CN107909120A - Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method - Google Patents

Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method Download PDF

Info

Publication number
CN107909120A
CN107909120A CN201711454166.0A CN201711454166A CN107909120A CN 107909120 A CN107909120 A CN 107909120A CN 201711454166 A CN201711454166 A CN 201711454166A CN 107909120 A CN107909120 A CN 107909120A
Authority
CN
China
Prior art keywords
sample
mrow
dictionary
msub
class
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
Application number
CN201711454166.0A
Other languages
Chinese (zh)
Inventor
纪则轩
谢梦蓝
孙权森
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN201711454166.0A priority Critical patent/CN107909120A/en
Publication of CN107909120A publication Critical patent/CN107909120A/en
Pending legal-status Critical Current

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
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches

Abstract

The invention discloses a kind of based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method.EO-1 hyperion 3-D view is converted into two-dimensional matrix, pixel is extracted as training sample according to same ratio to every a kind of sample, remaining as test sample, training sample is carried out to change by class extension and label, according to the training sample after extension and target dictionary structure mutual exclusion tag entry, K SVD dictionary learnings are carried out after deformation, establishing multiple dimensioned neighborhood union for test sample obtains sparse coding, determines that pixel class completes classification according to minimal reconstruction error.This method expands sample and dictionary atom by the innovation structure of dictionary model, breaches original dictionary learning and number of training purpose is relied on, solve the problems, such as small sample problem and imbalanced training sets；Dictionary learning early period neighborhood coding multiple dimensioned with the later stage is combined at the same time so that the advantage of two kinds of algorithms is played to greatest extent, realizes the high-precision quantitative analysis of classification hyperspectral imagery.

Description

Based on alternative label K-SVD and multiple dimensioned sparse hyperspectral image classification method
Technical field
The present invention relates to hyperspectral image classification method, it is particularly a kind of based on alternative label K-SVD dictionary learnings and The rarefaction representation sorting technique of multiple dimensioned neighborhood.
Background technology
In recent years, the development of classification hyperspectral imagery is quite varied, in the application of military, agricultural, environmental monitoring etc. It is more and more urgent.Particularly supervised classification method, can make full use of the spectral signature and label information of high spectrum image, such as most Small distance classification, maximum likelihood classification etc..But since these methods do not account for image sparse characteristic in itself and space letter Breath, therefore classifying quality is bad.Be subject to " human visual attention mechanism have openness " this inspiration, rarefaction representation meet the tendency of and Raw, the application in fields such as noise reduction, fusion, pattern-recognitions has been shown to have powerful ability.In high spectrum image, belong to Of a sort pixel is usually located at same low-rank subspace, that is, the dictionary atom being turned into belongs to same class, therefore rarefaction representation is same Sample is suitably applied in the classification of high spectrum image.That is, it is desirable to determine the correct label of unknown pixel, can be by right Unknown pixel carries out the solution based on dictionary atom and respective weights (i.e. sparse coefficient), this namely rarefaction representation sorting technique Basic thought.The research for the rarefaction representation sorting technique of high spectrum image mainly had following two directions in recent years：
(1) dictionary learning., directly by training sample vector composition structure dictionary, develop into and training sample is carried out from initially Study dictionary is obtained after study.The K-SVD not learnt from no labeling learns dictionary, develops into the linear discriminant from tape label The mark that dictionary (Fisher Discrimination Dictionary Learning, FDDL), dictionary and grader learn at the same time Sign uniformity KSVD (Label Consistent K-SVD, LC-KSVD) study dictionaries etc..
(2) sparse coding.The utilization for being mainly focused on image space information is studied in this part, and certain methods are to EO-1 hyperion picture Element carry out pixel-by-pixel sparse coding, this kind of method fail the space characteristics of image and spectral signature being combined, classifying quality compared with Difference；Other methods then consider to establish neighborhood using test pixel as center pixel, to pixel progress sparse coefficient in neighborhood Joint solves, to improve nicety of grading.
Therefore, in terms of dictionary learning, the quality of dictionary learning is influenced be subject to number of training purpose, if certain a kind of sample This training sample number far fewer than other classifications, or it is of all categories between number of samples it is unbalanced, then the classification of these classes Precision is possible to differ greatly.And in terms of sparse coding, if only considering to encode or only establish single scale neighborhood pixel-by-pixel, The otherness of different zones (homogenous region and fringe region) cannot be then made full use of, and then influences the overall classification essence of image Degree.
The content of the invention
The K-SVD dictionary learning sides influenced it is an object of the invention to provide a kind of tape label and from small sample problem Method, and existing multiple dimensioned neighborhood sparse coding is combined, complete the rarefaction representation high-precision classification of high spectrum image.
The technical solution for realizing the object of the invention is：It is a kind of based on alternative label K-SVD and multiple dimensioned sparse Hyperspectral image classification method, comprises the following steps：
Step 1, by EO-1 hyperion 3-D view be converted to two-dimensional matrix；
Step 2, extract pixel as training sample to every a kind of sample according to same ratio, remaining as test sample；
Step 3, carry out training sample to change by class extension and label；
Training sample and target dictionary structure mutual exclusion tag entry after step 4, foundation extension, carry out K-SVD words after deformation Allusion quotation learns；
Step 5, be that test sample establishes multiple dimensioned neighborhood union acquisition sparse coding；
Step 6, determine pixel class completion classification according to minimal reconstruction error.
Compared with prior art, remarkable advantage of the invention is：By carrying out positive negative sample extension to training sample, build The alternative tag entry for allusion quotation of signing an agreement, a contract, a receipt, etc., under the premise of training sample is unbalanced, breaches in traditional dictionary learning method, due to portion The problem of such dictionary ability to express is weaker caused by sub-category training sample number is small, it is dilute in combination with multiple dimensioned neighborhood Presentation technology is dredged, realizes the high-precision classification of high spectrum image.
Brief description of the drawings
Fig. 1 is the flow of the invention based on alternative label K-SVD and multiple dimensioned sparse hyperspectral image classification method Figure.
Fig. 2 is the imaging cube and standard category block diagram of Indian_Pines EO-1 hyperion 3-D views.Zuo Wei The three-dimensional imaging cube of Indian_Pines, the right side is standard category block diagram.
Fig. 3 be 16 class label color lump legend of Indian_Pines images and according to 10% ratio extract training sample with Test sample number exemplary plot.
Fig. 4 is the classification results exemplary plot that the present invention obtains.
Fig. 5 is the comparative example figure for the classification results that three kinds of methods obtain.
Embodiment
The invention will be further described below in conjunction with the accompanying drawings.
With reference to Fig. 1, the present invention is included based on alternative label K-SVD and multiple dimensioned sparse hyperspectral image classification method Following steps：
Step 1, conversion EO-1 hyperion 3-D view, using its spectral signature as atom dimension, image pixel is arranged by row As total atom number.The imaging cube and standard category piecemeal of EO-1 hyperion 3-D view are as shown in Figure 2.
Step 2, for a high spectrum image for including M class atural objects, extracted according to some percentage per a kind of image slices Element is used as training sample, and residual pixel is as test sample.Training sample number exemplary plot is as shown in Figure 3.
Step 3, will extend to overall training sample number per a kind of number of training mesh, wherein belonging to such pixel " just " class is defined as, corresponding label is " 1 ", and the pixel definition for being not belonging to such and belonging to remaining (M-1) class is " negative " class, corresponding Label is " 0 ".Sample and sample extension are as shown in Figure 1.
Step 4, based on classical K-SVD dictionary learnings algorithm, structure has in class alternative tag entry between uniformity and class, That is the nonzero value (and nonzero value is 1) of tag entry appears in training sample signal and dictionary atom shares those ropes of same label Draw place.So that dictionary atomic tag is corresponded with the positive negative training sample label in step 3.
Step 5, for each test sample establish multiple dimensioned neighborhood union and be coupled all scale neighborhoods, with SOMP Algorithm for Solving The sparse coefficient of test sample.Multiple dimensioned neighborhood is coupled as shown in Figure 1 with coding.
Step 6, by minimal reconstruction error of the code coefficient on dictionary determine current pixel generic, determines institute Whole image classification is completed after having test sample generic.
With reference to example, the present invention will be further described.
The present invention expresses 3-D view using EO-1 hyperion 3-D view as input, by dimension transformation, generates two dimension classification Image.
It is as shown in Figure 1 that the present invention implements exemplary flow.
(1) downloaded in GIC (Grupo De Inteligencia Computacional) EO-1 hyperion scene image on website Obtained Indian_Pines images size is 145 × 145 × 224, by removing the wave band of covering suction zone by wave band number Amount is reduced to 200, that is, the usable image size after correcting gives EO-1 hyperion 3-D view for 145 × 145 × 200, Fig. 2 It is imaged cube and standard category block diagram.A left side is the three-dimensional imaging cube of Indian_Pines, and the right side is standard category piecemeal Scheme, totally 16 class.
(2) for the three-dimensional high-spectral data Indian_Pines that a size is 145 × 145 × 200, it is by size Each pixel in 145 × 145 two-dimensional projection image is considered as a sample, and each sample has 200 dimensional features, With reference to carrying class label (Groundtruth), from it is every it is a kind of in randomly select 10% sample composing training collection (training sample be total Number is 1027), remaining sample is as test set.Fig. 3 give 16 class label color lump legend of Indian_Pines images and according to The training sample and test sample number exemplary plot of 10% ratio extraction.
(3) Indian_Pines images include 16 class atural objects, will extend to overall training per a kind of number of training mesh Number of samples 1027, wherein the original pixel for belonging to such is defined as " just " class, corresponding label is " 1 ", is not belonging to such and belongs to The pixel definition of remaining 15 classes is " negative " class, and corresponding label is " 0 ".Training sample sum after extension is 1027 × 16, sample And sample extension is as shown in Figure 1.
(4) the 3rd plate of Fig. 1 top halfs gives the core dictionary learning schematic diagram of the present invention.Based on classical K- SVD dictionary learning algorithms, structure is with alternative tag entry H, the i.e. nonzero value of H appear in trained sample between uniformity and class in class This signal and dictionary atom are shared at those indexes of same label so that dictionary atomic tag and the positive and negative training in step 3 Sample label corresponds.Assuming that the first kind has 1 training sample, the second class has 2 training samples, and three classes have 3 training Sample, then the tag entry h1 of the first kind is：
The total tag entry H of three classes is：
(5) Fig. 1 the latter half gives Test code stage schematic diagram.Due to adjacent pixel maximum probability with belonging to same class Thing, and large scale neighborhood is suitable for smooth (or large area) region, small scale neighborhood is suitable for border (or small area) region, because This pixel centered on each test sample pixel establishes multiple dimensioned neighborhood, and it is seven kinds big to be specifically divided into 3*3,5*5 ..., 15*15 It is small, the pixel in these regions is arranged and is coupled by row, for the overall multiple dimensioned Neighborhood matrix of each test sample structure, is used in combination SOMP algorithms obtain the sparse coefficient of each center pixel (i.e. test sample).
(6) according to the product of gained study dictionary and sparse coefficient in abovementioned steps, can be done with test specimens primitive vector Difference tries to achieve reconstructed error, and the correspondence classification (the class label of i.e. corresponding sub- dictionary) of the minimum value in reconstructed error is current test The judgement classification of sample.The kind judging for completing all test samples obtains the classification knot of whole Indian_Pines images Fruit.
(7) Fig. 4 gives the part classifying of the present invention as a result, wherein first is classified as three width sizes, wave band and scene not Same three frequency color composite image of original EO-1 hyperion, second is classified as criteria classification as a result, the 3rd is classified as classification results of the present invention.From Understood in figure：The present invention overcomes small sample problem and imbalanced training sets problem, realizes every a kind of high-precision classification.Fig. 5 The comparative example figure for the classification results that three kinds of methods obtain is given, wherein first is classified as criteria classification as a result, secondary series and Three are classified as the classification results of two kinds of comparative approach, and the 4th is classified as classification results of the present invention.As can be known from Fig. 5：It is compared to tradition Sorting technique for, the present invention from sample extension, K-SVD dictionary learning methods are improved, by each category dictionary Positive class and negative class are resorted to as, while being described in itself with such, using being not belonging to such other classification reverse scans Such is stated, the dictionary for learn from positive and negative two angles has in class alternative between uniformity and class at the same time, overcomes sample This problem and imbalanced training sets problem so that the sample of each classification can realize high-precision classification.The present invention obtains flat Equal nicety of grading and the stability of algorithm are better than existing method, realize the high-precision fixed of high spectrum image terrain classification Amount analysis, has important practical significance for crops monitoring etc..

Claims (5)

  1. It is 1. a kind of based on alternative label K-SVD and multiple dimensioned sparse hyperspectral image classification method, it is characterised in that including Following steps：
    Step 1, by EO-1 hyperion 3-D view be converted to two-dimensional matrix；
    Step 2, extract pixel as training sample to every a kind of sample according to same ratio, remaining as test sample；
    Step 3, carry out training sample to change by class extension and label；
    Training sample and target dictionary structure mutual exclusion tag entry after step 4, foundation extension, carry out K-SVD dictionaries after deformation Practise；
    Step 5, be that test sample establishes multiple dimensioned neighborhood union acquisition sparse coding；
    Step 6, determine pixel class completion classification according to minimal reconstruction error.
  2. 2. hyperspectral image classification method according to claim 1, it is characterised in that：To training sample described in step 3 Extended by class and the specific method of label conversion is：
    Be M for a classification number, first kind number of training be N1, high spectrum image that total number of training is N, by first Class number of training is extended, wherein the N of script1A first kind sample is still expressed as such, is changed in form " just " class, corresponding " 1 " label, and belong to the (N-N of other (M-1) classes1) a sample is unified for " negative class ", it is extended to the first kind , corresponding " 0 " label；
    When extending remaining (M-1) class, Nc c classes samples of script are still expressed as such, are changed in form " just " Class, corresponding " 1 " label, (N-Nc) a sample for belonging to other (M-1) classes is unified for " negative class ", is extended to c classes, right Answer " 0 " label.
  3. 3. hyperspectral image classification method according to claim 1, it is characterised in that：The concrete methods of realizing of step 4 is：
    For a target dictionary, make dictionary atomicity consistent with number of training, corresponding one of each dictionary item is specific Label, dictionary are expressed as D=[D1,D2,...,DC], wherein Dc corresponds to the sub- dictionary of c classes；Define a classification mutual exclusion The correspondence of property label, the clear and definite dictionary of the label and training sample；
    Objective function is as follows：
    <mrow> <mo>&lt;</mo> <mi>D</mi> <mo>,</mo> <mi>A</mi> <mo>&gt;</mo> <mo>=</mo> <mi>arg</mi> <munder> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <mrow> <mi>D</mi> <mo>,</mo> <mi>A</mi> </mrow> </munder> <mo>{</mo> <mo>|</mo> <mo>|</mo> <msub> <mi>Y</mi> <mrow> <mi>n</mi> <mi>e</mi> <mi>w</mi> </mrow> </msub> <mo>-</mo> <mi>D</mi> <mi>A</mi> <mo>|</mo> <msubsup> <mo>|</mo> <mn>2</mn> <mn>2</mn> </msubsup> <mo>+</mo> <mi>&amp;alpha;</mi> <mo>|</mo> <mo>|</mo> <mi>H</mi> <mo>-</mo> <mi>P</mi> <mi>A</mi> <mo>|</mo> <msubsup> <mo>|</mo> <mn>2</mn> <mn>2</mn> </msubsup> <mo>}</mo> <mo>,</mo> <mi>s</mi> <mo>.</mo> <mi>t</mi> <mo>.</mo> <mo>&amp;ForAll;</mo> <mi>i</mi> <mo>,</mo> <mo>|</mo> <mo>|</mo> <msub> <mi>A</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mo>|</mo> <mn>0</mn> </msub> <mo>&amp;le;</mo> <msub> <mi>T</mi> <mn>0</mn> </msub> </mrow>
    In above formula, YnewIt is the training sample set after being extended by step 3；Alternative tag entry H=[h are built for dictionary1, h2,...,hC], each class tag entryEach list be shown asIts nonzero value Appear in input signal yiWith dictionary item diAt those indexes of shared same label, and nonzero value is 1, claims hiIt is to believe with input Number yiCorresponding " alternative " Sparse Code；ɑ obtains optimal value as a balance parameters by cross validation；A is training sample Corresponding sparse coding；P is solved by right formula in the algorithm as a linear transformation matrix, P：P=HAt(AAt+I)-1, its Middle I is and AAtThe identical unit matrix of size；Alternative sparse coding error is generally designated as, after linear transformation Sparse coding PA be similar to alternative Sparse Code H, it using exclusive message increase from positive and negative two in terms of come from inhomogeneity letter Number identification；
    After following deformation is done to object function, you can complete dictionary learning according to classical K-SVD algorithms：
    <mrow> <mo>&lt;</mo> <mi>D</mi> <mo>,</mo> <mi>A</mi> <mo>&gt;</mo> <mo>=</mo> <mi>arg</mi> <munder> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <mrow> <mi>D</mi> <mo>,</mo> <mi>A</mi> </mrow> </munder> <mo>|</mo> <mo>|</mo> <mfenced open = "(" close = ")"> <mtable> <mtr> <mtd> <msub> <mi>Y</mi> <mrow> <mi>n</mi> <mi>e</mi> <mi>w</mi> </mrow> </msub> </mtd> </mtr> <mtr> <mtd> <mrow> <msqrt> <mi>&amp;alpha;</mi> </msqrt> <mi>H</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mfenced open = "(" close = ")"> <mtable> <mtr> <mtd> <mi>D</mi> </mtd> </mtr> <mtr> <mtd> <mrow> <msqrt> <mi>&amp;alpha;</mi> </msqrt> <mi>P</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> <mi>A</mi> <mo>|</mo> <msubsup> <mo>|</mo> <mn>2</mn> <mn>2</mn> </msubsup> <mo>,</mo> <mi>s</mi> <mo>.</mo> <mi>t</mi> <mo>.</mo> <mo>&amp;ForAll;</mo> <mi>i</mi> <mo>,</mo> <mo>|</mo> <mo>|</mo> <msub> <mi>A</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mo>|</mo> <mn>0</mn> </msub> <mo>&amp;le;</mo> <msub> <mi>T</mi> <mn>0</mn> </msub> <mo>.</mo> </mrow>
  4. 4. hyperspectral image classification method according to claim 1, it is characterised in that：Built described in step 5 for test sample Vertical multiple dimensioned neighborhood union obtains sparse coding, and specific method is：
    Pixel establishes multiple dimensioned neighborhood centered on each test sample pixel, and the pixel in each dimensional area is arranged by row And it is coupled all scales, overall Neighborhood matrix is built for each test sample, and each test sample is obtained with SOMP algorithms Sparse coefficient.
  5. 5. hyperspectral image classification method according to claim 1, it is characterised in that：According to minimal reconstruction described in step 6 Error determines that pixel class completes classification, and specific method is：
    Product obtained by test sample vector subtracts sparse coefficient that step 5 obtains and step 4 per category dictionary, that is, obtain each The reconstructed error of class, the corresponding sub- dictionary classification of minimum value is the sample generic in reconstructed error；Complete each survey Classification is completed after the kind judging of sample sheet.
CN201711454166.0A 2017-12-28 2017-12-28 Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method Pending CN107909120A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711454166.0A CN107909120A (en) 2017-12-28 2017-12-28 Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711454166.0A CN107909120A (en) 2017-12-28 2017-12-28 Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method

Publications (1)

Publication Number Publication Date
CN107909120A true CN107909120A (en) 2018-04-13

Family

ID=61871724

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711454166.0A Pending CN107909120A (en) 2017-12-28 2017-12-28 Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method

Country Status (1)

Country Link
CN (1) CN107909120A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109409432A (en) * 2018-10-31 2019-03-01 腾讯科技（深圳）有限公司 A kind of image processing method, device and storage medium
CN110363089A (en) * 2019-06-13 2019-10-22 南京理工大学 Based on the hyperspectral image classification method for differentiating sub- dictionary and multiple dimensioned super-pixel
CN110717354A (en) * 2018-07-11 2020-01-21 哈尔滨工业大学 Superpixel classification method based on semi-supervised K-SVD and multi-scale sparse representation
CN114332534A (en) * 2021-12-29 2022-04-12 山东省科学院海洋仪器仪表研究所 Hyperspectral image small sample classification method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104392463A (en) * 2014-12-16 2015-03-04 西安电子科技大学 Image salient region detection method based on joint sparse multi-scale fusion
CN104408478A (en) * 2014-11-14 2015-03-11 西安电子科技大学 Hyperspectral image classification method based on hierarchical sparse discriminant feature learning
CN105303198A (en) * 2015-11-17 2016-02-03 福州大学 Remote-sensing image semi-supervision classification method based on customized step-size learning
CN105528623A (en) * 2016-01-09 2016-04-27 北京工业大学 Imaging spectrum image sparse representation method based on ground object class classification redundant dictionary

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104408478A (en) * 2014-11-14 2015-03-11 西安电子科技大学 Hyperspectral image classification method based on hierarchical sparse discriminant feature learning
CN104392463A (en) * 2014-12-16 2015-03-04 西安电子科技大学 Image salient region detection method based on joint sparse multi-scale fusion
CN105303198A (en) * 2015-11-17 2016-02-03 福州大学 Remote-sensing image semi-supervision classification method based on customized step-size learning
CN105528623A (en) * 2016-01-09 2016-04-27 北京工业大学 Imaging spectrum image sparse representation method based on ground object class classification redundant dictionary

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
LEYUAN FANG ET AL.: "Spectral-spatial hyperspectral image classification via multiscale adaptive sparse representation", 《IEEE TRANSACTIONS ON GEOSCIENCES AND REMOTE SENSING》 *
ZIYU WANG ET AL.: "Joint sparse model-based discriminative K-svd for hyperspectral image classification", 《SIGNAL PROCESSING》 *
张晓龙: "基于判别式K-SVD的车标识别方法研究", 《万方硕士学位论文全文数据库》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110717354A (en) * 2018-07-11 2020-01-21 哈尔滨工业大学 Superpixel classification method based on semi-supervised K-SVD and multi-scale sparse representation
US10691974B2 (en) * 2018-07-11 2020-06-23 Harbin Institute Of Technology Superpixel classification method based on semi-supervised K-SVD and multiscale sparse representation
CN109409432A (en) * 2018-10-31 2019-03-01 腾讯科技（深圳）有限公司 A kind of image processing method, device and storage medium
CN109409432B (en) * 2018-10-31 2019-11-26 腾讯科技（深圳）有限公司 A kind of image processing method, device and storage medium
CN110363089A (en) * 2019-06-13 2019-10-22 南京理工大学 Based on the hyperspectral image classification method for differentiating sub- dictionary and multiple dimensioned super-pixel
CN114332534A (en) * 2021-12-29 2022-04-12 山东省科学院海洋仪器仪表研究所 Hyperspectral image small sample classification method
CN114332534B (en) * 2021-12-29 2024-03-29 山东省科学院海洋仪器仪表研究所 Hyperspectral image small sample classification method

Similar Documents

Publication Publication Date Title
Hafemann et al. Forest species recognition using deep convolutional neural networks
Zou et al. Deep learning based feature selection for remote sensing scene classification
Kirk et al. L* a* b* fruits: A rapid and robust outdoor fruit detection system combining bio-inspired features with one-stage deep learning networks
CN104166859B (en) Based on SSAE and FSALS SVM Classification of Polarimetric SAR Image
CN103714354B (en) Hyperspectral image wave band selection method based on quantum-behaved particle swarm optimization algorithm
CN105069478B (en) High-spectrum remote-sensing terrain classification method based on super-pixel tensor sparse coding
CN107909120A (en) Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method
CN107977932A (en) It is a kind of based on can differentiate attribute constraint generation confrontation network face image super-resolution reconstruction method
CN108520275A (en) A kind of regular system of link information based on adjacency matrix, figure Feature Extraction System, figure categorizing system and method
CN106845418A (en) A kind of hyperspectral image classification method based on deep learning
CN108960260B (en) Classification model generation method, medical image classification method and medical image classification device
CN103886342A (en) Hyperspectral image classification method based on spectrums and neighbourhood information dictionary learning
CN104484681A (en) Hyperspectral remote sensing image classification method based on space information and ensemble learning
CN106156798B (en) Scene image classification method based on annular space pyramid and Multiple Kernel Learning
Ensafi et al. A bag of words based approach for classification of HEp-2 cell images
CN108460391A (en) Based on the unsupervised feature extracting method of high spectrum image for generating confrontation network
CN104298974A (en) Human body behavior recognition method based on depth video sequence
CN105718932A (en) Colorful image classification method based on fruit fly optimization algorithm and smooth twinborn support vector machine and system thereof
CN105046272A (en) Image classification method based on concise unsupervised convolutional network
CN108108719A (en) A kind of Weighted Kernel is sparse and cooperates with the Hyperspectral Image Classification method for representing coefficient
CN106408009B (en) Neighborhood weighted average hyperspectral image classification method based on depth confidence network
CN104573699A (en) Trypetid identification method based on medium field intensity magnetic resonance dissection imaging
CN116912708A (en) Remote sensing image building extraction method based on deep learning
Patel et al. Convolutional neural network with transfer learning for rice type classification
Xiang et al. MC-JAFN: Multilevel contexts-based joint attentive fusion network for pansharpening

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: 20180413