
Introduce an intermediate variable a, then above formula (2) is:
(3)
Update a, h and p respectively,
H and p fixes, and updates a, that is, require:
(4)
The closed solutions obtaining a are as follows:
(5)
A and h fixes, and updates p, that is, require:
(6)
The closed solutions obtaining dictionary p are as follows:
(7)
A and p fixes, and updates dictionary h, that is, require:
(8)
Can be in the hope of the optimal solution of (8) formula using admm algorithm:
(9)
Step s3, determines sorter model;
Represent the sorter model based on mixing dictionary model algorithm with reconstructed error, as the formula:
(10).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610707854.2A CN106355196A (en) | 2016-08-23 | 2016-08-23 | Method of identifying synthetic aperture radar image targets based on coupled dictionary learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610707854.2A CN106355196A (en) | 2016-08-23 | 2016-08-23 | Method of identifying synthetic aperture radar image targets based on coupled dictionary learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106355196A true CN106355196A (en) | 2017-01-25 |
Family
ID=57844454
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610707854.2A Pending CN106355196A (en) | 2016-08-23 | 2016-08-23 | Method of identifying synthetic aperture radar image targets based on coupled dictionary learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106355196A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107220936A (en) * | 2017-05-26 | 2017-09-29 | 首都师范大学 | A kind of image super-resolution reconstructing method and system |
CN110244299A (en) * | 2019-06-21 | 2019-09-17 | 西安交通大学 | A kind of distributed method that the SAR image based on ADMM is restored |
CN110275166A (en) * | 2019-07-12 | 2019-09-24 | 中国人民解放军国防科技大学 | ADMM-based rapid sparse aperture ISAR self-focusing and imaging method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102024152A (en) * | 2010-12-14 | 2011-04-20 | 浙江大学 | Method for recognizing traffic sings based on sparse expression and dictionary study |
CN102609681A (en) * | 2012-01-12 | 2012-07-25 | 北京大学 | Face recognition method based on dictionary learning models |
-
2016
- 2016-08-23 CN CN201610707854.2A patent/CN106355196A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102024152A (en) * | 2010-12-14 | 2011-04-20 | 浙江大学 | Method for recognizing traffic sings based on sparse expression and dictionary study |
CN102609681A (en) * | 2012-01-12 | 2012-07-25 | 北京大学 | Face recognition method based on dictionary learning models |
Non-Patent Citations (2)
Title |
---|
ZHIHAO XU ET AL.: "Hybrid Dictionary Learning for JPEG Steganalysis", 《2015 ASIA-PACIFIC SIGNAL AND INFORMATION PROCESSING ASSOCIATION ANNUAL SUMMIT AND CONFERENCE (APSIPA)》 * |
郭艳卿 等: "局部保持"字典对"学习算法及其应用", 《信息安全研究》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107220936A (en) * | 2017-05-26 | 2017-09-29 | 首都师范大学 | A kind of image super-resolution reconstructing method and system |
CN110244299A (en) * | 2019-06-21 | 2019-09-17 | 西安交通大学 | A kind of distributed method that the SAR image based on ADMM is restored |
CN110275166A (en) * | 2019-07-12 | 2019-09-24 | 中国人民解放军国防科技大学 | ADMM-based rapid sparse aperture ISAR self-focusing and imaging method |
CN110275166B (en) * | 2019-07-12 | 2021-03-19 | 中国人民解放军国防科技大学 | ADMM-based rapid sparse aperture ISAR self-focusing and imaging method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109086700B (en) | Radar one-dimensional range profile target identification method based on deep convolutional neural network | |
CN103955702B (en) | SAR image terrain classification method based on depth RBF network | |
Yao et al. | Application of convolutional neural network in classification of high resolution agricultural remote sensing images | |
CN112446388A (en) | Multi-category vegetable seedling identification method and system based on lightweight two-stage detection model | |
CN111191583B (en) | Space target recognition system and method based on convolutional neural network | |
CN109934166A (en) | Unmanned plane image change detection method based on semantic segmentation and twin neural network | |
CN108416318A (en) | Diameter radar image target depth method of model identification based on data enhancing | |
CN108960330A (en) | Remote sensing images semanteme generation method based on fast area convolutional neural networks | |
CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
CN105005789B (en) | A kind of remote sensing images terrain classification method of view-based access control model vocabulary | |
CN104751166A (en) | Spectral angle and Euclidean distance based remote-sensing image classification method | |
CN105718963B (en) | SAR image classification method based on elongated incremental extreme learning machine | |
CN107944370A (en) | Classification of Polarimetric SAR Image method based on DCCGAN models | |
CN111126332B (en) | Frequency hopping signal classification method based on contour features | |
CN105913081A (en) | Improved PCAnet-based SAR image classification method | |
CN108447057A (en) | SAR image change detection based on conspicuousness and depth convolutional network | |
CN101807258A (en) | SAR (Synthetic Aperture Radar) image target recognizing method based on nuclear scale tangent dimensionality reduction | |
CN110060273A (en) | Remote sensing image landslide plotting method based on deep neural network | |
Liu et al. | Coastline extraction method based on convolutional neural networks—A case study of Jiaozhou Bay in Qingdao, China | |
CN106355196A (en) | Method of identifying synthetic aperture radar image targets based on coupled dictionary learning | |
CN104680169A (en) | Semi-supervised diagnostic characteristic selecting method aiming at thematic information extraction of high-spatial resolution remote sensing image | |
CN110969121A (en) | High-resolution radar target recognition algorithm based on deep learning | |
CN110852358A (en) | Vehicle type distinguishing method based on deep learning | |
CN113435254A (en) | Sentinel second image-based farmland deep learning extraction method | |
CN114241307A (en) | Synthetic aperture radar aircraft target identification method based on self-attention network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170125 |