CN109598284A - A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics - Google Patents
A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics Download PDFInfo
- Publication number
- CN109598284A CN109598284A CN201811238577.0A CN201811238577A CN109598284A CN 109598284 A CN109598284 A CN 109598284A CN 201811238577 A CN201811238577 A CN 201811238577A CN 109598284 A CN109598284 A CN 109598284A
- Authority
- CN
- China
- Prior art keywords
- ldm
- information
- spatial
- svm
- hyperspectral image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 100
- 238000001228 spectrum Methods 0.000 claims abstract description 36
- 238000013480 data collection Methods 0.000 claims abstract description 22
- 230000004927 fusion Effects 0.000 claims abstract description 21
- 230000009467 reduction Effects 0.000 claims abstract description 13
- 239000000284 extract Substances 0.000 claims abstract description 7
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 29
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000001914 filtration Methods 0.000 claims description 11
- 230000002146 bilateral effect Effects 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 7
- 239000012141 concentrate Substances 0.000 claims description 6
- 238000002790 cross-validation Methods 0.000 claims description 6
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 description 13
- 241001466077 Salina Species 0.000 description 8
- 238000002474 experimental method Methods 0.000 description 4
- 230000003595 spectral effect Effects 0.000 description 4
- 238000010521 absorption reaction Methods 0.000 description 2
- 238000003331 infrared imaging Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000012729 kappa analysis Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
-
- 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
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a kind of hyperspectral image classification methods based on large-spacing distribution and space characteristics, the following steps are included: the high spectrum image of input is normalized first, obtain the hyperspectral image data collection of information content redistribution, then the high-spectral data collection after normalized will be subjected to PCA dimensionality reduction, data set after dimension-reduction treatment is filtered, and extracts spatial texture characteristic information；Two EO-1 hyperion linear space correlation information matrixes are defined for the high spectrum image to be processed, spatial texture characteristic information is merged with spatial coherence information, fusion spatial information data collection is trained and is classified using large-spacing distribution machine LDM.The present invention is merged by a kind of spatial coherence feature and textural characteristics, overcomes the problem of texture feature extraction filter is easily lost spatial coherence feature, while the classification of high spectrum image is realized with large-spacing distribution machine LDM, improves nicety of grading.
Description
Technical field
The present invention relates to technical field of image processing, more particularly, to one kind based on large-spacing distribution and space characteristics
Hyperspectral image classification method.
Background technique
Bloom spectrum sensor obtains the reflected radiation information of atural object by a spectrum channels up to a hundred, and wavelength band covers
From visible light to near-infrared or even LONG WAVE INFRARED region, high spectrum image contain spatial information, reflection or the spoke of atural object simultaneously
Information and spectral information are penetrated, feature is commonly known as " collection of illustrative plates ".And spectral image data provides nearly continuity
Spectrum sample information, can recorde the reflection differences of atural object spectrally very little.The diagnosis that this characteristic is referred to as atural object is special
Property, it can be used as the foundation classified to atural object and detected.Classification hyperspectral imagery new technology is studied, there is important theory
Meaning and application value.
The technology of current classification hyperspectral imagery is primarily present following problems: not obtaining sufficiently during hyperspectral classification empty
Between feature, be easily lost using filter texture feature extraction atural object spatial coherence information, do not consider spatial texture spy
Spatial coherence Fusion Features of seeking peace get up to constitute more complete space characteristics, in classification hyperspectral imagery, it is traditional most
The disaggregated model of large-spacing model optimization cannot represent the interval point of entire training dataset both for some single interval
Cloth, it is difficult to further increase nicety of grading.Therefore the image classification of the EO-1 hyperion based on large-spacing distribution and space characteristics is still
It is the direction for being worth research.
Summary of the invention
The present invention in order to overcome at least one of the drawbacks of the prior art described above, provides a kind of based on large-spacing distribution and empty
Between feature hyperspectral image classification method.
The present invention is directed to solve above-mentioned technical problem at least to a certain extent.
Primary and foremost purpose of the invention is to construct a kind of spatial coherence feature, is averaged with maximization space characteristics interval
Value minimizes the large-spacing distribution machine LDM of space characteristics interval Variance feature to realize the classification of high spectrum image, effectively simultaneously
Improve nicety of grading.
To realize the above goal of the invention, technical scheme is as follows:
A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics, includes the following steps:
S1: inputting high spectrum image to be processed, and hyperspectral image data collection is normalized, information content is obtained
The hyperspectral image data collection G of redistribution；
S2: the high-spectral data collection after normalized is subjected to PCA dimensionality reduction: H=Pca (G), selection selects the data
The data of the preceding n dimension of collection form new data set H；
S3: being filtered the data set H after dimension-reduction treatment with bilateral filtering method, and extracts spatial texture feature letter
Cease Dt；
S4: two EO-1 hyperion linear space correlation information matrix Ds are defined for the high spectrum image to be processedl, Dv, high
Spectrum picture pixel (x, y) linear space correlation information: Dc=Dl+DV；
S5: spatial texture characteristic information is merged with spatial coherence information: W=Dt+Dc, obtain fusion spatial information number
According to collection W；
S6: fusion spatial information data collection W is trained and is classified using large-spacing distribution machine LDM.
Further, normalized described in step S1, calculates according to following formula:
Wherein, R represents hyperspectral image data collection reflected intensity numerical value, RmaxIndicate that hyperspectral image data concentrates reflection
Maximum of intensity, RminIndicate that hyperspectral image data concentrates reflected intensity minimum value, G is the EO-1 hyperion of information content redistribution
Image data set.
Further, two EO-1 hyperion linear space correlation information matrix Ds described in step S4l, Dv, respectively indicate are as follows:
Wherein, DlThe linear space correlation information matrix of one horizontal direction, DvThe linear space phase of one vertical direction
Closing property information matrix Dv, (x, y) is pixel in the position of high spectrum image.
Further, training and classifying step described in step S5 include:
S5.1: training set W is randomly selected from spatial information data collection W ratio D% at randoms, remaining (1-D%) part conduct
Training set Wt；
S5.2: large-spacing Distributed learning machine LDM cross validation is used, optimal parameter combination is found；
S5.3: with large-spacing Distributed learning machine LDM to WsIt is trained, obtains training pattern；
S5.4: after obtaining training pattern, with large-spacing Distributed learning machine LDM to test set WtClassify.
Further, step S5.2, the cross validation include: method 1: svm classifier is used only；Method 2: PCA is utilized
After hyperspectral information dimensionality reduction, classification PCA-SVM is carried out with SVM；Method 3: Gabor filter, two-sided filter and guiding are used
Filter is respectively to preceding 20 Principle component extraction spatial information of the high-spectral data after PCA dimensionality reduction, and the space letter that will acquire
After breath and spectral information linearly combine, with svm classifier, GBF-SVM, tri- kinds of classification methods of BF-SVM, GDF-SVM are formed；Method
4:EPF algorithm classifies to high spectrum image, there is EPF-B-g and EPF-G-g；Method 5: LDM is used only and classifies；Method 6: benefit
With PCA to hyperspectral information dimensionality reduction after, carry out classification method: PCA-LDM with LDM；Method 7: with recursive filtering to high-spectrum
As classifying, classification LDM-FL then is realized with LDM；Method 8: after the fusion of spatial texture characteristic line, the bloom of SVM is incorporated
Compose image classification method PBF-SVM；Method 9: after texture characteristic line fusion, the hyperspectral image classification method of LDM is incorporated
PBF-LDM；Method 10: after spatial coherence feature and the fusion of spatial texture characteristic line, the classification hyperspectral imagery of SVM is incorporated
Method PBFC-SVM；Method 11: after spatial coherence feature and the fusion of spatial texture characteristic line, the high-spectrum of LDM is incorporated
As classification method PBFC-LDM.
Compared with prior art, the beneficial effect of technical solution of the present invention is: the present invention extracts bloom by bilateral filtering
Spatial coherence feature and spatial texture Fusion Features are maximized space characteristics with having by the spatial texture feature of spectrogram picture
Interval averages minimize the large-spacing distribution machine LDM of space characteristics interval Variance feature simultaneously to realize point of high spectrum image
Class improves nicety of grading.
Detailed description of the invention
Fig. 1 is algorithm flow chart.
Fig. 2 is Indian agricultural data images classification data statistical chart.
Fig. 3 is Indian agricultural classifying quality figure.
Fig. 4 is that Indian agricultural classifying quality counts histogram.
Fig. 5 is OA, AA and KAPPA line chart after the classification of Indian agricultural difference training sample ratio.
Fig. 6 is Salinas mountain valley data images classification data statistical chart.
Fig. 7 is Salinas mountain valley classifying quality figure.
Fig. 8 is that Salinas mountain valley classifying quality counts histogram.
Fig. 9 is OA, AA and KAPPA line chart after the classification of Salinas mountain valley difference training sample ratio.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent；
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
It is algorithm flow chart with reference to Fig. 1, Fig. 1.
A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics, includes the following steps:
S1: inputting high spectrum image to be processed, and hyperspectral image data collection is normalized, information content is obtained
The hyperspectral image data collection G of redistribution；
S2: the high-spectral data collection after normalized is subjected to PCA dimensionality reduction: H=Pca (G), selection selects the data
The data of the preceding n dimension of collection form new data set H；
S3: being filtered the data set H after dimension-reduction treatment with bilateral filtering method, and extracts spatial texture feature letter
Cease Dt；
S4: rate is M × N respectively for the high-spectrum image space to be processed, defines two EO-1 hyperion linear space correlations
Information matrix Dl, Dv, high spectrum image pixel (x, y) linear space correlation information: Dc=Dl+DV；
S5: spatial texture characteristic information is merged with spatial coherence information: W=Dt+Dc, obtain fusion spatial information number
According to collection W；
S6: fusion spatial information data collection W is trained and is classified using large-spacing distribution machine LDM.
Normalized described in step S1, calculates according to following formula:
Wherein, R represents hyperspectral image data collection reflected intensity numerical value, RmaxIndicate that hyperspectral image data concentrates reflection
Maximum of intensity, RminIndicate that hyperspectral image data concentrates reflected intensity minimum value, G is the EO-1 hyperion of information content redistribution
Image data set.
Two EO-1 hyperion linear space correlation information matrix Ds described in step S4l, Dv, respectively indicate are as follows:
Wherein, DlThe linear space correlation information matrix of one horizontal direction, DvThe linear space phase of one vertical direction
Closing property information matrix Dv, (x, y) is pixel in the position of high spectrum image.
Training described in step S5 and classifying step include:
S5.1: training set W is randomly selected from spatial information data collection W ratio D% at randoms, remaining (1-D%) part conduct
Training set Wt；
S5.2: large-spacing Distributed learning machine LDM cross validation is used, optimal parameter combination is found；
S5.3: with large-spacing Distributed learning machine LDM to WsIt is trained, obtains training pattern；
S5.4: after obtaining training pattern, with large-spacing Distributed learning machine LDM to test set WtClassify.
Step S5.2, the method that the cross validation uses includes: method 1: svm classifier is used only；Method 2: it utilizes
After PCA is to hyperspectral information dimensionality reduction, classification PCA-SVM is carried out with SVM；Method 3: it with Gabor filter, two-sided filter and leads
To filter respectively to preceding 20 Principle component extraction spatial information of the high-spectral data after PCA dimensionality reduction, and the space that will acquire
After information and spectral information linearly combine, with svm classifier, GBF-SVM, tri- kinds of classification methods of BF-SVM, GDF-SVM are formed；Side
Method 4:EPF algorithm classifies to high spectrum image, there is EPF-B-g and EPF-G-g；Method 5: LDM is used only and classifies；Method 6:
Using PCA to hyperspectral information dimensionality reduction after, carry out classification method: PCA-LDM with LDM；Method 7: with recursive filtering to EO-1 hyperion
Image is classified, and then realizes classification LDM-FL with LDM；Method 8: after the fusion of spatial texture characteristic line, the height of SVM is incorporated
Spectrum picture classification method PBF-SVM；Method 9: after texture characteristic line fusion, the classification hyperspectral imagery side of LDM is incorporated
Method PBF-LDM；Method 10: after spatial coherence feature and the fusion of spatial texture characteristic line, the high spectrum image point of SVM is incorporated
Class method PBFC-SVM；Method 11: after spatial coherence feature and the fusion of spatial texture characteristic line, the EO-1 hyperion of LDM is incorporated
Image classification method PBFC-LDM, i.e., classification method of the present invention.
In terms of data statistics, the present invention is using whole nicety of grading (Overall accuracy, OA), average classification essence
It spends (Average accuracy, AA) and Kappa counts coefficient (Kappa statistic, Kappa) Lai Hengliang sorting algorithm
Precision, in order to avoid random deviation, each experiment is repeated 10 times record average result, and verification platform uses Matlab
The experiment porch of R2012b, i7-6700CPU, 8GBRAM.
Present invention employs two groups of embodiments to be described in detail.
Embodiment 1
Indian agricultural high-spectrum remote sensing comes from spectrometer (Airborne Visible Infrared Imaging
Spectrometer), it is the high-spectrum remote sensing being collected into the Indian agricultural in the state of Indiana northwestward in 1992, has
20 meters of spatial resolution, it includes 144x144 pixel, 220 wave bands, since the factors such as noise and water absorption remove wherein
20 wave bands, remaining 200 wave bands include 16 kinds of atural objects, specifically species not and number of samples referring to fig. 2.
Indian agricultural data images, atural object be distributed as shown in (a) figure of Fig. 3, choose whole 16 classifications, every class with
Machine, which chooses 5% sample composition, label training set, remaining 95% is used as test set, and the three classes atural object 20% of atural object negligible amounts is made
For training set.
Figure (a) indicates atural object as shown in Figure 3；Scheming (b) indicates only SVM method, whole nicety of grading OA=78.43%；
Scheming (c) indicates PCA-SVM method, whole nicety of grading OA=77.99%；Scheming (d) indicates GBF-SVM method, whole classification essence
Spend OA=78.20%；Scheming (e) indicates BF-SVM method, whole nicety of grading OA=86.99%；Scheming (f) indicates the side GDF-SVM
Method, whole nicety of grading OA=88.56%；Scheming (g) indicates EPF-B-g method, whole nicety of grading OA=90.01%；Scheme (h)
Indicate EPF-G-g method, whole nicety of grading OA=90.84%；Scheming (i) indicates IFRF method, whole nicety of grading OA=
92.37%；Scheming (j) indicates LDM method, whole nicety of grading OA=80.45%；Scheming (k) indicates PCA-LDM method, whole to divide
Class precision OA=79.97%；Scheming (l) indicates LDM-FL method, whole nicety of grading OA=93.77%；Scheming (m) indicates PBF-
SVM method, whole nicety of grading OA=91.58%；Scheming (n) indicates PBF-LDM method, whole nicety of grading OA=95.3%；
Scheming (o) indicates PBFC-SVM method, whole nicety of grading OA=94.96%；Scheming (p) indicates PBFC-LDM, whole nicety of grading
OA=96.87%.
Classifying quality counts histogram, as shown in figure 4, from the point of view of the experiment of Indian agricultural data set, LDM and PCA-LDM's
OA is 80.45% and 79.97%, is higher by 2.02 and 1.98 percentage points, the OA of PBF-LDM and PBFC-LDM than SVM and PCA-SVM
3.73 and 1.91 percentage points are higher by than PBF-SVM and PBFC-SVM, LDM is demonstrated and is carried out by space characteristics based between maximization
Minimize the validity of interval variance simultaneously every average value.Classified with PBF-LDM to two kinds of data sets, wherein Indian agriculture
The OA of woods data set is 95.30%, is higher by 14.85 than LDM, demonstrates and is favorably assisted with the space characteristics that bilateral filtering extracts
LDM carries out classification hyperspectral imagery.Classified with PBFC-LDM to two kinds of data sets, wherein the OA of Indian agricultural data set
Be 96.87%, be higher by 1.56 respectively than PBF-LDM, demonstrate spatial autocorrelation information proposed by the present invention can effectively supplement it is double
The deficiency of side filtering.
In order to verify influence of the monitoring data to algorithm, the nicety of grading of different training sample testing algorithms is selected, such as
Shown in Fig. 5, wherein the digital label on curve is OA numerical value.Training of the Indian woods overall classification accuracy in training sample 2%
Sample proportion OA just reaches 92.10%, and 7% training sample OA has been more than 97%, demonstrates PBFC-LDM algorithm in low training
Also preferably nicety of grading can be obtained in the case where sample, and has certain stability.
Embodiment 2
Salinas mountain valley high-spectrum comes from spectrometer (Airborne Visible Infrared Imaging
It Spectrometer), is the image for adding the state Li Fuliya Salinas mountain valley to be collected into the U.S. in 1992, with 3.7 meters
Spatial resolution, it includes 512 × 217 pixels, 224 wave bands, since the factors such as noise and water absorption remove therein 20
A wave band, remaining 204 wave bands, include 16 kinds of atural objects, specific vegetation classification and number of samples are referring to Fig. 6.
Salinas mountain valley data images are chosen in all 16 kinds of vegetation classifications, and every class randomly selects 1% sample composition
There is label training set, remaining 99% is used as test set, and table 2 is various classification methods to the classification of Salinas mountain valley data set essence
Degree counts, and (a) indicates atural object in Fig. 7,；Scheming (b) indicates only SVM method, whole nicety of grading OA=85.93%；Scheming (c) indicates
PCA-SVM method, whole nicety of grading OA=85.72%；Scheming (d) indicates GBF-SVM method, whole nicety of grading OA=
83.94%；Scheming (e) indicates BF-SVM method, whole nicety of grading OA=87.34%；Scheming (f) indicates GDF-SVM method, whole
Nicety of grading OA=88.88%；Scheming (g) indicates EPF-B-g method, whole nicety of grading OA=88.91%；Scheming (h) indicates
EPF-G-g method, whole nicety of grading OA=89.59%；Scheming (i) indicates IFRF method, whole nicety of grading OA=
95.97%；Scheming (j) indicates LDM method, whole nicety of grading OA=87.59%；Scheming (k) indicates PCA-LDM method, whole to divide
Class precision OA=87.11%；Scheming (l) indicates LDM-FL method, whole nicety of grading OA=97.81%；Scheming (m) indicates PBF-
SVM method, whole nicety of grading OA=90.34%；Scheming (n) indicates PBF-LDM method, whole nicety of grading OA=96.02%；
Scheming (o) indicates PBFC-SVM method, whole nicety of grading OA=96.88%；Scheming (p) indicates PBFC-LDM, whole nicety of grading
OA=98.55%.
Classifying quality counts histogram, as shown in Figure 8.This mountain valley data set classification experiments of Surrey, LDM and PCA-LDM's
OA is 87.59% and 87.11%, is higher by 1.66% and 1.39% percentage point, PBF-LDM and PBFC-LDM than SVM and PCA-SVM
OA ratio PBF-SVM and PBFC-SVM be higher by 5.68 and 1.67 percentage points.LDM is demonstrated to carry out by space characteristics based on maximum
Change the validity that interval averages minimize interval variance simultaneously.
Classified with PBF-LDM to two kinds of data sets, the OA of this mountain valley data set of Surrey is 96.02%, is higher by than LDM
It 8.44 percentage points, demonstrates and favorably assists LDM to carry out classification hyperspectral imagery with the space characteristics that bilateral filtering extracts.
Classified with PBFC-LDM to two kinds of data sets, the OA of this mountain valley data set of Surrey is 98.28%, compares PBF-
LDM is higher by 2.53 percentage points respectively, and bilateral filtering can effectively be supplemented by demonstrating spatial autocorrelation information proposed by the present invention
It is insufficient.
In order to verify influence of the monitoring data to algorithm, the nicety of grading of different training sample testing algorithms is selected, such as
Shown in Fig. 9, wherein the digital label on curve is OA numerical value.This mountain valley of Surrey overall classification accuracy OA is in training sample
Just reach 93.42% and 98.28% when 0.2% and 0.5%, demonstrates PBFC-LDM algorithm in the case where low training sample
Also preferably nicety of grading can be obtained, and has certain stability.
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (5)
1. a kind of hyperspectral image classification method based on large-spacing distribution and space characteristics, which is characterized in that including walking as follows
It is rapid:
S1: inputting high spectrum image to be processed, and hyperspectral image data collection is normalized, and obtains information content again
The hyperspectral image data collection G of distribution；
S2: the high-spectral data collection after normalized is subjected to PCA dimensionality reduction: H=Pca (G), selection selects the data set
The data of preceding n dimension form new data set H；
S3: the data set H after dimension-reduction treatment is filtered with bilateral filtering method, and extracts spatial texture characteristic information Dt；
S4: two EO-1 hyperion linear space correlation information matrix Ds are defined for the high spectrum image to be processedl, Dv, EO-1 hyperion
Image picture elements (x, y) linear space correlation information: Dc=Dl+DV；
S5: spatial texture characteristic information is merged with spatial coherence information: W=Dt+Dc, obtain fusion spatial information data collection
W；
S6: fusion spatial information data collection W is trained and is classified using large-spacing distribution machine LDM.
2. a kind of hyperspectral image classification method based on large-spacing distribution and space characteristics according to claim 1,
It is characterized in that, normalized described in step S1 is calculated according to following formula:
Wherein, R represents hyperspectral image data collection reflected intensity numerical value, RmaxIndicate that hyperspectral image data concentrates reflected intensity
Maximum value, RminIndicate that hyperspectral image data concentrates reflected intensity minimum value, G is the high spectrum image of information content redistribution
Data set.
3. a kind of hyperspectral image classification method based on large-spacing distribution and space characteristics according to claim 1,
It is characterized in that, two EO-1 hyperion linear space correlation information matrix Ds described in step S4l, Dv, respectively indicate are as follows:
Wherein, DlThe linear space correlation information matrix of one horizontal direction, DvThe linear space correlation of one vertical direction
Information matrix Dv, (x, y) is pixel in the position of high spectrum image.
4. a kind of hyperspectral image classification method based on large-spacing distribution and space characteristics according to claim 1,
It is characterized in that, training described in step S5 and classifying step include:
S5.1: training set W is randomly selected from spatial information data collection W ratio D% at randoms, remaining (1-D%) is partially as training
Collect Wt；
S5.2: large-spacing Distributed learning machine LDM cross validation is used, optimal parameter combination is found；
S5.3: with large-spacing Distributed learning machine LDM to WsIt is trained, obtains training pattern；
S5.4: after obtaining training pattern, with large-spacing Distributed learning machine LDM to test set WtClassify.
5. a kind of hyperspectral image classification method based on large-spacing distribution and space characteristics according to claim 4,
It is characterized in that, step S5.2 cross validation application method includes method 1: svm classifier is used only；Method 2: using PCA to bloom
After spectrum information dimensionality reduction, classification PCA-SVM is carried out with SVM；Method 3: Gabor filter, two-sided filter and Steerable filter device are used
Respectively to preceding 20 Principle component extraction spatial information of the high-spectral data after PCA dimensionality reduction, and the spatial information and light that will acquire
After spectrum information linearly combines, with svm classifier, GBF-SVM, tri- kinds of classification methods of BF-SVM, GDF-SVM are formed；Method 4:EPF is calculated
Method classifies to high spectrum image, there is EPF-B-g and EPF-G-g；Method 5: LDM is used only and classifies；Method 6: PCA pairs is utilized
After hyperspectral information dimensionality reduction, classification method: PCA-LDM is carried out with LDM；Method 7: high spectrum image is divided with recursive filtering
Then class realizes classification LDM-FL with LDM；Method 8: after the fusion of spatial texture characteristic line, the high spectrum image point of SVM is incorporated
Class method PBF-SVM；Method 9: after texture characteristic line fusion, the hyperspectral image classification method PBF-LDM of LDM is incorporated；
Method 10: after spatial coherence feature and the fusion of spatial texture characteristic line, the hyperspectral image classification method of SVM is incorporated
PBFC-SVM；Method 11: after spatial coherence feature and the fusion of spatial texture characteristic line, the high spectrum image point of LDM is incorporated
Class method PBFC-LDM.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811238577.0A CN109598284A (en) | 2018-10-23 | 2018-10-23 | A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811238577.0A CN109598284A (en) | 2018-10-23 | 2018-10-23 | A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109598284A true CN109598284A (en) | 2019-04-09 |
Family
ID=65957973
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811238577.0A Pending CN109598284A (en) | 2018-10-23 | 2018-10-23 | A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109598284A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070137A (en) * | 2019-04-26 | 2019-07-30 | 广东交通职业技术学院 | A kind of hyperspectral image classification method based on adaptive manifold filtering and domain transfer standard convolutional filtering |
| CN110334731A (en) * | 2019-05-09 | 2019-10-15 | 云南大学 | A kind of the extraction of spatial information method, apparatus and electronic equipment of spectrum picture |
| CN110414438A (en) * | 2019-07-30 | 2019-11-05 | 西南交通大学 | High spectrum image recognition methods based on space clustering Information revision |
| CN110472682A (en) * | 2019-08-13 | 2019-11-19 | 辽宁工程技术大学 | A kind of Hyperspectral Remote Sensing Imagery Classification method for taking space and local feature into account |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104102705A (en) * | 2014-07-09 | 2014-10-15 | 南京大学 | Digital media object classification method based on large margin distributed learning |
| CN107066959A (en) * | 2017-03-29 | 2017-08-18 | 广东交通职业技术学院 | A kind of hyperspectral image classification method based on Steerable filter and linear space correlation information |
| WO2018045626A1 (en) * | 2016-09-07 | 2018-03-15 | 深圳大学 | Super-pixel level information fusion-based hyperspectral image classification method and system |
| CN108399355A (en) * | 2017-02-08 | 2018-08-14 | 广东交通职业技术学院 | A kind of hyperspectral image classification method that spatial information adaptively merges |
-
2018
- 2018-10-23 CN CN201811238577.0A patent/CN109598284A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104102705A (en) * | 2014-07-09 | 2014-10-15 | 南京大学 | Digital media object classification method based on large margin distributed learning |
| WO2018045626A1 (en) * | 2016-09-07 | 2018-03-15 | 深圳大学 | Super-pixel level information fusion-based hyperspectral image classification method and system |
| CN108399355A (en) * | 2017-02-08 | 2018-08-14 | 广东交通职业技术学院 | A kind of hyperspectral image classification method that spatial information adaptively merges |
| CN107066959A (en) * | 2017-03-29 | 2017-08-18 | 广东交通职业技术学院 | A kind of hyperspectral image classification method based on Steerable filter and linear space correlation information |
Non-Patent Citations (1)
| Title |
|---|
| ZHAN KUN等: "Large margin distribution machine for hyperspectral image classification", 《JOURNAL OF ELECTRONIC IMAGING》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070137A (en) * | 2019-04-26 | 2019-07-30 | 广东交通职业技术学院 | A kind of hyperspectral image classification method based on adaptive manifold filtering and domain transfer standard convolutional filtering |
| CN110334731A (en) * | 2019-05-09 | 2019-10-15 | 云南大学 | A kind of the extraction of spatial information method, apparatus and electronic equipment of spectrum picture |
| CN110334731B (en) * | 2019-05-09 | 2022-04-12 | 云南大学 | Spectral image spatial information extraction method and device and electronic equipment |
| CN110414438A (en) * | 2019-07-30 | 2019-11-05 | 西南交通大学 | High spectrum image recognition methods based on space clustering Information revision |
| CN110472682A (en) * | 2019-08-13 | 2019-11-19 | 辽宁工程技术大学 | A kind of Hyperspectral Remote Sensing Imagery Classification method for taking space and local feature into account |
| CN110472682B (en) * | 2019-08-13 | 2023-04-18 | 辽宁工程技术大学 | Hyperspectral remote sensing image classification method considering spatial and local characteristics |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pourreza et al. | An optimum method for real-time in-field detection of Huanglongbing disease using a vision sensor | |
| Okamoto et al. | Green citrus detection using hyperspectral imaging | |
| CN109598284A (en) | A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics | |
| Kurtulmus et al. | Green citrus detection using ‘eigenfruit’, color and circular Gabor texture features under natural outdoor conditions | |
| Coburn et al. | A multiscale texture analysis procedure for improved forest stand classification | |
| US10408740B2 (en) | Method for huanglongbing (HLB) detection | |
| Goodall et al. | Tasking on natural statistics of infrared images | |
| Syazwani et al. | Automated image identification, detection and fruit counting of top-view pineapple crown using machine learning | |
| CN109325431B (en) | Method and device for detecting vegetation coverage in feeding path of grassland grazing sheep | |
| Roscher et al. | Detection of disease symptoms on hyperspectral 3D plant models | |
| CN110376202B (en) | Tea tree anthracnose lesion identification method based on imaging hyperspectral technology | |
| Dubey et al. | Fruit disease recognition using improved sum and difference histogram from images | |
| CN111161362A (en) | Tea tree growth state spectral image identification method | |
| Ji et al. | In-field automatic detection of maize tassels using computer vision | |
| CN107871132A (en) | A kind of hyperspectral image classification method of the adaptive optimizing of space characteristics | |
| Ma et al. | Multiscale 2-D singular spectrum analysis and principal component analysis for spatial–spectral noise-robust feature extraction and classification of hyperspectral images | |
| Ekanayake et al. | A semi-supervised algorithm to map major vegetation zones using satellite hyperspectral data | |
| Liu et al. | An attention-guided and wavelet-constrained generative adversarial network for infrared and visible image fusion | |
| CN108764285A (en) | A kind of recognition methods of pine tree Deceased wood and system based on high resolution image | |
| CN108256557A (en) | The hyperspectral image classification method integrated with reference to deep learning and neighborhood | |
| CN105869161B (en) | Hyperspectral image band selection method based on image quality evaluation | |
| CN107527001A (en) | A kind of hyperspectral image classification method based on Steerable filter and linear space correlation information | |
| CN104239895B (en) | SAR target identification method based on feature dimension reduction | |
| Suzuki et al. | Image segmentation between crop and weed using hyperspectral imaging for weed detection in soybean field | |
| Hollaus et al. | MultiSpectral image binarization using GMMs |
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 |
Application publication date: 20190409 |
|
| RJ01 | Rejection of invention patent application after publication |