CN107292336A - A kind of Classification of Polarimetric SAR Image method based on DCGAN - Google Patents
A kind of Classification of Polarimetric SAR Image method based on DCGAN Download PDFInfo
- Publication number
- CN107292336A CN107292336A CN201710440090.XA CN201710440090A CN107292336A CN 107292336 A CN107292336 A CN 107292336A CN 201710440090 A CN201710440090 A CN 201710440090A CN 107292336 A CN107292336 A CN 107292336A
- Authority
- CN
- China
- Prior art keywords
- mrow
- mtd
- eigenmatrix
- msub
- mtr
- 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The invention discloses a kind of Classification of Polarimetric SAR Image method based on DCGAN, comprise the following steps：1) odd scattering coefficient, even scattering coefficient and volume scattering coefficient are obtained, then builds the eigenmatrix F based on pixel；2) each element value in the eigenmatrix F based on pixel is normalized in [0,1], and normalized result is denoted as eigenmatrix F1；3) each element in eigenmatrix F1 is replaced by around it 64 × 64 image block, obtains the eigenmatrix F2 based on image block；4) construct the eigenmatrix W1 without label training dataset D1 and have label training dataset D2 eigenmatrix W2；5) the eigenmatrix W3 of construction test data set T super-pixel cluster centre；6) the training network model DCGAN after being trained；7) identification and classification network model is built, then eigenmatrix W3 is classified by identification and classification network model, this method can realize the classification of Polarimetric SAR Image, and nicety of grading is higher.
Description
Technical field
The invention belongs to technical field of image processing, it is related to a kind of Classification of Polarimetric SAR Image method based on DCGAN.
Background technology
Polarization SAR is a kind of high-resolution active-mode active microwave remote sensing imaging radar, with round-the-clock, round-the-clock, is divided
Resolution is high, can side view the advantages of be imaged, the more rich information of target can be obtained.The purpose of Classification of Polarimetric SAR Image is to utilize machine
Carry or polarization measurement data that borne polarization SAR sensor is obtained determine classification belonging to each pixel, agricultural, forestry,
There is extensive research and application value in terms of military affairs, geology, hydrology and ocean.Classical Classification of Polarimetric SAR Image side
Method has：
1992, the research such as Lee was thought, the form of polarization covariance matrix can be expressed as depending on Polarimetric SAR Image more, and
And the approximate matrix obeys multiple Wishart distributions, on this basis, he proposes a kind of simple and effective Wishart sorting algorithms
And for classifying to types of ground objects such as forest, city, ocean, sea ice.
1998, the feature that Lee etc. is extracted with H/Alpha decomposition methods carried out initial clustering to image, obtained in 8 clusters
The heart；Then classified (abbreviation H/Alpha- with many Wishart Iterative classifications devices depending on covariance matrix of description to image
Wishart graders).
2000, Pottier etc. proposed H/Alpha/A-Wishart graders, on the basis of H/Alpha decomposition,
A features are added, image is polymerized to 16 classes, Wishart Iterative classifications then are carried out to image again.
Polarization SAR develops also immature, many core technologies, such as filtering technique, polarizing target at present due to starting late
Decomposition technique, sorting technique are in urgent need to be improved, and particularly Classification of Polarimetric SAR Image also lacks the algorithm of high efficient and reliable at present, some
Advanced machine Learning Theory and method are not yet applied in Classification of Polarimetric SAR Image.Classical Classification of Polarimetric SAR Image
Method, it is difficult to adapt to increasing polarization SAR data, so as to be difficult to the distribution spy that fully study uses polarization SAR data
Property, it is difficult to the feature extracted, do not reach very high nicety of grading.
The content of the invention
It is an object of the invention to the shortcoming for overcoming above-mentioned prior art, there is provided a kind of polarization SAR figure based on DCGAN
As sorting technique, this method can realize the classification of Polarimetric SAR Image, and nicety of grading is higher.
To reach above-mentioned purpose, the Classification of Polarimetric SAR Image method of the present invention based on DCGAN includes following step
Suddenly：
1) polarization scattering matrix S is obtained, Pauli decomposition is carried out to polarization scattering matrix S, odd scattering coefficient, even is obtained
Scattering coefficient and volume scattering coefficient, then it regard odd scattering coefficient, even scattering coefficient and volume scattering coefficient as polarization to be sorted
Eigenmatrix F of the 3-D view feature construction of SAR image based on pixel；
2) each element value in the eigenmatrix F based on pixel is normalized in [0,1], and by normalized result
It is denoted as eigenmatrix F1；
3) each element in eigenmatrix F1 is replaced by around it 64 × 64 image block, obtained based on image
The eigenmatrix F2 of block；
4) construct the eigenmatrix W1 without label training dataset D1 using the eigenmatrix F2 based on image block and have mark
Sign training dataset D2 eigenmatrix W2；
5) using the eigenmatrix F2 construction test data set T based on image block, then in the eigenmatrix F based on pixel
Middle utilization SLIC super-pixel algorithm partition super-pixel block, obtains the cluster centre of super-pixel block, then in the feature based on image block
The eigenmatrix W3 of test data set T super-pixel cluster centre is constructed in matrix F 2；
6) by being trained without label training dataset D1 to training network model DCGAN, the training after being trained
Network model DCGAN；
7) two graders in arbiter D in the training network model DCGAN after training are replaced by softmax classification
Device, then it regard the arbiter D after replacing as sorter network model；
8) the eigenmatrix W2 for having label training dataset D2 is input in sorter network model, and updates softmax
The parameter of grader, then by there is label training dataset D2 eigenmatrix W2 to update the ginseng of whole sorter network model
Number, is then classified by identification and classification network model to the eigenmatrix W3 of test data set T super-pixel cluster centre,
Labeled test data set T category, realizes the Classification of Polarimetric SAR Image based on DCGAN again.
Step 1) in polarization scattering matrix S carry out Pauli decomposition, obtain odd scattering coefficient, even scattering coefficient and
The operation of volume scattering coefficient is：
Pauli bases { S 1a) is set1,S2,S3, wherein,
Wherein, S1Scattered for odd, S2Scattered for even, S3For volume scattering；
1b) decomposed and defined by Pauli：
Wherein, a is odd scattering coefficient, and b is even scattering coefficient, and c is volume scattering coefficient；
Formula (2) 1c) is solved, odd scattering coefficient a, even scattering coefficient b and volume scattering coefficient c is obtained, wherein,
Step 1) in build the eigenmatrix F based on pixel concrete operations be：
If size is the eigenmatrix of M1 × M2 × 3, then by odd scattering coefficient a, even scattering coefficient b and volume scattering system
Number c is assigned to the eigenmatrix that size is M1 × M2 × 3, obtains the eigenmatrix F based on pixel, wherein, M1 is polarization to be sorted
The length of SAR image, M2 is the width of Polarimetric SAR Image to be sorted.
Step 2) concrete operations be：The maximum max (F) of the eigenmatrix F based on pixel is solved, then picture will be based on
Each element in the eigenmatrix F of vegetarian refreshments is equal divided by the maximum max (F), obtains eigenmatrix F1.
Step 6) in pass through the concrete operations that are trained without label training dataset D1 to training network model DCGAN
For：
6a) setting the maker G in training network model DCGAN includes being sequentially connected the input layer a connect, the first deconvolution
Layer, the second warp lamination, the 3rd warp lamination, the 4th warp lamination and output layer, wherein, input layer a input is made an uproar for 100 dimensions
Sound vector；The Feature Mapping map number of first warp lamination is 512, and the filter size of the first warp lamination is 5；Second warp
The Feature Mapping map number of lamination is 256, and the filter size of the second warp lamination is 5；The Feature Mapping of 3rd warp lamination
Map number is 128, and the filter size of the 3rd warp lamination is 5；The Feature Mapping map number of 4th warp lamination is the 64, the 4th
The filter size of warp lamination is 5；Output layer exports the pcolor of 64 × 64 × 3 sizes；
6b) set arbiter D in training network model DCGAN include being sequentially connected the input layer b connect, the first convolutional layer a,
Second convolutional layer a, the 3rd convolutional layer a, Volume Four lamination a and two graders, wherein, input layer b Feature Mapping map number is
3；First layer convolutional layer a Feature Mapping map number is 64, and first layer convolutional layer a filter size is 5；Second layer convolutional layer
A Feature Mapping map number is 128, and second layer convolutional layer a filter size is 5；Third layer convolutional layer a Feature Mapping figure
Number is 256, and third layer convolutional layer a filter size is 5；4th layer of convolutional layer a Feature Mapping map number is 512, the
Four layers of convolutional layer a filter size is 5；Two graders export a scalar；
6c) the Uniform noise that input 100 is tieed up into training network model DCGAN maker G, will be without label training data
The output for collecting D1 eigenmatrix W1 and maker G is input in training network model DCGAN in arbiter D, passes through training net
Maker G and arbiter D in network model DCGAN compete with one another for resisting learning training, complete training network model DCGAN instruction
Practice.
Identification and classification network model includes being sequentially connected the input layer c connect, the first convolutional layer b, the second convolutional layer b, the 3rd
Convolutional layer b, Volume Four lamination b and softmax grader, wherein the Feature Mapping map number of softmax graders are 5.
The invention has the advantages that：
Classification of Polarimetric SAR Image method of the present invention based on DCGAN is in concrete operations, by training network
Model DCGAN is trained, and then reuses the arbiter D in training network model DCGAN, and two graders are replaced by
Softmax graders construct identification and classification network model, and realize dividing for Polarimetric SAR Image by identification and classification network model
Class, relative to other deep learning feature extracting methods, the present invention can be also realized to polarization without heuristic loss function
The classification of SAR image, so as to effectively improve the nicety of grading of Polarimetric SAR Image.In addition, it is necessary to which explanation, the present invention is logical
Cross and training network model DCGAN is trained, enable training network model DCGAN from a large amount of data untagged sample middle schools
Practise out the distribution character of data so that can still reach very high nicety of grading in the case where there is mark training sample less,
Overcoming in conventional polar SAR image assorting process has the problem of exemplar is less, nicety of grading is poor.
Brief description of the drawings
Fig. 1 is implementation process figure of the invention；
Fig. 2 is schemes to the handmarking of image to be classified in the present invention；
Fig. 3 is the classification results figure to image to be classified with the present invention.
Embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings：
With reference to Fig. 1, the Classification of Polarimetric SAR Image method of the present invention based on DCGAN comprises the following steps：
1) polarization scattering matrix S is obtained, Pauli decomposition is carried out to polarization scattering matrix S, odd scattering coefficient, even is obtained
Scattering coefficient and volume scattering coefficient, then it regard odd scattering coefficient, even scattering coefficient and volume scattering coefficient as polarization to be sorted
Eigenmatrix F of the 3-D view feature construction of SAR image based on pixel；
Wherein, Polarimetric SAR Image to be sorted in the present embodiment selects in April, 2008, SF Bay San
Francisco Bay full polarimetric SAR data images, resolution ratio is 50m, and image is L-band, and image size is 1800*1380, is owned
Pixel is divided into 5 classes.
Step 1) in polarization scattering matrix S carry out Pauli decomposition, obtain odd scattering coefficient, even scattering coefficient and
The operation of volume scattering coefficient is：
Pauli bases { S 1a) is set1,S2,S3, wherein,
Wherein, S1Scattered for odd, S2Scattered for even, S3For volume scattering；
1b) decomposed and defined by Pauli：
Wherein, a is odd scattering coefficient, and b is even scattering coefficient, and c is volume scattering coefficient；
Formula (2) 1c) is solved, odd scattering coefficient a, even scattering coefficient b and volume scattering coefficient c is obtained, wherein,
Step 1) in build the eigenmatrix F based on pixel concrete operations be：
If size is the eigenmatrix of M1 × M2 × 3, then by odd scattering coefficient a, even scattering coefficient b and volume scattering system
Number c is assigned to the eigenmatrix that size is M1 × M2 × 3, obtains the eigenmatrix F based on pixel, wherein, M1 is polarization to be sorted
The length of SAR image, M2 is the width of Polarimetric SAR Image to be sorted.
2) each element value in the eigenmatrix F based on pixel is normalized in [0,1], and by normalized result
It is denoted as eigenmatrix F1；
Wherein, conventional method for normalizing has characteristic line pantography, feature normalization and feature albefaction.In the present embodiment
Step 2) concrete operations be：The maximum max (F) of the eigenmatrix F based on pixel is solved, then by the spy based on pixel
Levy that each element in matrix F is equal divided by the maximum max (F), obtain eigenmatrix F1.
3) each element in eigenmatrix F1 is replaced by around it 64 × 64 image block, obtained based on image
The eigenmatrix F2 of block；
4) construct the eigenmatrix W1 without label training dataset D1 using the eigenmatrix F2 based on image block and have mark
Sign training dataset D2 eigenmatrix W2；
In the present embodiment, step 4) concrete operations be：
4a) Polarimetric SAR Image atural object is fallen into 5 types, the corresponding pixel of each classification is recorded in image to be classified
Position, position A1, A2, A3, A4, A5 of 5 kinds of correspondences of generation heterogeneously image vegetarian refreshments, wherein, A1 the 1st class atural object pixels of correspondence
Position of the point in image to be classified, position of A2 the 2nd class atural object pixels of correspondence in image to be classified, A3 the 3rd classes of correspondence
Position of the atural object pixel in image to be classified, position of A4 the 4th class atural object pixels of correspondence in image to be classified, A5 pairs
Answer position of the 5th class atural object pixel in image to be classified；
4b) randomly selected from described A1, A2, A3, A4, A5 0.5% element, generate 5 kinds of correspondence inhomogeneity atural object quilts
Position B1, B2, B3, B4, B5 of the pixel of training dataset are elected to be, wherein, B1 is to correspond to be selected as training in the 1st class atural object
Position of the pixel of data set in image to be classified, B2 is to correspond to the pixel that training dataset is selected as in the 2nd class atural object
The position in image to be classified is put, B3 is to correspond in the 3rd class atural object to be selected as the pixel of training dataset in figure to be sorted
Position as in, B4 is to correspond in the 4th class atural object to be selected as position of the pixel of training dataset in image to be classified, B5
To be selected as position of the pixel of training dataset in image to be classified in the 5th class atural object of correspondence, and by B1, B2, B3,
Element in B4, B5 merges position L1 of all pixels point of composition training dataset in image to be classified；
No label training dataset D1 eigenmatrix W1 4c) is defined, it is random in the eigenmatrix F2 based on image block
3.5% element of total pixel is selected to constitute the eigenmatrix W1 without label training dataset D1；
Label training dataset D2 eigenmatrix W2 4d) is defined, the foundation in the eigenmatrix F2 based on image block
L1 takes the value on correspondence position, and is assigned to training dataset D2 eigenmatrix W2.
5) using the eigenmatrix F2 construction test data set T based on image block, then in the eigenmatrix F based on pixel
Middle utilization SLIC super-pixel algorithm partition super-pixel block, obtains the cluster centre of super-pixel block, then in the feature based on image block
The eigenmatrix W3 of test data set T super-pixel cluster centre is constructed in matrix F 2；
Step 5) concrete operations be：
5a) with 5 kinds of correspondence inhomogeneity atural object quilts of remaining 99.5% Element generation in A1, A2, A3, A4, A5 described in step 4
Position C1, C2, C3, C4, C5 of the pixel of test data set are elected to be, wherein, C1 is to correspond to be selected as test in the 1st class atural object
Position of the pixel of data set in image to be classified, C2 is to correspond to the pixel that test data set is selected as in the 2nd class atural object
The position in image to be classified is put, C3 is to correspond in the 3rd class atural object to be selected as the pixel of test data set in figure to be sorted
Position as in, C4 is to correspond in the 4th class atural object to be selected as position of the pixel of test data set in image to be classified, C5
To be selected as position of the pixel of test data set in image to be classified in the 5th class atural object of correspondence, and by C1, C2, C3,
Element in C4, C5 merges position L2 of all pixels point of composition test data set in image to be classified；
5b) the pixel point set on correspondence position is taken to be test data set T according to L2；
5c) initialization seed point, according to the super-pixel number K=80000 of setting, the uniform distribution seed in image
Point, it is assumed that a total of N number of pixel of picture, pre-segmentation is the super-pixel of K identical sizes, then the size of each super-pixel is N/
K, then the distance (step-length) of neighboring seeds point be approximately：
The size for calculating each block of pixels is the ≈ 32 of (1800 × 1380) ÷ 80000, and the length of side of block of pixels is about 6；
Seed point 5d) is reselected in the n*n neighborhoods of seed point, specific method is：Calculate all pixels in the neighborhood
The Grad of point, the minimum position of the neighborhood inside gradient is moved on to by seed point；
It is 5e) which cluster each pixel distribution class label (belongs in the neighborhood around each seed point
The heart), hunting zone is limited to 2S*2S；
5f) distance metric include color distance and space length, for each pixel searched, calculate respectively it and
The distance of the seed point, distance calculating method is as follows：
Wherein, dcRepresent color distance, dcRepresent space length, NsIt is maximum space distance in class, is defined as Ns=S, is fitted
For each cluster, maximum color distance NcBoth it is different with picture difference, it is also different with cluster difference, due to each picture
Vegetarian refreshments can all be searched by multiple seed points, so each pixel can have the distance of one and surrounding seed point, take minimum
It is worth corresponding seed point as the cluster centre of the pixel；
5g) constantly iteration above-mentioned steps repartition artwork until error convergence using the length of side as 6, then have (1800 ×
1380) ÷ 36=69000 super-pixel block, here it is final super-pixel block number, records the middle imago of these super-pixel block
The position L3 of vegetarian refreshments；
The eigenmatrix W3 of test data set T super-pixel cluster centre 5h) is defined, in the eigenmatrix based on image block
Foundation L3 takes the value on correspondence position in F2, and is assigned to the eigenmatrix W3 of test data set T super-pixel cluster centre；
6) by being trained without label training dataset D1 to training network model DCGAN, the training after being trained
Network model DCGAN；
7) two graders in arbiter D in the training network model DCGAN after training are replaced by softmax classification
Device, then it regard the arbiter D after replacing as sorter network model；
8) the eigenmatrix W2 for having label training dataset D2 is input in sorter network model, and updates softmax
The parameter of grader, then by there is label training dataset D2 eigenmatrix W2 to update the ginseng of whole sorter network model
Number, is then classified by identification and classification network model to the eigenmatrix W3 of test data set T super-pixel cluster centre,
Labeled test data set T category, realizes the Classification of Polarimetric SAR Image based on DCGAN again.
Step 6) in pass through the concrete operations that are trained without label training dataset D1 to training network model DCGAN
For：
6a) setting the maker G in training network model DCGAN includes being sequentially connected the input layer a connect, the first deconvolution
Layer, the second warp lamination, the 3rd warp lamination, the 4th warp lamination and output layer, wherein, input layer a input is made an uproar for 100 dimensions
Sound vector；The Feature Mapping map number of first warp lamination is 512, and the filter size of the first warp lamination is 5；Second warp
The Feature Mapping map number of lamination is 256, and the filter size of the second warp lamination is 5；The Feature Mapping of 3rd warp lamination
Map number is 128, and the filter size of the 3rd warp lamination is 5；The Feature Mapping map number of 4th warp lamination is the 64, the 4th
The filter size of warp lamination is 5；Output layer exports the pcolor of 64 × 64 × 3 sizes；
6b) set arbiter D in training network model DCGAN include being sequentially connected the input layer b connect, the first convolutional layer a,
Second convolutional layer a, the 3rd convolutional layer a, Volume Four lamination a and two graders, wherein, input layer b Feature Mapping map number is
3；First layer convolutional layer a Feature Mapping map number is 64, and first layer convolutional layer a filter size is 5；Second layer convolutional layer
A Feature Mapping map number is 128, and second layer convolutional layer a filter size is 5；Third layer convolutional layer a Feature Mapping figure
Number is 256, and third layer convolutional layer a filter size is 5；4th layer of convolutional layer a Feature Mapping map number is 512, the
Four layers of convolutional layer a filter size is 5；Two graders export a scalar；
6c) the Uniform noise that input 100 is tieed up into training network model DCGAN maker G, will be without label training data
The output for collecting D1 eigenmatrix W1 and maker G is input in training network model DCGAN in arbiter D, passes through training net
Maker G and arbiter D in network model DCGAN compete with one another for resisting learning training, complete training network model DCGAN instruction
Practice.
Identification and classification network model includes being sequentially connected the input layer c connect, the first convolutional layer b, the second convolutional layer b, the 3rd
Convolutional layer b, Volume Four lamination b and softmax grader, wherein the Feature Mapping map number of softmax graders are 5；
The concrete operations that softmax graders are trained are：
Using the eigenmatrix W2 for having label training dataset D2 as the input of identification and classification network model, there is label training
The classification of each pixel trains softmax graders, by asking as the output of identification and classification network model in data set D2
Solve error between above-mentioned classification and the correct classification of handmarking and to error back-propagating, only update softmax graders
Parameter, the softmax graders trained, the correct category of handmarking is as shown in Figure 2.
Step 8) in pass through eigenmatrix W3 of the identification and classification network model to test data set T super-pixel cluster centre
The concrete operations classified are：
It regard the eigenmatrix W3 of test data set T super-pixel cluster centre as the identification and classification network model trained
Input, the identification and classification network model trained is output as the class categories of test data set T super-pixel cluster centre,
Then to classification of each pixel in test set T labeled as the cluster centre of super-pixel block where the pixel, so as to complete
The classification of test set.
Emulation experiment
Simulated conditions：Hardware platform is：HP Z840；Software platform is：TensorFlow；Emulation content and result：With this
Inventive method is tested under above-mentioned simulated conditions, and the unlabeled exemplars that 3.5% is chosen first (account for all total sample numbers
The 3.5% of 1800*1380, the block of about 8w pixel) unsupervised learning is carried out, then respectively from each of polarization SAR data
Randomly selected in classification 0.5% have after exemplar the 0.5% of sum (occupy exemplar), training softmax again to point
Class network is finely adjusted, and remaining markd pixel obtains the classification results such as Fig. 3, fall into 5 types ground as test sample
Thing.
As can be seen from Figure 3：Preferably, the edge clear after different zones are divided is distinguishable for the region consistency of classification results,
And it is also fewer to maintain noise in detailed information, classification results figure.
The present invention and convolutional neural networks CNN test data set nicety of grading are contrasted, as a result as shown in table 1：
Table 1
| Sorting technique | Convolutional neural networks | The present invention |
| Classification 1 (%) | 99.9849 | 99.9913 |
| Classification 2 (%) | 96.9208 | 98.5980 |
| Classification 3 (%) | 89.4749 | 98.5554 |
| Classification 4 (%) | 98.6352 | 99.6494 |
| Classification 5 (%) | 98.4097 | 99.0883 |
| Total accuracy rate | 97.254 | 99.4346 |
Reduction has exemplar to 0.2%, is contrasted with convolutional neural networks CNN test data set nicety of grading,
Nicety of grading is as shown in table 2 below：
Table 2
| Sorting technique | Convolutional neural networks | The present invention |
| Classification 1 (%) | 99.0036 | 99.9317 |
| Classification 2 (%) | 92.6015 | 97.6998 |
| Classification 3 (%) | 92.8231 | 98.8479 |
| Classification 4 (%) | 93.2422 | 97.8111 |
| Classification 5 (%) | 96.0703 | 97.3789 |
| Total accuracy rate | 95.9239 | 98.9805 |
From Table 1 and Table 2, under conditions of marked sample is 0.5% and 0.2%, test data set of the invention
The nicety of grading of each classification is above convolutional neural networks, it is possible to increase nicety of grading.
In summary, the present invention carries out feature extraction using DCGAN, can be from a large amount of data untagged learning data point
Cloth characteristic, with good character representation ability, i.e., can still be reached using a small amount of marked sample to the classification of polarization SAR data
To very high nicety of grading.
Claims (6)
1. a kind of Classification of Polarimetric SAR Image method based on DCGAN, it is characterised in that comprise the following steps：
1) polarization scattering matrix S is obtained, Pauli decomposition is carried out to polarization scattering matrix S, odd scattering coefficient, even scattering is obtained
Coefficient and volume scattering coefficient, then it regard odd scattering coefficient, even scattering coefficient and volume scattering coefficient as polarization SAR figure to be sorted
Eigenmatrix F of the 3-D view feature construction of picture based on pixel；
2) each element value in the eigenmatrix F based on pixel is normalized in [0,1], and normalized result is denoted as
Eigenmatrix F1；
3) each element in eigenmatrix F1 is replaced by around it 64 × 64 image block, obtained based on image block
Eigenmatrix F2；
4) construct the eigenmatrix W1 without label training dataset D1 using the eigenmatrix F2 based on image block and have label instruction
Practice data set D2 eigenmatrix W2；
5) the eigenmatrix F2 construction test data set T based on image block, then the profit in the eigenmatrix F based on pixel are utilized
With SLIC super-pixel algorithm partition super-pixel block, the cluster centre of super-pixel block is obtained, then in the eigenmatrix based on image block
The eigenmatrix W3 of test data set T super-pixel cluster centre is constructed in F2；
6) by being trained without label training dataset D1 to training network model DCGAN, the training network after being trained
Model DCGAN；
7) two graders in arbiter D in the training network model DCGAN after training are replaced by softmax graders, then
It regard the arbiter D after replacing as sorter network model；
8) the eigenmatrix W2 for having label training dataset D2 is input in sorter network model, and updates softmax classification
The parameter of device, then by there is label training dataset D2 eigenmatrix W2 to update the parameter of whole sorter network model, then
The eigenmatrix W3 of test data set T super-pixel cluster centre is classified by identification and classification network model, then marked
Test data set T category, realizes the Classification of Polarimetric SAR Image based on DCGAN.
2. the Classification of Polarimetric SAR Image method according to claim 1 based on DCGAN, it is characterised in that step 1) in it is right
Polarization scattering matrix S carries out Pauli decomposition, and the operation for obtaining odd scattering coefficient, even scattering coefficient and volume scattering coefficient is：
Pauli bases { S 1a) is set1,S2,S3, wherein,
<mrow>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>S</mi>
<mn>1</mn>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mn>2</mn>
</msqrt>
</mfrac>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>S</mi>
<mn>2</mn>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mn>2</mn>
</msqrt>
</mfrac>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>S</mi>
<mn>3</mn>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mn>2</mn>
</msqrt>
</mfrac>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, S1Scattered for odd, S2Scattered for even, S3For volume scattering；
1b) decomposed and defined by Pauli：
<mrow>
<mi>S</mi>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>S</mi>
<mrow>
<mi>H</mi>
<mi>H</mi>
</mrow>
</msub>
</mtd>
<mtd>
<msub>
<mi>S</mi>
<mrow>
<mi>H</mi>
<mi>V</mi>
</mrow>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>S</mi>
<mrow>
<mi>H</mi>
<mi>V</mi>
</mrow>
</msub>
</mtd>
<mtd>
<msub>
<mi>S</mi>
<mrow>
<mi>V</mi>
<mi>V</mi>
</mrow>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>=</mo>
<msub>
<mi>aS</mi>
<mn>1</mn>
</msub>
<mo>+</mo>
<msub>
<mi>bS</mi>
<mn>2</mn>
</msub>
<mo>+</mo>
<msub>
<mi>cS</mi>
<mn>3</mn>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, a is odd scattering coefficient, and b is even scattering coefficient, and c is volume scattering coefficient；
Formula (2) 1c) is solved, odd scattering coefficient a, even scattering coefficient b and volume scattering coefficient c is obtained, wherein,
<mrow>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>a</mi>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mn>2</mn>
</msqrt>
</mfrac>
<mrow>
<mo>(</mo>
<msub>
<mi>S</mi>
<mrow>
<mi>H</mi>
<mi>H</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>S</mi>
<mrow>
<mi>V</mi>
<mi>V</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>b</mi>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mn>2</mn>
</msqrt>
</mfrac>
<mrow>
<mo>(</mo>
<msub>
<mi>S</mi>
<mrow>
<mi>H</mi>
<mi>H</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>S</mi>
<mrow>
<mi>V</mi>
<mi>V</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>c</mi>
<mo>=</mo>
<msqrt>
<mn>2</mn>
</msqrt>
<msub>
<mi>S</mi>
<mrow>
<mi>H</mi>
<mi>V</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
3. the Classification of Polarimetric SAR Image method according to claim 1 based on DCGAN, it is characterised in that step 1) in structure
The concrete operations for building the eigenmatrix F based on pixel are：
If size is the eigenmatrix of M1 × M2 × 3, then by odd scattering coefficient a, even scattering coefficient b and volume scattering coefficient c
The eigenmatrix that size is M1 × M2 × 3 is assigned to, the eigenmatrix F based on pixel is obtained, wherein, M1 is polarization SAR to be sorted
The length of image, M2 is the width of Polarimetric SAR Image to be sorted.
4. the Classification of Polarimetric SAR Image method according to claim 1 based on DCGAN, it is characterised in that step 2) tool
Gymnastics conduct：The maximum max (F) of the eigenmatrix F based on pixel is solved, then by the eigenmatrix F based on pixel
Each element is equal divided by the maximum max (F), obtain eigenmatrix F1.
5. the Classification of Polarimetric SAR Image method according to claim 1 based on DCGAN, it is characterised in that step 6) in lead to
Cross no label training dataset D1 is to the training network model DCGAN concrete operations being trained：
6a) setting maker G in training network model DCGAN includes being sequentially connected the input layer a connect, the first warp lamination, the
Two warp laminations, the 3rd warp lamination, the 4th warp lamination and output layer, wherein, input layer a input for 100 dimension noises to
Amount；The Feature Mapping map number of first warp lamination is 512, and the filter size of the first warp lamination is 5；Second warp lamination
Feature Mapping map number be 256, the filter size of the second warp lamination is 5；The Feature Mapping figure number of 3rd warp lamination
Mesh is 128, and the filter size of the 3rd warp lamination is 5；The Feature Mapping map number of 4th warp lamination is 64, the 4th warp
The filter size of lamination is 5；Output layer exports the pcolor of 64 × 64 × 3 sizes；
6b) setting the arbiter D in training network model DCGAN includes being sequentially connected the input layer b connect, the first convolutional layer a, second
Convolutional layer a, the 3rd convolutional layer a, Volume Four lamination a and two graders, wherein, input layer b Feature Mapping map number is 3；The
One layer of convolutional layer a Feature Mapping map number is 64, and first layer convolutional layer a filter size is 5；Second layer convolutional layer a's
Feature Mapping map number is 128, and second layer convolutional layer a filter size is 5；Third layer convolutional layer a Feature Mapping figure number
Mesh is 256, and third layer convolutional layer a filter size is 5；4th layer of convolutional layer a Feature Mapping map number is the 512, the 4th
Layer convolutional layer a filter size is 5；Two graders export a scalar；
6c) the Uniform noise that input 100 is tieed up into training network model DCGAN maker G, will be without label training dataset D1
Eigenmatrix W1 and maker G output be input in training network model DCGAN in arbiter D, pass through training network mould
Maker G and arbiter D in type DCGAN compete with one another for resisting learning training, complete training network model DCGAN training.
6. the Classification of Polarimetric SAR Image method according to claim 1 based on DCGAN, it is characterised in that step 7) in sentence
Other sorter network model includes being sequentially connected the input layer c connect, the first convolutional layer b, the second convolutional layer b, the 3rd convolutional layer b, the
Four convolutional layer b and softmax graders, wherein the Feature Mapping map number of softmax graders are 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710440090.XA CN107292336A (en) | 2017-06-12 | 2017-06-12 | A kind of Classification of Polarimetric SAR Image method based on DCGAN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710440090.XA CN107292336A (en) | 2017-06-12 | 2017-06-12 | A kind of Classification of Polarimetric SAR Image method based on DCGAN |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107292336A true CN107292336A (en) | 2017-10-24 |
Family
ID=60096544
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710440090.XA Pending CN107292336A (en) | 2017-06-12 | 2017-06-12 | A kind of Classification of Polarimetric SAR Image method based on DCGAN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107292336A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944483A (en) * | 2017-11-17 | 2018-04-20 | 西安电子科技大学 | Classification of Multispectral Images method based on binary channels DCGAN and Fusion Features |
| CN107943751A (en) * | 2017-11-14 | 2018-04-20 | 华南理工大学 | A kind of autonomous channel convolution method based on depth convolution confrontation network model |
| CN108564006A (en) * | 2018-03-26 | 2018-09-21 | 西安电子科技大学 | Based on the polarization SAR terrain classification method from step study convolutional neural networks |
| CN109614979A (en) * | 2018-10-11 | 2019-04-12 | 北京大学 | A kind of data augmentation method and image classification method based on selection with generation |
| CN109784401A (en) * | 2019-01-15 | 2019-05-21 | 西安电子科技大学 | A kind of Classification of Polarimetric SAR Image method based on ACGAN |
| CN110009015A (en) * | 2019-03-25 | 2019-07-12 | 西北工业大学 | EO-1 hyperion small sample classification method based on lightweight network and semi-supervised clustering |
| WO2019237240A1 (en) * | 2018-06-12 | 2019-12-19 | 中国科学院深圳先进技术研究院 | Enhanced generative adversarial network and target sample identification method |
| CN110610207A (en) * | 2019-09-10 | 2019-12-24 | 重庆邮电大学 | Small sample SAR image ship classification method based on transfer learning |
| CN111192221A (en) * | 2020-01-07 | 2020-05-22 | 中南大学 | Aluminum electrolysis fire hole image repairing method based on deep convolution generation countermeasure network |
| CN112307679A (en) * | 2020-11-23 | 2021-02-02 | 内蒙古工业大学 | Method and device for constructing river ice thickness inversion microwave scattering model |
| CN116385813A (en) * | 2023-06-07 | 2023-07-04 | 南京隼眼电子科技有限公司 | ISAR image classification method, ISAR image classification device and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6943724B1 (en) * | 2002-10-30 | 2005-09-13 | Lockheed Martin Corporation | Identification and tracking of moving objects in detected synthetic aperture imagery |
| CN101488188A (en) * | 2008-11-10 | 2009-07-22 | 西安电子科技大学 | SAR image classification method based on SVM classifier of mixed nucleus function |
| CN102999908A (en) * | 2012-11-19 | 2013-03-27 | 西安电子科技大学 | Synthetic aperture radar (SAR) airport segmentation method based on improved visual attention model |
| CN104331707A (en) * | 2014-11-02 | 2015-02-04 | 西安电子科技大学 | Polarized SAR (synthetic aperture radar) image classification method based on depth PCA (principal component analysis) network and SVM (support vector machine) |
| 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 |
| CN105868793A (en) * | 2016-04-18 | 2016-08-17 | 西安电子科技大学 | Polarization SAR image classification method based on multi-scale depth filter |
-
2017
- 2017-06-12 CN CN201710440090.XA patent/CN107292336A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6943724B1 (en) * | 2002-10-30 | 2005-09-13 | Lockheed Martin Corporation | Identification and tracking of moving objects in detected synthetic aperture imagery |
| CN101488188A (en) * | 2008-11-10 | 2009-07-22 | 西安电子科技大学 | SAR image classification method based on SVM classifier of mixed nucleus function |
| CN102999908A (en) * | 2012-11-19 | 2013-03-27 | 西安电子科技大学 | Synthetic aperture radar (SAR) airport segmentation method based on improved visual attention model |
| CN104331707A (en) * | 2014-11-02 | 2015-02-04 | 西安电子科技大学 | Polarized SAR (synthetic aperture radar) image classification method based on depth PCA (principal component analysis) network and SVM (support vector machine) |
| 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 |
| CN105868793A (en) * | 2016-04-18 | 2016-08-17 | 西安电子科技大学 | Polarization SAR image classification method based on multi-scale depth filter |
Non-Patent Citations (3)
| Title |
|---|
| ALEC RADFORD等: "UNSUPERVISED REPRESENTATION LEARNING WITH DEEP CONVOLUTIONAL GENERATIVE ADVERSARIAL NETWORKS", 《 ARXIV》 * |
| 史彩娟等: "基于增强稀疏性特征选择的网络图像标注", 《软件学报》 * |
| 王鑫: "SAR图像显著性检测与分类算法研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107943751A (en) * | 2017-11-14 | 2018-04-20 | 华南理工大学 | A kind of autonomous channel convolution method based on depth convolution confrontation network model |
| CN107944483B (en) * | 2017-11-17 | 2020-02-07 | 西安电子科技大学 | Multispectral image classification method based on dual-channel DCGAN and feature fusion |
| CN107944483A (en) * | 2017-11-17 | 2018-04-20 | 西安电子科技大学 | Classification of Multispectral Images method based on binary channels DCGAN and Fusion Features |
| CN108564006A (en) * | 2018-03-26 | 2018-09-21 | 西安电子科技大学 | Based on the polarization SAR terrain classification method from step study convolutional neural networks |
| CN108564006B (en) * | 2018-03-26 | 2021-10-29 | 西安电子科技大学 | Polarized SAR terrain classification method based on self-learning convolutional neural network |
| WO2019237240A1 (en) * | 2018-06-12 | 2019-12-19 | 中国科学院深圳先进技术研究院 | Enhanced generative adversarial network and target sample identification method |
| CN109614979A (en) * | 2018-10-11 | 2019-04-12 | 北京大学 | A kind of data augmentation method and image classification method based on selection with generation |
| CN109614979B (en) * | 2018-10-11 | 2023-05-02 | 北京大学 | Data augmentation method and image classification method based on selection and generation |
| CN109784401A (en) * | 2019-01-15 | 2019-05-21 | 西安电子科技大学 | A kind of Classification of Polarimetric SAR Image method based on ACGAN |
| CN110009015A (en) * | 2019-03-25 | 2019-07-12 | 西北工业大学 | EO-1 hyperion small sample classification method based on lightweight network and semi-supervised clustering |
| CN110610207A (en) * | 2019-09-10 | 2019-12-24 | 重庆邮电大学 | Small sample SAR image ship classification method based on transfer learning |
| CN110610207B (en) * | 2019-09-10 | 2022-11-25 | 重庆邮电大学 | Small sample SAR image ship classification method based on transfer learning |
| CN111192221A (en) * | 2020-01-07 | 2020-05-22 | 中南大学 | Aluminum electrolysis fire hole image repairing method based on deep convolution generation countermeasure network |
| CN111192221B (en) * | 2020-01-07 | 2024-04-16 | 中南大学 | Aluminum electrolysis fire hole image repairing method based on deep convolution generation countermeasure network |
| CN112307679A (en) * | 2020-11-23 | 2021-02-02 | 内蒙古工业大学 | Method and device for constructing river ice thickness inversion microwave scattering model |
| CN116385813A (en) * | 2023-06-07 | 2023-07-04 | 南京隼眼电子科技有限公司 | ISAR image classification method, ISAR image classification device and storage medium |
| CN116385813B (en) * | 2023-06-07 | 2023-08-29 | 南京隼眼电子科技有限公司 | ISAR image space target classification method, device and storage medium based on unsupervised contrast learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107292336A (en) | A kind of Classification of Polarimetric SAR Image method based on DCGAN | |
| CN110210486B (en) | Sketch annotation information-based generation countermeasure transfer learning method | |
| CN107563428B (en) | Based on the Classification of Polarimetric SAR Image method for generating confrontation network | |
| CN105184309B (en) | Classification of Polarimetric SAR Image based on CNN and SVM | |
| CN107368852A (en) | A kind of Classification of Polarimetric SAR Image method based on non-down sampling contourlet DCGAN | |
| CN103413151B (en) | Hyperspectral image classification method based on figure canonical low-rank representation Dimensionality Reduction | |
| CN108491849A (en) | Hyperspectral image classification method based on three-dimensional dense connection convolutional neural networks | |
| CN107194433A (en) | A kind of Radar range profile's target identification method based on depth autoencoder network | |
| CN104484681B (en) | Hyperspectral Remote Sensing Imagery Classification method based on spatial information and integrated study | |
| CN106650830A (en) | Deep model and shallow model decision fusion-based pulmonary nodule CT image automatic classification method | |
| CN107622272A (en) | A kind of image classification method and device | |
| CN104200217B (en) | Hyperspectrum classification method based on composite kernel function | |
| CN104732244B (en) | The Classifying Method in Remote Sensing Image integrated based on wavelet transformation, how tactful PSO and SVM | |
| CN104299232B (en) | SAR image segmentation method based on self-adaptive window directionlet domain and improved FCM | |
| CN107229917A (en) | A kind of several remote sensing image general character well-marked target detection methods clustered based on iteration | |
| CN104517284A (en) | Polarimetric SAR (synthetic aperture radar) image segmentation based on DBN (deep belief network) | |
| CN104166859A (en) | Polarization SAR image classification based on SSAE and FSALS-SVM | |
| CN105138970A (en) | Spatial information-based polarization SAR image classification method | |
| CN102999762B (en) | Decompose and the Classification of Polarimetric SAR Image method of spectral clustering based on Freeman | |
| CN106600595A (en) | Human body characteristic dimension automatic measuring method based on artificial intelligence algorithm | |
| CN107169492A (en) | Polarization SAR object detection method based on FCN CRF master-slave networks | |
| CN103996047A (en) | Hyperspectral image classification method based on compression spectrum clustering integration | |
| CN105760900A (en) | Hyperspectral image classification method based on affinity propagation clustering and sparse multiple kernel learning | |
| CN104298974A (en) | Human body behavior recognition method based on depth video sequence | |
| CN106548445A (en) | Spatial domain picture general steganalysis method based on content |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171024 |