CN101894269B - Multi-classifier system-based synthetic aperture radar automatic target recognition method - Google Patents
Multi-classifier system-based synthetic aperture radar automatic target recognition method Download PDFInfo
- Publication number
- CN101894269B CN101894269B CN2010102309488A CN201010230948A CN101894269B CN 101894269 B CN101894269 B CN 101894269B CN 2010102309488 A CN2010102309488 A CN 2010102309488A CN 201010230948 A CN201010230948 A CN 201010230948A CN 101894269 B CN101894269 B CN 101894269B
- Authority
- CN
- China
- Prior art keywords
- aperture radar
- target
- training
- image
- pca
- 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.)
- Expired - Fee Related
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Radar Systems Or Details Thereof (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a synthetic aperture radar automatic target recognition method which belongs to the target recognition field and mainly solves the problem that the space complexity of the existing synthetic aperture radar automatic target recognition technology is higher and single classifier has low recognition rate. The method comprises the following recognition steps: preprocessing, extracting characteristics, training classifiers and identifying target, wherein the step of extracting characteristics is to extract PCA characteristics of the synthetic aperture radar image, elliptic Fourier descriptor and two-dimensional Fourier transform; the step of training classifiers is based on the extracted three characteristics to separately use K-nearest neighbor method, support vector machine and MINACE filter theory to train three classifiers; and the step of identifying target is to input the extracted synthetic aperture radar image to be identified in the trained three classifiers for classification and finally adopting the Dempster-Shafer evidence theory to fuse the recognition results of the three classifiers. The method has the advantages of high recognition rate and low space complexity and can be used in the target tracking of the military or civilian field.
Description
Technical field
The invention belongs to the Target Recognition field, particularly synthetic-aperture radar automatic target identification can be used for the synthetic-aperture radar automatic target identification of military or civil area.
Background technology
At military field or civil area, usually need be, differentiate such as the classification or the attribute of tank, automobile to target, so that target is followed the tracks of, judge its intention etc.The identification of synthetic-aperture radar automatic target is called for short SAR ATR, lets computing machine discern the technology of unknown object according to prior imformation exactly.Let computing machine from existing data or information, learn a model, then the target of the unknown is discerned, i.e. classification, this is the problem that automatic target identification needs solution.Synthetic-aperture radar SAR is widely used owing to it has many good performances, is the important source that obtains target information.In recent years, the identification of synthetic-aperture radar automatic target became an one of advanced subject in military and civilian field.
Since U.S. Department of Defense's ARPA issue can be used for the MSTAR database of object identification test, many target identification technologies to this database were suggested.Synthetic aperture radar automatic target recognition method mainly is divided three classes at present:
(1) based on the method for template matches, such as the MSE sorter, it is that every class targets makes up several templates, calculates the similarity between test sample book and these templates, then with this test sample book be grouped into the maximum template of its similarity under class in.The advantage of this method is simple, yet owing to need to make up many templates, and the size of template is general and original image big or small identical, thereby its space complexity is bigger.
(2) based on the method for model, this method is carried out modeling from statistical angle to image, and existing model has rayleigh model, condition Gauss model etc.For a test sample book, according to its ownership of Bayes's maximum a posteriori probability decision.The same with preceding method, there is the higher shortcoming of space complexity equally in this method.
(3) based on the method for pattern-recognition, this method mainly is the correlation technique solution synthetic-aperture radar automatic target recognition technology by area of pattern recognition.Many technology such as the supporting vector machine in the pattern-recognition, K-neighbour have been used the Target Recognition field.Because adopt the less characteristic of dimension, this method can overcome preceding two kinds of shortcomings that the method space complexity is higher, but its discrimination is lower.
Summary of the invention
The objective of the invention is to overcome above-mentioned existing methods defective, propose a kind of synthetic-aperture radar automatic target recognition system and method,, improve the identification of targets rate to reduce space complexity based on multi-classifier system.
For realizing above-mentioned purpose, the present invention is based on the synthetic aperture radar automatic target recognition method of multi-classifier system, comprising:
Pre-treatment step: synthetic aperture radar training image is carried out the log conversion, the pre-service that normalization and objective contour extract;
Characteristic extraction step: pretreated synthetic-aperture radar training image is carried out principal component analysis PCA respectively, these three kinds of Feature Extraction of oval Fourier descriptor characteristic and two-dimensional Fourier transform;
Sorter training step: train three sorters, promptly adopt the K-nearest neighbor algorithm to use a K-nearest neighbour classification of PCA features training device; Adopt the supporting vector machine algorithm to use a K-nearest neighbour classification of oval Fourier descriptor features training device, adopt the MINIACE algorithm filter to use two-dimensional Fourier transform training MINACE wave filter;
Target Recognition step: extract the PCA characteristic of diameter radar image to be identified, oval Fourier descriptor characteristic and two-dimensional Fourier transform respectively; And output in corresponding K-nearest neighbour classification device, supporting vector machine sorter and the MINIACE wave filter, each sorter provides a recognition result; After the discount operation and of the recognition result fusion of Dempster rule of combination through the Dempster-Shafer evidence theory, obtain final recognition result with described three sorters.
For realizing above-mentioned purpose, the present invention is based on the synthetic-aperture radar automatic target recognition system of multi-classifier system, comprising:
Pretreatment unit: be used for synthetic aperture radar training image is carried out the log conversion pre-service that normalization and objective contour extract;
Feature deriving means: be used for pretreated synthetic-aperture radar training image is carried out principal component analysis PCA respectively these three kinds of Feature Extraction of oval Fourier descriptor characteristic and two-dimensional Fourier transform;
Sorter trainer: be used to train three sorters, promptly adopt the K-nearest neighbor algorithm to use a K-nearest neighbour classification of PCA features training device; Adopt the supporting vector machine algorithm to use a K-nearest neighbour classification of oval Fourier descriptor features training device, adopt the MINIACE algorithm filter to use two-dimensional Fourier transform training MINACE wave filter;
Target Identification Unit: be used for extracting respectively the PCA characteristic of diameter radar image to be identified, oval Fourier descriptor characteristic and two-dimensional Fourier transform; And output in corresponding K-nearest neighbour classification device, supporting vector machine sorter and the MINIACE wave filter, each sorter provides a recognition result; After the discount operation and of the recognition result fusion of Dempster rule of combination through the Dempster-Shafer evidence theory, obtain final recognition result with described three sorters.
The present invention has the following advantages compared with prior art:
1. the present invention is owing to use the features training sorter that extracts from image, rather than directly uses original image, thereby space complexity is lower.
2. the present invention is owing to trained K-nearest neighbour classification device, supporting vector machine sorter and three kinds of sorters of MINACE wave filter; And merge through the recognition result of Dempster-Shafer evidence theory to these three kinds of sorters, have higher discrimination than single sorter.
Description of drawings
Fig. 1 is a synthetic-aperture radar automatic target recognition system synoptic diagram of the present invention;
Fig. 2 is a synthetic aperture radar automatic target recognition method process flow diagram of the present invention.
Embodiment
With reference to Fig. 1, multi-classifier system of the present invention comprises: pretreatment unit, feature deriving means, sorter trainer and Target Identification Unit.Wherein:
Pretreatment unit carries out the log conversion to synthetic aperture radar training image, the pre-service that normalization and objective contour extract; To the different character extractive technique; Preprocess method is also different, before synthetic-aperture radar is carried out the PCA feature extraction, with the maximal value of the pixel value of every width of cloth image with its normalization; Be about to the pixel value of each pixel value of image, to avoid the calculating of big numerical value divided by maximum; Before synthetic-aperture radar is extracted oval Fourier descriptor, use the constant false alarm rate Threshold Segmentation to isolate the target area earlier, extract objective contour through the canny rim detection then; Before diameter radar image is carried out two-dimensional Fourier transform, carry out the log conversion earlier, carry out energy normalized then.
Feature deriving means extracts these three kinds of characteristics of main PCA characteristic, oval Fourier descriptor characteristic and two-dimensional Fourier transform respectively to pretreated synthetic-aperture radar training image; When extracting the PCA characteristic, the number of principal component is chosen according to 90% energy; When extracting oval Fourier descriptor characteristic, get 7 rank coefficients totally 26 dimensional features.
The sorter trainer adopts three kinds of different algorithms, trains three different sorters respectively, promptly uses a K-nearest neighbour classification of PCA features training device with the K-nearest neighbor algorithm, and the value of K is 3; Use a K-nearest neighbour classification of oval Fourier descriptor features training device with the supporting vector machine algorithm, the supporting vector machine algorithm uses the LIBSVM routine package to realize that the optimized parameter of supporting vector machine obtains through exhaustive search; Use two-dimensional Fourier transform training MINACE wave filter with the MINIACE algorithm filter, the training image of each class targets is divided into 6 parts according to 60 degree position angles, on each part, make up a wave filter, each class has 6 wave filters like this.
Target Identification Unit; Extract the PCA characteristic of diameter radar image to be identified, oval Fourier descriptor characteristic and two-dimensional Fourier transform respectively; And output in corresponding K-nearest neighbour classification device, supporting vector machine sorter and the MINIACE wave filter, each sorter provides a recognition result; After the discount operation and of the recognition result fusion of Dempster rule of combination through the Dempster-Shafer evidence theory, obtain final recognition result with described three sorters.
With reference to Fig. 2, the concrete performing step of synthetic aperture radar automatic target recognition method of the present invention comprises:
Step 1 is carried out the log conversion to synthetic aperture radar training image, the pre-service that normalization and objective contour extract; To the different character extractive technique; Preprocess method is also different, before synthetic-aperture radar is carried out the PCA feature extraction, with the maximal value of the pixel value of every width of cloth image with its normalization; Be about to the pixel value of each pixel value of image, to avoid the calculating of big numerical value divided by maximum; Before synthetic-aperture radar is extracted oval Fourier descriptor, use the constant false alarm rate Threshold Segmentation to isolate the target area earlier, extract objective contour through the canny rim detection then; Before synthetic-aperture radar is carried out two-dimensional Fourier transform, carry out the log conversion earlier, carry out energy normalized then.
Step 2 is extracted these three kinds of characteristics of PCA characteristic, oval Fourier descriptor characteristic and two-dimensional Fourier transform respectively to pretreated synthetic-aperture radar training image.
(2.1) PCA feature extraction:
Existing principal component method is adopted in PCA feature extraction of the present invention, and concrete steps are following:
(2.1a) pretreated each width of cloth diameter radar image is pulled into a dimensional vector, its length is L, and the value of L equals total number of each width of cloth training image pixel;
(2.1b) calculate sample average
X wherein
iRepresent the corresponding column vector of i width of cloth training image, R is the training image sum;
(2.1d) sample covariance matrix C is carried out feature decomposition, obtain L eigenvalue
1, λ
2... λ
LWith the characteristic of correspondence vector v
1, v
1... v
L
(2.1f) L eigenwert pressed rank order from big to small, k biggest characteristic value characteristic of correspondence vector formed projection matrix V={v before selecting
1, v
1... v
k, wherein the value of k is chosen according to 90% energy, the k that promptly chooses a biggest characteristic value with account for 90% of all eigenwert summations;
(2.1g) that each width of cloth training image is corresponding column vector is carried out projection on projection matrix V:
Y=V
TX
i,i=1,2,...R
Y is the PCA characteristic of extraction;
(2.2) oval Fourier descriptor feature extraction:
The realization of this step; According to document " Shape-based Recognition of Targets in SyntheticAperture Radar Images using Elliptical Fourier Descriptors, Proceedings of SPIE, vol.6967; 2008 ", specifically comprise as follows:
The objective contour that (2.2a) step 1 is extracted is expressed as the form of following closed curve:
[v(t)]=[x(t),y(t)],t∈[0,2π)
Wherein t is the phasing degree, x (t), and y (t) is illustrated respectively in the value of t place horizontal ordinate and ordinate;
(2.2b) closed curve of being represented by following formula is carried out conversion by following formula:
F wherein
kBe a matrix of coefficients, it has four element a
k, b
k, c
k, d
k, wherein subscript k is an exponent number;
(2.2c) k is got 6 from 0, obtain seven rank coefficient F
0, F
1, F
2, F
3, F
4, F
5, F
6, this seven rank coefficient is called as oval Fourier descriptor.
(2.3) two-dimensional Fourier transform:
The formula that adopts following two-dimensional Fourier transform is got in two-dimensional Fourier transform of the present invention:
u=0,1,...M-1;v=0,1,...N-1
Wherein (x y) represents a secondary training image, x to f; Y is respectively the abscissa value and the ordinate value of image, and M and N are respectively the height and the width of image, and symbol e represents exponent arithmetic; F (u; V) be the coefficient of two-dimensional Fourier transform, u, v are respectively the abscissa value and the ordinate value of two-dimensional Fourier transform.
Step 3 adopts three kinds of different algorithms, trains three different sorters respectively.
(3.1) adopt the K-nearest neighbor algorithm to use a K-nearest neighbour classification of PCA features training device, the K value is 3;
(3.2) use supporting vector machine sorter of oval Fourier descriptor features training with the supporting vector machine algorithm, the supporting vector machine algorithm uses the LIBSVM routine package to realize that the optimized parameter of supporting vector machine obtains through exhaustive search;
(3.3) use two-dimensional Fourier transform training MINACE wave filter with the MINIACE algorithm filter, the training image of each class targets is divided into 6 parts according to 60 degree position angles, on each part, make up a wave filter, each class has 6 wave filters like this.
Step 4 is discerned the target of the unknown.
(4.1) diameter radar image of treating recognition objective carries out the log conversion, the pre-service that normalization and objective contour extract;
(4.2) extract the PCA characteristic of diameter radar image to be identified, oval Fourier descriptor characteristic and two-dimensional Fourier transform respectively; The method for distilling of wherein oval Fourier descriptor characteristic and two-dimensional Fourier transform is identical with step 2, but when extracting the PCA characteristic, only needs image to be identified is pulled into a column vector, then the projection matrix V projection of this column vector in step 2.1f is got final product;
PCA characteristic, oval Fourier descriptor characteristic and the two-dimensional Fourier transform of (4.3) step (4.2) being extracted is input in corresponding K-nearest neighbour classification device, supporting vector machine sorter and the MINIACE wave filter, and each sorter provides a recognition result:
ψ
i={s
i1,...s
ij...s
iM},i=1,2,3
ψ
1, ψ
2, ψ
3Represent the recognition result of K-nearest neighbour classification device, supporting vector machine sorter and MINACE wave filter respectively, s
IjPresentation class device ψ
iUnknown object is assigned to j class C
jProbability;
(4.4) after the discount operation and the recognition result fusion of Dempster rule of combination with described three sorters through the Dempster-Shafer evidence theory, obtain final recognition result, concrete performing step is following:
(4.4a) definition framework of identification
θ={C
1,C
2,...C
j,...C
M};
C wherein
jRepresent the class label of j class targets, M is total classification number of target;
(4.4b) use the recognition result of step (4.3) to define probability assignment function m
1, m
2, m
3:
Wherein A is subclass and the A ≠ C of θ
j, j=1 ... M; m
1, m
2, m
3Represent an evidence respectively;
(4.4c) calculate the factor of conflicting of each evidence and other evidences, obtain the collision vector of each evidence:
K
1=(k
12,k
13)
K
2=(k
21,k
23)
K
3=(k
31,k
32)
i wherein; J=1; The conflict factor of 2,3 i evidences of expression and j evidence; B and C are the subclass of θ;
Each collision vector that (4.4d) step (4.4c) is calculated carries out normalization:
K
1=(k
12,k
13)/(k
12+k
13)
K
2=(k
21,k
23)/(k
21+k
23);
K
3=(k
31,k
32)/(k
31+k
32)
(4.4e) entropy of normalization collision vector in the calculation procedure (4.4d):
H
1=k
12n(k
12)+k
13ln(k
13)
H
2=k
21ln(k
21)+k
13ln(k
23);
H
3=k
31ln(k
31)+k
13ln(k
32)
(4.4f) inverse of entropy in the calculation procedure (4.4e):
(4.4g) calculate the weight of each evidence:
(4.4h) use above-mentioned weight that each evidence is carried out the discount operation:
(4.4i) evidence that finishes the discount operation is made up:
(4.4j) the evidence m after the combination is a vector that comprises M element, i.e. m={m (C
1), m (C
2) ..., m (C
j) ... m (C
M), m (C wherein
j) the expression unknown object assigned to the probability of j class, unknown object is identified as the target in the greatest member corresponding class of m, promptly accomplished identifying.
Effect of the present invention can further specify through following emulation experiment:
1. used data of emulation
The used data of emulation are the MSTAR data.These data are the static military targets in actual measurement SAR ground that provided by U.S. DARPA/AFRL MSTAR project team.The MSTAR data are collected by X-band, 0.3m * 0.3m high score rate bunching type synthetic aperture radar.Target image size is 128 * 128.Data have been divided into training data and test data two parts, and wherein training data obtains when the angle of pitch is 17 °, and test data obtains when the angle of pitch is 15 °.The part of this database has only been used in this emulation, and only uses tertiary target: BMP2, BTR70, T72.Wherein BMP2 and T72 respectively have three models.Sn-812, the sn-s7 of the sn-9563 of BMP2, sn-9566 and T72 do not participate in the training stage, but participate in test phase.Table 1 has provided the details of data.
The used data of table 1 emulation
2. emulation content
Emulation content 1 is used in method of the present invention and carries out experiment for target identification on the MSTAR data;
Emulation content 2 uses the supporting vector machine sorter based on oval Fourier descriptor characteristic on the MSTAR data, to carry out Target Recognition;
Emulation content 3 uses the K-nearest neighbour classification device based on the PCA characteristic on the MSTAR data, to carry out experiment for target identification;
Emulation content 4 uses the MINACE wave filter based on two-dimensional Fourier transform on the MSATR data, to carry out experiment for target identification.
3. simulation result
Table 2 has provided the recognition result of the Adaboost method of emulation content 1 in the simulation result of emulation content 4 and the template matching method in the existing document " n 2; pp.643-654; April 2001 for Support VectorsMachines for SAR Automatic Target Recognition; IEEE Transactions on Aerospace andElectronic Systems, vol.37 " and " vol.43; No.1, January 2007 for Synthetic Aperture Radar Target Recognition Using Adaptive Boosting, IEEETransactions on Aerospace and Electronic Systems ".
Table 2 simulation result
Can find out that from table 2 discrimination of the present invention is the highest, wherein on this class targets of BTR70, can reach 100% discrimination; The discrimination of recognition methods of the present invention on each class targets all obviously is superior to supporting vector machine, K-neighbour, MINACE wave filter and template matching method; The present invention simultaneously also is superior to another kind of integrated approach-Adaboost; The present invention not only exceeds about one percentage point than Adaboost on average recognition rate; Need train a large amount of sorters owing to Adaboost in addition, thereby the present invention will be well below Adaboost on time complexity.
Claims (3)
1. synthetic aperture radar automatic target recognition method comprises:
Pre-treatment step: the diameter radar image to synthetic aperture radar training image and target to be identified carries out the log conversion, the pre-service that normalization and objective contour extract;
Characteristic extraction step: pretreated synthetic-aperture radar training image is carried out principal component analysis PCA respectively, these three kinds of Feature Extraction of oval Fourier descriptor characteristic and two-dimensional Fourier transform;
Sorter training step: train three sorters, promptly adopt the K-nearest neighbor algorithm to use a K-nearest neighbour classification of PCA features training device; Adopt the supporting vector machine algorithm to use supporting vector machine sorter of oval Fourier descriptor features training, adopt the MINIACE algorithm filter to use two-dimensional Fourier transform training MINIACE wave filter;
Target Recognition step: extract the PCA characteristic of diameter radar image to be identified, oval Fourier descriptor characteristic and two-dimensional Fourier transform respectively; And output in corresponding K-nearest neighbour classification device, supporting vector machine sorter and the MINIACE wave filter, each sorter provides a recognition result; After the discount operation and of the recognition result fusion of Dempster rule of combination through the Dempster-Shafer evidence theory, obtain final recognition result with described three sorters;
The PCA characteristic of described extraction diameter radar image to be identified only needs image to be identified is pulled into a column vector, then this column vector is got final product to the projection matrix V projection that pretreated synthetic-aperture radar training image carries out in the PCA leaching process.
2. target identification method according to claim 1; Wherein the discount operation of passing through the Dempster-Shafer evidence theory described in the Target Recognition step and Dempster rule of combination merge the recognition result of described three sorters, carry out as follows:
(2a) definition framework of identification
θ={C
1,C
2,...C
j,...C
M}
C wherein
jRepresent the class label of j class targets, M is total classification number of target;
(2b) the output result with K-nearest neighbour classification device, supporting vector machine sorter and MINIACE wave filter is designated as ψ respectively
1, ψ
2, ψ
3, these output result's form is following:
ψ
i={s
i1,...s
ij...s
iM},i=1,2,3
S wherein
IjPresentation class device ψ
iTo C
jDegree of support;
Definition probability assignment function m
1, m
2, m
3:
Wherein A is subclass and the A ≠ C of θ
j, j=1 ... M; m
1, m
2, m
3Represent an evidence respectively;
(2c) calculate the factor of conflicting of each evidence and other evidences, obtain the collision vector of each evidence:
K
1=(k
12,k
13)
K
2=(k
21,k
23)
K
3=(k
31,k
32)
i wherein; J=1; The conflict factor of 2,3 i evidences of expression and j evidence; B and C are the subclass of θ;
Each collision vector that (2d) step (2c) is calculated carries out normalization:
K
1=(k
12,k
13)/(k
12+k
13)
K
2=(k
21,k
23)/(k
21+k
23)
K
3=(k
31,k
32)/(k
31+k
32)
(2e) entropy of normalization collision vector in the calculation procedure (2d):
H
1=k
12ln(k
12)+k
13ln(k
13)
H
2=k
21ln(k
21)+k
13ln(k
23)
H
3=k
31ln(k
31)+k
13ln(k
32)
(2f) inverse of entropy in the calculation procedure (2e):
(2g) calculate the weight of each evidence:
(2h) use above-mentioned weight that each evidence is carried out the discount operation:
(2i) evidence that finishes the discount operation is made up
Wherein A, B, C are the subclass of θ; M is the result after the combination.
3. synthetic-aperture radar automatic target recognition system comprises:
Pretreatment unit: be used for the diameter radar image of synthetic aperture radar training image and target to be identified is carried out the log conversion pre-service that normalization and objective contour extract;
Feature deriving means: be used for pretreated synthetic-aperture radar training image is carried out principal component analysis PCA respectively these three kinds of Feature Extraction of oval Fourier descriptor characteristic and two-dimensional Fourier transform;
Sorter trainer: be used to train three sorters, promptly adopt the K-nearest neighbor algorithm to use a K-nearest neighbour classification of PCA features training device; Adopt the supporting vector machine algorithm to use supporting vector machine sorter of oval Fourier descriptor features training, adopt the MINIACE algorithm filter to use two-dimensional Fourier transform training MINIACE wave filter;
Target Identification Unit: be used for extracting respectively the PCA characteristic of diameter radar image to be identified, oval Fourier descriptor characteristic and two-dimensional Fourier transform; And output in corresponding K-nearest neighbour classification device, supporting vector machine sorter and the MINIACE wave filter, each sorter provides a recognition result; After the discount operation and of the recognition result fusion of Dempster rule of combination through the Dempster-Shafer evidence theory, obtain final recognition result with described three sorters;
The PCA characteristic of described extraction diameter radar image to be identified only needs image to be identified is pulled into a column vector, then this column vector is got final product to the projection matrix V projection that pretreated synthetic-aperture radar training image carries out in the PCA leaching process.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010102309488A CN101894269B (en) | 2010-07-16 | 2010-07-16 | Multi-classifier system-based synthetic aperture radar automatic target recognition method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010102309488A CN101894269B (en) | 2010-07-16 | 2010-07-16 | Multi-classifier system-based synthetic aperture radar automatic target recognition method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101894269A CN101894269A (en) | 2010-11-24 |
CN101894269B true CN101894269B (en) | 2012-07-04 |
Family
ID=43103457
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2010102309488A Expired - Fee Related CN101894269B (en) | 2010-07-16 | 2010-07-16 | Multi-classifier system-based synthetic aperture radar automatic target recognition method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101894269B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103870836A (en) * | 2014-01-10 | 2014-06-18 | 浙江工业大学 | POCS (Projections Onto Convex Sets) super-resolution reconstruction-based SAR (Synthetic Aperture Radar) image target recognition method |
Families Citing this family (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102798867B (en) * | 2012-09-10 | 2015-01-07 | 电子科技大学 | Correlation method for flight tracks of airborne radar and infrared sensor |
CN103258213B (en) * | 2013-04-22 | 2016-04-27 | 中国石油大学(华东) | A kind of for the dynamic vehicle model recognizing method in intelligent transportation system |
CN103679212A (en) * | 2013-12-06 | 2014-03-26 | 无锡清华信息科学与技术国家实验室物联网技术中心 | Method for detecting and counting personnel based on video image |
CN104239900B (en) * | 2014-09-11 | 2017-03-29 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on K averages and depth S VM |
CN104392465A (en) * | 2014-11-13 | 2015-03-04 | 南京航空航天大学 | Multi-core target tracking method based on D-S evidence theory information integration |
CN104809230A (en) * | 2015-05-08 | 2015-07-29 | 东北大学 | Cigarette sensory quality evaluation method based on multi-classifier integration |
CN104992180B (en) * | 2015-06-26 | 2019-01-29 | 武汉大学 | A kind of multiple features fusion automobile logo identification method and system towards traffic block port |
CN104965199B (en) * | 2015-07-28 | 2017-11-17 | 中国人民解放军海军航空工程学院 | Radar video moving target Fusion Features decision method |
CN105913081B (en) * | 2016-04-08 | 2019-01-15 | 西安电子科技大学 | SAR image classification method based on improved PCAnet |
CN106022366B (en) * | 2016-07-04 | 2019-03-08 | 杭州电子科技大学 | A kind of rotating machinery method for diagnosing faults based on neighbour's evidence fusion |
CN106324577B (en) * | 2016-08-18 | 2019-03-12 | 北京理工大学 | A kind of high resolution radar test point method for congregating based on standard deviation ellipse |
DE102018222800A1 (en) * | 2018-12-21 | 2020-06-25 | Robert Bosch Gmbh | Locating and / or classifying objects from radar data with improved reliability at different distances |
CN110458064B (en) * | 2019-07-31 | 2022-09-27 | 南京理工大学 | Low-altitude target detection and identification method combining data driving type and knowledge driving type |
CN110940959B (en) * | 2019-12-13 | 2022-05-24 | 中国电子科技集团公司第五十四研究所 | Man-vehicle classification and identification method for low-resolution radar ground target |
CN110992357B (en) * | 2019-12-17 | 2023-04-14 | 云南中烟工业有限责任公司 | Radar chart analysis method, medium and single chip microcomputer for storing radar chart analysis method and respective application |
CN113608172B (en) * | 2021-06-23 | 2023-12-29 | 西安电子科技大学 | Airborne multifunctional radar working mode identification method based on improved K nearest neighbor |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100426314C (en) * | 2005-08-02 | 2008-10-15 | 中国科学院计算技术研究所 | Feature classification based multiple classifiers combined people face recognition method |
FR2898709B1 (en) * | 2006-03-17 | 2008-09-26 | Solystic Sas | METHOD FOR MERGING POSTAL OCR OPERATING FUNCTIONS OF BELIEF |
CN101551864B (en) * | 2009-05-13 | 2012-01-04 | 西安电子科技大学 | Image classification method based on feature correlation of frequency domain direction |
CN101561865B (en) * | 2009-05-22 | 2012-03-28 | 西安电子科技大学 | Synthetic aperture radar image target identification method based on multi-parameter spectrum feature |
-
2010
- 2010-07-16 CN CN2010102309488A patent/CN101894269B/en not_active Expired - Fee Related
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103870836A (en) * | 2014-01-10 | 2014-06-18 | 浙江工业大学 | POCS (Projections Onto Convex Sets) super-resolution reconstruction-based SAR (Synthetic Aperture Radar) image target recognition method |
Also Published As
Publication number | Publication date |
---|---|
CN101894269A (en) | 2010-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101894269B (en) | Multi-classifier system-based synthetic aperture radar automatic target recognition method | |
Zhang et al. | Airport detection and aircraft recognition based on two-layer saliency model in high spatial resolution remote-sensing images | |
Tuzel et al. | Human detection via classification on riemannian manifolds | |
Wen et al. | Efficient feature selection and classification for vehicle detection | |
Sivaraman et al. | Active learning for on-road vehicle detection: A comparative study | |
Lee et al. | Adaboost for text detection in natural scene | |
Razavi et al. | Latent hough transform for object detection | |
EP2003596A2 (en) | Computer implemented method for constructing classifier from training data detecting moving objects in test data using classifier | |
CN104680127A (en) | Gesture identification method and gesture identification system | |
Ullah et al. | Pedestrian detection in infrared images using fast RCNN | |
CN102651073A (en) | Sparse dynamic ensemble selection-based SAR (synthetic aperture radar) image terrain classification method | |
Kadhm et al. | Handwriting word recognition based on SVM classifier | |
CN101551809A (en) | Search method of SAR images classified based on Gauss hybrid model | |
Haloi | A novel plsa based traffic signs classification system | |
CN103049767B (en) | Aurora image classification method based on biological stimulation characteristic and manifold learning | |
CN109344880B (en) | SAR image classification method based on multiple features and composite kernels | |
CN103295032B (en) | Based on the image classification method of spatial Fisher vector | |
Nguyen et al. | Satellite image classification using convolutional learning | |
Masita et al. | Pedestrian detection using R-CNN object detector | |
Thubsaeng et al. | Vehicle logo detection using convolutional neural network and pyramid of histogram of oriented gradients | |
Elmikaty et al. | Car detection in aerial images of dense urban areas | |
Lin et al. | Using CNN to classify hyperspectral data based on spatial-spectral information | |
Moreno et al. | A comparative study of local descriptors for object category recognition: SIFT vs HMAX | |
Ghahnavieh et al. | Enhancing the license plates character recognition methods by means of SVM | |
Rodríguez-Serrano et al. | Data-driven vehicle identification by image matching |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120704 Termination date: 20180716 |
|
CF01 | Termination of patent right due to non-payment of annual fee |