CN106934419A - Classification of Polarimetric SAR Image method based on plural profile ripple convolutional neural networks - Google Patents
Classification of Polarimetric SAR Image method based on plural profile ripple convolutional neural networks Download PDFInfo
- Publication number
- CN106934419A CN106934419A CN201710137886.8A CN201710137886A CN106934419A CN 106934419 A CN106934419 A CN 106934419A CN 201710137886 A CN201710137886 A CN 201710137886A CN 106934419 A CN106934419 A CN 106934419A
- Authority
- CN
- China
- Prior art keywords
- layer
- eigenmatrix
- plural
- pond
- convolutional neural
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of Classification of Polarimetric SAR Image method based on plural profile ripple convolutional neural networks, mainly solve the problems, such as that prior art nicety of grading is relatively low.Implementation step is：1. it is input into the polarization coherence matrix T of Polarimetric SAR Image to be sorted and normalizes；2., according to the matrix after normalization, the eigenmatrix of training dataset and test data set is constructed respectively；3. complex convolution neutral net is constructed, and then obtains plural profile ripple convolutional neural networks；4. plural profile ripple convolutional neural networks, the model for being trained are trained with training dataset；5. the eigenmatrix of test data set is input in the model for training and is classified, obtain classification results.Convolutional neural networks continuation to complex field is carried out computing and extracts the characteristics of image of multiple dimensioned, multi-direction, many resolution characteristics by the present invention, effectively increases the nicety of grading of Polarimetric SAR Image, can be used for object detection and recognition.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to a kind of Classification of Polarimetric SAR Image method, can be used for target
Detection and identification.
Background technology
Polarization SAR is a kind of relevant microwave imaging system of multichannel, is the extension system of single polarization SAR, is surveyed by vector
Amount method obtains ground object target information, comprising amplitude, phase component.The premise that Polarimetric SAR Image is correctly classified is to polarization
SAR image carries out sufficient feature extraction, with the attribute of atural object in this phenogram picture.
The existing target's feature-extraction method based on scattering properties, mainly includes that Cloude is decomposed, Freeman is decomposed
Deng.
1997, Cloude et al. proposed H/ alpha taxonomy methods, and polarization is characterized with Terrain Scattering entropy H and angle of scattering α two
The characteristic value of data is that each pixel is divided into respective classes by foundation.
2004, Lee et al. proposed a kind of feature extracting method decomposed based on Freeman, and the method is utilized
The power that Freeman decomposes the three kinds of scattering mechanisms for obtaining is classified, and the Polarization scattering of Polarimetric SAR Image can be kept special
Property.
But these feature extracting methods due to do not account for Polarimetric SAR Image phase information and it is multiple dimensioned,
Multi-direction, many resolution characteristics, thus the complicated Polarimetric SAR Image of background is difficult to obtain nicety of grading higher.
The content of the invention
It is an object of the invention to regarding to the issue above, propose a kind of polarization based on plural profile ripple convolutional neural networks
SAR image sorting technique, to improve the nicety of grading of Polarimetric SAR Image.
Technological core of the invention is that convolutional neural networks continuation to complex field is carried out into computing, and in answering that construction is obtained
Multiple dimensioned depth filter is introduced in number convolutional neural networks to extract the image with multiple dimensioned, multi-direction, many resolution characteristics
Feature, its technical scheme includes as follows：
(1) it is input into the polarization coherence matrix T of Polarimetric SAR Image to be sorted；
(2) it is real part eigenmatrix T1 and imaginary part eigenmatrix T2 by T points of polarization coherence matrix, respectively by real part feature square
Element value in battle array T1 and imaginary part eigenmatrix T2 is normalized between [0,1], the real part eigenmatrix F1 after being normalized
With the imaginary part eigenmatrix F2 after normalization；
(3) real part eigenmatrix F1 after normalization and normalization after imaginary part eigenmatrix F2 in each element week
Enclose and take 37 × 37 block and represent central element, constitute the real part eigenmatrix F3 based on image block and the imaginary part based on image block is special
Levy matrix F 4；
(4) Polarimetric SAR Image atural object to be sorted is divided into 15 classes, in the real part eigenmatrix F3 based on image block
200 markd elements are randomly selected as training sample in per class, and remaining obtains training dataset as test sample
The real part eigenmatrix W2 of real part eigenmatrix W1 and test data set；It is every in the imaginary part eigenmatrix F4 based on image block
200 markd elements are randomly selected in class as training sample, remaining obtains the void of training dataset as test sample
The imaginary part eigenmatrix W4 of portion eigenmatrix W3 and test data set；
(5) complex convolution neutral net is constructed：
Construction one is by input layer → complex convolution layer → plural pond layer → complex convolution layer → plural pond layer → again
10 layers of complex convolution nerve of number convolutional layer → plural number pond layer → full articulamentum → full articulamentum → softmax graders composition
Network, gives the Feature Mapping figure of each layer, determines the filter size and random initializtion wave filter of complex convolution layer；
(6) plural profile ripple convolutional neural networks are constructed：
Multiple dimensioned depth filter is constructed with the scaling filter in non-down sampling contourlet transform and anisotropic filter, and
The wave filter of random initializtion in first complex convolution layer of complex convolution neutral net is replaced, is obtained by input layer → multiple dimensioned
Depth filter layer → plural pond layer → complex convolution layer → plural pond layer → complex convolution layer → plural pond layer → complete
The plural profile ripple convolutional neural networks of articulamentum → full articulamentum → softmax graders this 10 Rotating fields composition；
(7) it is trained with training data set pair complex wheel exterior feature ripple convolutional neural networks, the model for being trained；
(8) the real part eigenmatrix W2 and imaginary part eigenmatrix W4 of test data set are trained into model as step (7)
Input, obtain the classification that test data concentrates each element.
The present invention has advantages below compared with prior art：
1. the present invention by convolutional neural networks continuation to complex field due to carrying out computing, thus the complex convolution that construction is obtained
Neutral net can make full use of the phase information of Polarimetric SAR Image；
2. the present invention in complex convolution neutral net due to introducing multiple dimensioned depth filter, therefore can obtain with many
The characteristics of image of yardstick, multi-direction, many resolution characteristics, enhances the generalization ability of plural profile ripple convolutional neural networks.
3. the present invention is due to taking full advantage of the phase information of Polarimetric SAR Image in plural profile ripple convolutional neural networks,
And it is extracted the characteristics of image of multiple dimensioned, multi-direction, many resolution characteristics so that the terrestrial object information that the network extraction is arrived is more rich
Richness, effectively increases the nicety of grading of Polarimetric SAR Image.
Brief description of the drawings
Fig. 1 is of the invention to realize flow chart；
Fig. 2 is handmarking figure of the present invention to image to be classified；
Fig. 3 is the classification results figure to image to be classified with the present invention.
Specific embodiment
Realize that step and experiment effect are described in further detail to of the invention below in conjunction with accompanying drawing：
Reference picture 1, of the invention to implement step as follows：
Step 1, is input into the polarization coherence matrix T of Polarimetric SAR Image to be sorted.
The complete polarization in the Dutch Flevoland areas that Polarimetric SAR Image to be sorted is shot from NASA/JPL laboratories
Image, image size is 750 × 1024.
Step 2, by polarization coherence matrix, T points is real part eigenmatrix T1 and imaginary part eigenmatrix T2, special to real part respectively
Matrix T1 and imaginary part eigenmatrix T2 normalization is levied, the imaginary part after real part eigenmatrix F1 and normalization after being normalized is special
Levy matrix F 2.
Conventional method for normalizing has：Characteristic line pantography, feature normalization and feature albefaction.
This example use characteristic line pantography, i.e., first obtain respectively real part eigenmatrix T1 maximum max (T1) and
The maximum max (T2) of imaginary part eigenmatrix T2；Again by each element in real part eigenmatrix T1 and imaginary part eigenmatrix T2
Corresponding maximum max (T1) and max (T2) are respectively divided by, after the real part eigenmatrix F1 after being normalized and normalization
Imaginary part eigenmatrix F2.
Step 3, constitutes real part eigenmatrix F3 and the imaginary part eigenmatrix F4 based on image block based on image block.
The block that 37 × 37 are taken around each element of real part eigenmatrix F1 after normalization represents central element, constitutes
Real part eigenmatrix F3 based on image block；
The block that 37 × 37 are taken around each element of imaginary part eigenmatrix F2 after normalization represents central element, constitutes
Imaginary part eigenmatrix F4 based on image block.
Step 4, obtains the eigenmatrix of training dataset and test data set.
Polarimetric SAR Image atural object to be sorted is divided into 15 classes；
200 markd elements are randomly selected in every class in the real part eigenmatrix F3 based on image block as instruction
Practice sample, remaining obtains the real part eigenmatrix W1 of training dataset and the real part feature of test data set as test sample
Matrix W 2.
200 markd elements are randomly selected in every class in the imaginary part eigenmatrix F4 based on image block as instruction
Practice sample, remaining obtains the imaginary part eigenmatrix W3 of training dataset and the imaginary part feature of test data set as test sample
Matrix W 4.
Step 5, constructs complex convolution neutral net.
5a) press data transfer direction and set one by input layer → complex convolution layer → plural pond layer → complex convolution layer
10 layers of → plural pond layer → complex convolution layer → plural pond layer → full articulamentum → full articulamentum → softmax graders
Complex convolution neutral net, wherein, the transmission direction of arrow → refer to input data.
The Feature Mapping figure of each layer 5b) is given, and determines the filter size of complex convolution layer：
For the 1st layer of input layer, it is 18 to set Feature Mapping map number；
For the 2nd layer of complex convolution layer, it is 72 to set Feature Mapping map number；
For the 3rd layer of plural pond layer, it is 2 to set pond radius；
For the 4th layer of complex convolution layer, it is 48 to set Feature Mapping map number, and it is 4 to set filter size；
For the 5th layer of plural pond layer, it is 2 to set pond radius；
For the 6th layer of complex convolution layer, it is 16 to set Feature Mapping map number, and it is 4 to set filter size；
For the 7th layer of plural pond layer, it is 2 to set pond radius；
For the 8th layer of full articulamentum, it is 128 to set Feature Mapping map number；
For the 9th layer of full articulamentum, it is 50 to set Feature Mapping map number；
For the 10th layer of softmax grader, it is 15 to set Feature Mapping map number.
Step 6, the plural profile ripple convolutional neural networks of construction.
6a) multiple dimensioned depth filtering is constructed with the scaling filter and anisotropic filter in non-down sampling contourlet transform
Device；
6b) with random initializtion in first complex convolution layer of multiple dimensioned depth filter replacement complex convolution neutral net
Wave filter, obtain by input layer → multiple dimensioned depth filter layer → plural pond layer → complex convolution layer → plural number pond layer
What → complex convolution layer → plural pond layer → full articulamentum → full articulamentum → softmax graders this 10 Rotating fields were constituted answers
Number profile ripple convolutional neural networks；
The parameter for 6c) setting plural profile each layer of ripple convolutional neural networks is as follows：
It is 18 to set the 1st layer of Feature Mapping map number of input layer；
The Feature Mapping map number for setting the 2nd layer of multiple dimensioned depth filter layer is 72；
The pond radius for setting the 3rd layer of plural pond layer is 2；
The Feature Mapping map number for setting the 4th layer of complex convolution layer is 48, and filter size is 4；
The pond radius for setting the 5th layer of plural pond layer is 2；
The Feature Mapping map number for setting the 6th layer of complex convolution layer is 16, and filter size is 4；
The pond radius for setting the 7th layer of plural pond layer is 2；
It is 128 to set the 8th layer of Feature Mapping map number of full articulamentum；
It is 50 to set the 9th layer of Feature Mapping map number of full articulamentum；
It is 15 to set the 10th layer of Feature Mapping map number of softmax graders.
Step 7, is trained, the model for being trained with training data set pair complex wheel exterior feature ripple convolutional neural networks.
7a) using the real part eigenmatrix W1 and imaginary part eigenmatrix W3 of training dataset as plural profile ripple convolutional Neural
The input of network, is expressed as x=W1+jW3；
7b) x is sequentially input to each layer of plural profile ripple convolutional neural networks, the output of network is obtained, the network is defeated
Go out the training classification of correspondence x；
The error between training classification and the correct classification of handmarking 7c) is solved, and is carried out reversely by the error
Propagate to optimize network parameter, the model for being trained.
The correct category of the handmarking, as shown in Figure 2.
Step 8, is classified using the model for training to test data set.
The real part eigenmatrix W2 and imaginary part eigenmatrix W4 of test data set are trained into the defeated of model as step (7)
Enter, the output of the model is the classification results that test data concentrates each element classification to obtain.
Effect of the invention can be further illustrated by following emulation experiment：
1. simulated conditions：
Hardware platform is：Intel (R) Xeon (R) CPU E5-2630,2.40GHz*16, inside saves as 64G.
Software platform is：MXNet.
2. emulation content and result：
Emulation 1, tested under above-mentioned simulated conditions with the inventive method, i.e., respectively from polarization SAR data each
200 markd elements are randomly selected in classification as training sample, remaining markd element is carried out as test sample
Classification, as a result as shown in Figure 3.
As can be seen from Figure 3：The edge of classification results very clear and region consistency preferably, shows proposed by the invention
Plural profile ripple convolutional neural networks be applied to treatment Classification of Polarimetric SAR Image problem.
Emulation 2, reduces training sample successively, 100 is randomly selected from every class, 50 markd elements are used as training
Sample, nicety of grading of the present invention with convolutional neural networks in test data set is contrasted, as a result as shown in table 1：
Table 1
| Per class training sample number | Training sample proportion | Convolutional neural networks | The present invention |
| 200 | 1.8% | 99.00% | 99.41% |
| 100 | 0.9% | 97.60% | 97.94% |
| 50 | 0.5% | 95.33% | 96.37% |
As seen from Table 1, when training sample accounts for 1.8%, 0.9%, the 0.5% of total sample number, the present invention is in test data set
On nicety of grading obviously higher than convolutional neural networks.It is with the obvious advantage in the case where training sample number is less.
To sum up, the present invention carries out computing by by convolutional neural networks continuation to complex field, and introduces multiple dimensioned depth filter
Ripple device, can effectively improve the ability to express of characteristics of image, strengthen the generalization ability of model, make the nicety of grading of Polarimetric SAR Image
It is obviously improved.
Claims (5)
1. a kind of Classification of Polarimetric SAR Image method based on plural profile ripple convolutional neural networks, including：
(1) it is input into the polarization coherence matrix T of Polarimetric SAR Image to be sorted；
(2) it is real part eigenmatrix T1 and imaginary part eigenmatrix T2 by T points of polarization coherence matrix, respectively by real part eigenmatrix T1
Normalized between [0,1] with the element value in imaginary part eigenmatrix T2, real part eigenmatrix F1 after being normalized and returned
Imaginary part eigenmatrix F2 after one change；
(3) taken around each element in imaginary part eigenmatrix F2 after real part eigenmatrix F1 after normalization and normalization
37 × 37 block represents central element, constitutes the real part eigenmatrix F3 based on image block and the imaginary part feature square based on image block
Battle array F4；
(4) Polarimetric SAR Image atural object to be sorted is divided into 15 classes, the every class in the real part eigenmatrix F3 based on image block
In randomly select 200 markd elements as training sample, remaining obtains the real part of training dataset as test sample
The real part eigenmatrix W2 of eigenmatrix W1 and test data set；In every class in the imaginary part eigenmatrix F4 based on image block
200 markd elements are randomly selected as training sample, remaining is used as test sample, and the imaginary part for obtaining training dataset is special
Levy the imaginary part eigenmatrix W4 of matrix W 3 and test data set；
(5) complex convolution neutral net is constructed：
Construction one is rolled up by input layer → complex convolution layer → plural pond layer → complex convolution layer → plural pond layer → plural number
10 layers of complex convolution nerve net of lamination → plural number pond layer → full articulamentum → full articulamentum → softmax graders composition
Network, gives the Feature Mapping figure of each layer, determines the filter size and random initializtion wave filter of complex convolution layer；
(6) plural profile ripple convolutional neural networks are constructed：
Multiple dimensioned depth filter is constructed with the scaling filter in non-down sampling contourlet transform and anisotropic filter, and is replaced
The wave filter of random initializtion, obtains by input layer → multiple dimensioned depth in first complex convolution layer of complex convolution neutral net
Filter layer → plural number pond layer → complex convolution layer → plural pond layer → complex convolution layer → plural pond layer → full connection
The plural profile ripple convolutional neural networks of layer → full articulamentum → softmax graders this 10 Rotating fields composition；
(7) it is trained with training data set pair complex wheel exterior feature ripple convolutional neural networks, the model for being trained；
(8) the real part eigenmatrix W2 and imaginary part eigenmatrix W4 of test data set are trained into the defeated of model as step (7)
Enter, obtain the classification that test data concentrates each element.
2. method according to claim 1, wherein returns in step (2) to real part eigenmatrix T1 and imaginary part eigenmatrix T2
One changes, and using characteristic line pantography, i.e., first obtains the maximum max (T1) and imaginary part feature square of real part eigenmatrix T1 respectively
The maximum max (T2) of battle array T2；Each element in real part eigenmatrix T1 and imaginary part eigenmatrix T2 is respectively divided by again right
The maximum max (T1) and max (T2) for answering, the imaginary part feature square after real part eigenmatrix F1 and normalization after being normalized
Battle array F2.
3. method according to claim 1, wherein constructs the 10 layers of complex convolution neutral net for obtaining in step (5), its
Parameter setting is as follows：
For the 1st layer of input layer, it is 18 to set Feature Mapping map number；
For the 2nd layer of complex convolution layer, it is 72 to set Feature Mapping map number；
For the 3rd layer of plural pond layer, it is 2 to set pond radius；
For the 4th layer of complex convolution layer, it is 48 to set Feature Mapping map number, and it is 4 to set filter size；
For the 5th layer of plural pond layer, it is 2 to set pond radius；
For the 6th layer of complex convolution layer, it is 16 to set Feature Mapping map number, and it is 4 to set filter size；
For the 7th layer of plural pond layer, it is 2 to set pond radius；
For the 8th layer of full articulamentum, it is 128 to set Feature Mapping map number；
For the 9th layer of full articulamentum, it is 50 to set Feature Mapping map number；
For the 10th layer of softmax grader, it is 15 to set Feature Mapping map number.
4. method according to claim 1, wherein constructs the plural profile ripple convolutional Neural net for obtaining 10 layers in step (6)
Network, the parameter setting of its each layer is as follows：
It is 18 to set the 1st layer of Feature Mapping map number of input layer；
The Feature Mapping map number for setting the 2nd layer of multiple dimensioned depth filter layer is 72；
The pond radius for setting the 3rd layer of plural pond layer is 2；
The Feature Mapping map number for setting the 4th layer of complex convolution layer is 48, and filter size is 4；
The pond radius for setting the 5th layer of plural pond layer is 2；
The Feature Mapping map number for setting the 6th layer of complex convolution layer is 16, and filter size is 4；
The pond radius for setting the 7th layer of plural pond layer is 2；
It is 128 to set the 8th layer of Feature Mapping map number of full articulamentum；
It is 50 to set the 9th layer of Feature Mapping map number of full articulamentum；
It is 15 to set the 10th layer of Feature Mapping map number of softmax graders.
5. method according to claim 1, wherein uses training data set pair complex wheel exterior feature ripple convolutional Neural net in step (7)
Network is trained, and carries out as follows：
(1) using the real part eigenmatrix W1 and imaginary part eigenmatrix W3 of training dataset as plural profile ripple convolutional neural networks
Input, be expressed as x=W1+jW3；
(2) x is sequentially input to each layer of plural profile ripple convolutional neural networks, obtains the output of network, network output is right
Answer the training classification of x；
(3) error between training classification and the correct classification of handmarking is solved, and backpropagation is carried out by the error
To optimize network parameter, the model for being trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710137886.8A CN106934419A (en) | 2017-03-09 | 2017-03-09 | Classification of Polarimetric SAR Image method based on plural profile ripple convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710137886.8A CN106934419A (en) | 2017-03-09 | 2017-03-09 | Classification of Polarimetric SAR Image method based on plural profile ripple convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106934419A true CN106934419A (en) | 2017-07-07 |
Family
ID=59433074
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710137886.8A Pending CN106934419A (en) | 2017-03-09 | 2017-03-09 | Classification of Polarimetric SAR Image method based on plural profile ripple convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106934419A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273938A (en) * | 2017-07-13 | 2017-10-20 | 西安电子科技大学 | Multi-source Remote Sensing Images terrain classification method based on binary channels convolution ladder net |
| CN107368852A (en) * | 2017-07-13 | 2017-11-21 | 西安电子科技大学 | A kind of Classification of Polarimetric SAR Image method based on non-down sampling contourlet DCGAN |
| CN107491793A (en) * | 2017-09-04 | 2017-12-19 | 西安电子科技大学 | A kind of Classification of Polarimetric SAR Image method based on the sparse full convolution of scattering |
| CN107833180A (en) * | 2017-10-27 | 2018-03-23 | 北京大学 | A kind of method using complex field neutral net rapid solving nonlinear electromagnetic inverse Problem |
| CN107944353A (en) * | 2017-11-10 | 2018-04-20 | 西安电子科技大学 | SAR image change detection based on profile ripple BSPP networks |
| CN109102015A (en) * | 2018-08-06 | 2018-12-28 | 西安电子科技大学 | A kind of SAR image change detection based on complex-valued neural networks |
| CN109658469A (en) * | 2018-12-13 | 2019-04-19 | 深圳先进技术研究院 | A kind of neck joint imaging method and device based on the study of depth priori |
| CN110210558A (en) * | 2019-05-31 | 2019-09-06 | 北京市商汤科技开发有限公司 | Assess the method and device of neural network performance |
| CN110728324A (en) * | 2019-10-12 | 2020-01-24 | 西安电子科技大学 | Depth complex value full convolution neural network-based polarimetric SAR image classification method |
| CN111123183A (en) * | 2019-12-27 | 2020-05-08 | 杭州电子科技大学 | Rapid magnetic resonance imaging method based on complex R2U _ Net network |
| CN112329538A (en) * | 2020-10-10 | 2021-02-05 | 杭州电子科技大学 | Target classification method based on microwave vision |
| CN113030902A (en) * | 2021-05-08 | 2021-06-25 | 电子科技大学 | Twin complex network-based few-sample radar vehicle target identification method |
| CN113109780A (en) * | 2021-03-02 | 2021-07-13 | 西安电子科技大学 | High-resolution range profile target identification method based on complex number dense connection neural network |
| CN113191361A (en) * | 2021-04-19 | 2021-07-30 | 苏州大学 | Shape recognition method |
| 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 |
| CN113408628A (en) * | 2021-06-22 | 2021-09-17 | 西安邮电大学 | PolSAR image classification method based on multi-model joint learning network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102708364A (en) * | 2012-05-31 | 2012-10-03 | 西安电子科技大学 | Cascade-classifier-based fingerprint image classification method |
| WO2015054666A1 (en) * | 2013-10-10 | 2015-04-16 | Board Of Regents, The University Of Texas System | Systems and methods for quantitative analysis of histopathology images using multi-classifier ensemble schemes |
| CN105142177A (en) * | 2015-08-05 | 2015-12-09 | 西安电子科技大学 | Complex neural network channel prediction method |
| US20160132768A1 (en) * | 2014-11-10 | 2016-05-12 | The Boeing Company | Systems and methods for training multipath filtering systems |
| CN105718957A (en) * | 2016-01-26 | 2016-06-29 | 西安电子科技大学 | Polarized SAR image classification method based on nonsubsampled contourlet convolutional neural network |
-
2017
- 2017-03-09 CN CN201710137886.8A patent/CN106934419A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102708364A (en) * | 2012-05-31 | 2012-10-03 | 西安电子科技大学 | Cascade-classifier-based fingerprint image classification method |
| WO2015054666A1 (en) * | 2013-10-10 | 2015-04-16 | Board Of Regents, The University Of Texas System | Systems and methods for quantitative analysis of histopathology images using multi-classifier ensemble schemes |
| US20160132768A1 (en) * | 2014-11-10 | 2016-05-12 | The Boeing Company | Systems and methods for training multipath filtering systems |
| CN105142177A (en) * | 2015-08-05 | 2015-12-09 | 西安电子科技大学 | Complex neural network channel prediction method |
| CN105718957A (en) * | 2016-01-26 | 2016-06-29 | 西安电子科技大学 | Polarized SAR image classification method based on nonsubsampled contourlet convolutional neural network |
Non-Patent Citations (1)
| Title |
|---|
| GUBERMAN,NITZAN: "On Complex Valued Convolutional Neural Networks", 《EPRINT ARXIV:1602.09046》 * |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107368852A (en) * | 2017-07-13 | 2017-11-21 | 西安电子科技大学 | A kind of Classification of Polarimetric SAR Image method based on non-down sampling contourlet DCGAN |
| CN107273938A (en) * | 2017-07-13 | 2017-10-20 | 西安电子科技大学 | Multi-source Remote Sensing Images terrain classification method based on binary channels convolution ladder net |
| CN107273938B (en) * | 2017-07-13 | 2020-05-29 | 西安电子科技大学 | Multi-source remote sensing image ground object classification method based on two-channel convolution ladder network |
| CN107491793B (en) * | 2017-09-04 | 2020-05-01 | 西安电子科技大学 | Polarized SAR image classification method based on sparse scattering complete convolution |
| CN107491793A (en) * | 2017-09-04 | 2017-12-19 | 西安电子科技大学 | A kind of Classification of Polarimetric SAR Image method based on the sparse full convolution of scattering |
| CN107833180A (en) * | 2017-10-27 | 2018-03-23 | 北京大学 | A kind of method using complex field neutral net rapid solving nonlinear electromagnetic inverse Problem |
| CN107944353A (en) * | 2017-11-10 | 2018-04-20 | 西安电子科技大学 | SAR image change detection based on profile ripple BSPP networks |
| CN107944353B (en) * | 2017-11-10 | 2019-12-24 | 西安电子科技大学 | SAR image change detection method based on contour wave BSPP network |
| CN109102015A (en) * | 2018-08-06 | 2018-12-28 | 西安电子科技大学 | A kind of SAR image change detection based on complex-valued neural networks |
| CN109658469A (en) * | 2018-12-13 | 2019-04-19 | 深圳先进技术研究院 | A kind of neck joint imaging method and device based on the study of depth priori |
| CN110210558A (en) * | 2019-05-31 | 2019-09-06 | 北京市商汤科技开发有限公司 | Assess the method and device of neural network performance |
| CN110728324A (en) * | 2019-10-12 | 2020-01-24 | 西安电子科技大学 | Depth complex value full convolution neural network-based polarimetric SAR image classification method |
| CN110728324B (en) * | 2019-10-12 | 2022-03-04 | 西安电子科技大学 | Depth complex value full convolution neural network-based polarimetric SAR image classification method |
| CN111123183B (en) * | 2019-12-27 | 2022-04-15 | 杭州电子科技大学 | Rapid magnetic resonance imaging method based on complex R2U _ Net network |
| CN111123183A (en) * | 2019-12-27 | 2020-05-08 | 杭州电子科技大学 | Rapid magnetic resonance imaging method based on complex R2U _ Net network |
| CN112329538A (en) * | 2020-10-10 | 2021-02-05 | 杭州电子科技大学 | Target classification method based on microwave vision |
| CN113109780A (en) * | 2021-03-02 | 2021-07-13 | 西安电子科技大学 | High-resolution range profile target identification method based on complex number dense connection neural network |
| CN113109780B (en) * | 2021-03-02 | 2022-08-05 | 西安电子科技大学 | High-resolution range profile target identification method based on complex number dense connection neural network |
| CN113191361A (en) * | 2021-04-19 | 2021-07-30 | 苏州大学 | Shape recognition method |
| CN113191361B (en) * | 2021-04-19 | 2023-08-01 | 苏州大学 | Shape recognition method |
| CN113030902A (en) * | 2021-05-08 | 2021-06-25 | 电子科技大学 | Twin complex network-based few-sample radar vehicle target identification method |
| CN113030902B (en) * | 2021-05-08 | 2022-05-17 | 电子科技大学 | Twin complex network-based few-sample radar vehicle target identification method |
| 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 |
| 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 |
| CN113408628A (en) * | 2021-06-22 | 2021-09-17 | 西安邮电大学 | PolSAR image classification method based on multi-model joint learning network |
| CN113408628B (en) * | 2021-06-22 | 2023-01-31 | 西安邮电大学 | PolSAR image classification method based on multi-model joint learning network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106934419A (en) | Classification of Polarimetric SAR Image method based on plural profile ripple convolutional neural networks | |
| CN110660038B (en) | Multispectral image and full-color image fusion method based on generation countermeasure network | |
| CN105868793B (en) | Classification of Polarimetric SAR Image method based on multiple dimensioned depth filter | |
| CN108564006B (en) | Polarized SAR terrain classification method based on self-learning convolutional neural network | |
| CN107239751B (en) | High-resolution SAR image classification method based on non-subsampled contourlet full convolution network | |
| CN108830330B (en) | Multispectral image classification method based on self-adaptive feature fusion residual error network | |
| CN108460342A (en) | Hyperspectral image classification method based on convolution net and Recognition with Recurrent Neural Network | |
| CN107463948A (en) | Classification of Multispectral Images method based on binary channels multiple features fusion network | |
| CN109767412A (en) | A kind of remote sensing image fusing method and system based on depth residual error neural network | |
| CN106934404A (en) | A kind of image flame identifying system based on CNN convolutional neural networks | |
| CN110516728B (en) | Polarized SAR terrain classification method based on denoising convolutional neural network | |
| CN107292317A (en) | Polarization SAR sorting technique based on shallow-layer feature Yu T matrix deep learnings | |
| CN110020693B (en) | Polarimetric SAR image classification method based on feature attention and feature improvement network | |
| CN107292336A (en) | A kind of Classification of Polarimetric SAR Image method based on DCGAN | |
| CN104299232B (en) | SAR image segmentation method based on self-adaptive window directionlet domain and improved FCM | |
| CN107944370A (en) | Classification of Polarimetric SAR Image method based on DCCGAN models | |
| CN104751172B (en) | The sorting technique of Polarimetric SAR Image based on denoising autocoding | |
| CN104616280B (en) | Method for registering images based on maximum stable extremal region and phase equalization | |
| CN112949738B (en) | Multi-class unbalanced hyperspectral image classification method based on EECNN algorithm | |
| CN109784192A (en) | Hyperspectral Image Classification method based on super-pixel feature extraction neural network algorithm | |
| CN107169492A (en) | Polarization SAR object detection method based on FCN CRF master-slave networks | |
| CN114463637B (en) | Winter wheat remote sensing identification analysis method and system based on deep learning | |
| CN106127221A (en) | Classification of Polarimetric SAR Image method based on polarization textural characteristics with DPL | |
| CN105894013A (en) | Method for classifying polarized SAR image based on CNN and SMM | |
| CN109447111A (en) | A kind of remote sensing supervised classification method based on subclass training sample |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170707 |
|
| WD01 | Invention patent application deemed withdrawn after publication |