CN107862330A - A kind of hyperspectral image classification method of combination Steerable filter and maximum probability - Google Patents
A kind of hyperspectral image classification method of combination Steerable filter and maximum probability Download PDFInfo
- Publication number
- CN107862330A CN107862330A CN201711044035.5A CN201711044035A CN107862330A CN 107862330 A CN107862330 A CN 107862330A CN 201711044035 A CN201711044035 A CN 201711044035A CN 107862330 A CN107862330 A CN 107862330A
- Authority
- CN
- China
- Prior art keywords
- image
- steerable filter
- calculate
- maximum probability
- filtering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
Abstract
The invention discloses a kind of combination Steerable filter and the hyperspectral image classification method of maximum probability, by high spectrum image is normalized and PCA dimension-reduction treatment after carry out Steerable filter, probability calculation is then carried out to the navigational figure after Steerable filter and draws classification results.Utilization space information extraction texture information of the present invention improves nicety of grading, is had a distinct increment with reference to classification performance after Steerable filter and maximum probability optimization.
Description
Technical field
The present invention relates to field of image recognition, more particularly, to a kind of combination Steerable filter and the bloom of maximum probability
Compose image classification method.
Background technology
Sky spectrum improves high spectrum image classification performance after combining is the previous study hotspot of mesh.Extraction of spatial information at present
Method has：1) shape filtering feature extraction；2) markov random file feature extraction；3) image segmentation feature extraction.Utilize filter
The spatial texture information of wave method extraction high spectrum image starts gradually to increase.
High spectrum image spatial texture information, which is extracted, achieves certain effect in the research for classification, but there is also one
A little deficiencies：1) spatial texture information needs further to excavate；2) spatial texture information needs more effectively to combine with spectral information
Subsidiary classification improves nicety of grading.
The content of the invention
It is a kind of present invention aim to address the defects of current existing space information extracting method nicety of grading deficiency, proposing
With reference to Steerable filter and the hyperspectral image classification method of maximum probability.
To realize above goal of the invention, the technical scheme of use is：
The hyperspectral image classification method of a kind of combination Steerable filter and maximum probability, comprises the following steps：
S1：Input high spectrum image 1；
S2：High spectrum image 1 is normalized；
S3：PCA dimensionality reductions are carried out to high spectrum image 1, obtain the 1st principal component；
S4：Steerable filter is carried out to N number of principal component before PCA, obtains spatial information F；
S5：Spectral information H in high spectrum image 1 and spatial information F is subjected to linear fusion and obtains spatial information data
Collect W, wherein W=H+F；
S6：Svm classifier is carried out to spatial information data collection W；
S7：Obtain preliminary classification matrix of consequence Lable；
S8：The compositions of PCA the 1st are inputted as filtering image；
S9：Obtain type of ground objects sum L；
S10：I-th of atural object distribution matrix is obtained as filtering image P；
S11：Filtering is guided to filtering image P；
S12：Obtain L Two-dimensional Probabilistic matrix；
S13：Maximum probability determines atural object label, draws classification results and the output of atural object.
Steerable filter in step S4 comprises the following steps：
S4.1：Calculate the average for being oriented to image I：Im=BF (I, r)；
S4.2：Calculate filtering image P average：Pm=BF (P, r)；
S4.3：Calculate the auto-correlation coefficient for being oriented to image I：corrI=BF (I.*I, r)；
S4.4：Calculate the cross-correlation coefficient for being oriented to image I and filtering image P：corrIp=BF (I.*P, r)；
S4.5：Calculate the variance for being oriented to image I：varI=corrI-Im.*Im；
S4.6：Calculate the covariance for being oriented to image I and filtering image P：covIp=corrIp-Im.*Pm；
S4.7：Design factor α：α=covIp./(varI+ε)；
S4.8：Design factor β：β=Pm-α*Im；
S4.9：Calculate Steerable filter：Fi=α * G+ β.
Step S6 comprises the following steps：
S6.1：Training set W is randomly selected with certain proportion from spatial information data collection W at randoms, remainder is as survey
Examination collection Wt；
S6.2：The SVM method cross validations supported with RBF, find optimal parameter；
S6.3：The SVM supported with RBF is to WsIt is trained, obtains training pattern；
S6.4：After obtaining model, the SVM supported with RBF is to test set WtClassified.
The step of Steerable filter described in step S11 with step S4 the step of Steerable filter it is identical.
Compared with prior art, the beneficial effects of the invention are as follows：
1) preferable EO-1 hyperion spatial texture information is extracted using Steerable filter, can effectively aids in spectral information to improve classification
Precision；
2) pixel for combining Steerable filter extraction EO-1 hyperion spatial texture and spectral information is effectively classified, and is combined and is led for the later stage
Preferable basis of classification is provided to filtering and maximum probability approach Optimum Classification result.
Brief description of the drawings
Fig. 1 is flow chart of the present invention；
Fig. 2 is Indian agricultural data images grouped data statistical chart；
Fig. 3 is Indian agricultural data set atural object distribution map；
Fig. 4 is university of Pavia data images classification grouped data statistical chart；
Fig. 5 is university of Pavia data set atural object distribution map；
Fig. 6 is the nicety of grading comparison diagram of Indian forest land thing；
Fig. 7 is the nicety of grading comparison diagram of university of Pavia atural object.
Embodiment
Accompanying drawing being given for example only property explanation, it is impossible to be interpreted as the limitation to this patent；
Below in conjunction with drawings and examples, the present invention is further elaborated.
Embodiment 1
The hyperspectral image classification method of a kind of combination Steerable filter and maximum probability, it refer to Fig. 1, including following step
Suddenly：
S1：Input high spectrum image 1；
S2：High spectrum image 1 is normalized；
S3：PCA dimensionality reductions are carried out to high spectrum image 1, obtain the 1st principal component；
S4：Steerable filter is carried out to N number of principal component before PCA, obtains spatial information F；
S5：Spectral information H in high spectrum image 1 and spatial information F is subjected to linear fusion and obtains spatial information data
Collect W, wherein W=H+F；
S6：Svm classifier is carried out to spatial information data collection W；
S7：Obtain preliminary classification matrix of consequence Lable；
S8：The compositions of PCA the 1st are inputted as filtering image；
S9：Obtain type of ground objects sum L；
S10：I-th of atural object distribution matrix is obtained as filtering image P；
S11：Filtering is guided to filtering image P；
S12：Obtain L Two-dimensional Probabilistic matrix；
S13：Maximum probability determines atural object label, draws classification results and the output of atural object.
Steerable filter in step S4 comprises the following steps：
S4.1：Calculate the average for being oriented to image I：Im=BF (I, r)；
S4.2：Calculate filtering image P average：Pm=BF (P, r)；
S4.3：Calculate the auto-correlation coefficient for being oriented to image I：corrI=BF (I.*I, r)；
S4.4：Calculate the cross-correlation coefficient for being oriented to image I and filtering image P：corrIp=BF (I.*P, r)；
S4.5：Calculate the variance for being oriented to image I：varI=corrI-Im.*Im；
S4.6：Calculate the covariance for being oriented to image I and filtering image P：covIp=corrIp-Im.*Pm；
S4.7：Design factor α：α=covIp./(varI+ε)；
S4.8：Design factor β：β=Pm-α*Im；
S4.9：Calculate Steerable filter：Fi=α * G+ β.
Step S6 comprises the following steps：
S6.1：Training set W is randomly selected with certain proportion from spatial information data collection W at randoms, remainder is as survey
Examination collection Wt；
S6.2：The SVM method cross validations supported with RBF, find optimal parameter；
S6.3：The SVM supported with RBF is to WsIt is trained, obtains training pattern；
S6.4：After obtaining model, the SVM supported with RBF is to test set WtClassified.
The step of Steerable filter described in step S11 with step S4 the step of Steerable filter it is identical.
SGD-RSVM-GD algorithms proposed by the present invention combine Indian agricultural data set and university of Pavia data set, and
Using confusion matrix (Confusion Matrix), overall nicety of grading (Overall accuracy, OA), Kappa departments of statistic
(Kappa statistic, Kappa) and average nicety of grading (Average accuracy, AA) are counted to weigh sorting algorithm
Precision analyzed：
Indian agricultural data set 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, it is wrapped
It is remaining 200 because the factor such as noise and water absorption removes 20 wave bands therein containing 144 × 144 pixels, 220 wave bands
Wave band, comprising 16 classifications, specifically species are not and number of samples is referring to Fig. 2；All 16 classifications are wherein chosen, it is random per class
Choosing 10% sample composition has label training set, the few atural object selection 20% of discrete quantities, remaining 90% and 80% conduct test
Collection
University of Pavia data set：University of Pavia data set comes from spectrometer (Reflective Optics System
Imaging Spectrometer), shoot in the high-spectrum remote sensing of university of Pavia, comprising 610 × 340 pixels, 115
Individual wave band, because the factors such as noise remove 12 wave bands therein, remaining 103 wave bands, include 9 classifications, specific atural object
Classification and number of samples are referring to Fig. 4；All 9 classifications are chosen, 5% sample composition is randomly selected per class label training set, its
Remaining and 95% be used as test set.
Analysis is drawn：
1) classifying quality proposed by the present invention is more excellent, wherein Indian agricultural data set and university of Pavia data set are overall
Nicety of grading OA is respectively 97.97, and its classification performance is all higher than the nicety of grading of other algorithms, demonstrates the present invention and proposes classification
The validity of algorithm.
2) from the point of view of Indian agricultural data set classification results, the terrain classification effect promoting more than quantity is obvious, but to individual
The classifying quality of the few atural object of other quantity equally might as well, as clover (Alfalfa), quantity be only 54, nicety of grading reaches
95.20, spiced salt phenomenon can be effectively removed as can be seen from Figure 3, and the part especially irised out with square frame becomes apparent；Tieed up from pa
From the point of view of sub- university's data set classifying quality, training sample is only 5%, and this algorithm can equally reach preferably classifying quality, such as
Shown in Fig. 5, this algorithm can equally remove spiced salt phenomenon, fully demonstrate the validity of this algorithm；
3) overall classification accuracy OA, average nicety of grading AA of all kinds of graders to two class data sets are picked from experiment
Fig. 6,7 are constructed with Kappa coefficients, in terms of the classification results of spatial texture information, BL-RSVM classification performance compares GD-RSVM
To be got well with GB-RSVM classification performance, illustrating the spatial texture information of two-sided filter extraction can effectively classify, but be composed from sky
With reference to experiment from the point of view of, SGD-RSVM classification performance is more excellent than SGB-RSVM and SBL-RSVM performance, illustrates Steerable filter
The spatial texture information of extraction preferably can aid in spectral information to be classified.
4) from Fig. 6,7 as can be seen that nicety of grading OA, average nicety of grading AA and Kappa coefficient, GB-RSVM, BL-
RSVM, GD-RSVM, SGB-RSVM, SBL-RSVM and SGD-RSVM are better than without spatial information information, demonstrate spatial information with
And empty spectrum combining information is not bound with Steerable filter than early stage and carried to the importance of classification, SGD-RSVM-GD proposed by the present invention
Sorting algorithm EPF-B-c and the EPF-G-c nicety of grading of spatial texture information is taken to want high 2 to 3 percentage points, to university of Pavia
Overall nicety of grading close to 99%, demonstrate the validity of SGD-RSVM-GD algorithms.
Obviously, the above embodiment of the present invention is only intended to clearly illustrate example of the present invention, and is not pair
The restriction of embodiments of the present invention.For those of ordinary skill in the field, may be used also on the basis of the above description
To make other changes in different forms.There is no necessity and possibility to exhaust all the enbodiments.It is all this
All any modification, equivalent and improvement made within the spirit and principle of invention etc., should be included in the claims in the present invention
Protection domain within.
Claims (4)
1. the hyperspectral image classification method of a kind of combination Steerable filter and maximum probability, it is characterised in that comprise the following steps：
S1：Input high spectrum image 1；
S2：High spectrum image 1 is normalized；
S3：PCA dimensionality reductions are carried out to high spectrum image 1, obtain the 1st principal component；
S4：Steerable filter is carried out to N number of principal component before PCA, obtains spatial information F；
S5：Spectral information H in high spectrum image 1 and spatial information F is subjected to linear fusion and obtains spatial information data collection W,
Wherein W=H+F；
S6：Svm classifier is carried out to spatial information data collection W；
S7：Obtain preliminary classification matrix of consequence Lable；
S8：The compositions of PCA the 1st are inputted as filtering image；
S9：Obtain type of ground objects sum L；
S10：I-th of atural object distribution matrix is obtained as filtering image P；
S11：Filtering is guided to filtering image P；
S12：Obtain L Two-dimensional Probabilistic matrix；
S13：Maximum probability determines atural object label, draws classification results and the output of atural object.
2. the hyperspectral image classification method of a kind of combination Steerable filter according to claim 1 and maximum probability, it is special
Sign is that the Steerable filter in step S4 comprises the following steps：
S4.1：Calculate the average for being oriented to image I：Im=BF (I, r)；
S4.2：Calculate filtering image P average：Pm=BF (P, r)；
S4.3：Calculate the auto-correlation coefficient for being oriented to image I：corrI=BF (I.*I, r)；
S4.4：Calculate the cross-correlation coefficient for being oriented to image I and filtering image P：corrIp=BF (I.*P, r)；
S4.5：Calculate the variance for being oriented to image I：varI=corrI-Im.*Im；
S4.6：Calculate the covariance for being oriented to image I and filtering image P：covIp=corrIp-Im.*Pm；
S4.7：Design factor α：α=covIp./(varI+ε)；
S4.8：Design factor β：β=Pm-α*Im；
S4.9：Calculate Steerable filter：Fi=α * G+ β.
3. the hyperspectral image classification method of a kind of combination Steerable filter according to claim 1 and maximum probability, it is special
Sign is that step S6 comprises the following steps：
S6.1：Training set W is randomly selected with certain proportion from spatial information data collection W at randoms, remainder is as test set
Wt；
S6.2：The SVM method cross validations supported with RBF, find optimal parameter；
S6.3：The SVM supported with RBF is to WsIt is trained, obtains training pattern；
S6.4：After obtaining model, the SVM supported with RBF is to test set WtClassified.
4. the hyperspectral image classification method of a kind of combination Steerable filter according to claim 1 and maximum probability, it is special
Sign is, the step of Steerable filter described in step S11 with step S4 the step of Steerable filter it is identical.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711044035.5A CN107862330A (en) | 2017-10-31 | 2017-10-31 | A kind of hyperspectral image classification method of combination Steerable filter and maximum probability |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711044035.5A CN107862330A (en) | 2017-10-31 | 2017-10-31 | A kind of hyperspectral image classification method of combination Steerable filter and maximum probability |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107862330A true CN107862330A (en) | 2018-03-30 |
Family
ID=61696973
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711044035.5A Pending CN107862330A (en) | 2017-10-31 | 2017-10-31 | A kind of hyperspectral image classification method of combination Steerable filter and maximum probability |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107862330A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150055820A1 (en) * | 2013-08-22 | 2015-02-26 | Ut-Battelle, Llc | Model for mapping settlements |
| CN104683767A (en) * | 2015-02-10 | 2015-06-03 | 浙江宇视科技有限公司 | Fog penetrating image generation method and device |
| CN106503739A (en) * | 2016-10-31 | 2017-03-15 | 中国地质大学（武汉） | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics |
| 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 |
| CN107527001A (en) * | 2017-03-29 | 2017-12-29 | 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 |
-
2017
- 2017-10-31 CN CN201711044035.5A patent/CN107862330A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150055820A1 (en) * | 2013-08-22 | 2015-02-26 | Ut-Battelle, Llc | Model for mapping settlements |
| CN104683767A (en) * | 2015-02-10 | 2015-06-03 | 浙江宇视科技有限公司 | Fog penetrating image generation method and device |
| CN106503739A (en) * | 2016-10-31 | 2017-03-15 | 中国地质大学（武汉） | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics |
| 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 |
| CN107527001A (en) * | 2017-03-29 | 2017-12-29 | 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 (3)
| Title |
|---|
| 地质矿产部地质辞典办公室: "《地质大辞典 5 地质普查勘探技术方法分册 上》", 30 June 2005, 北京：地质出版社 * |
| 康旭东: "高光谱遥感影像空谱特征提取与分类方法研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 龚文引 等: "《智能算法在高光谱遥感数据处理中的应用》", 30 November 2014, 武汉：中国地质大学出版社 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lopatin et al. | Mapping plant species in mixed grassland communities using close range imaging spectroscopy | |
| CN106339674B (en) | The Hyperspectral Image Classification method that model is cut with figure is kept based on edge | |
| CN106503739A (en) | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics | |
| Nery et al. | Comparing supervised algorithms in Land Use and Land Cover classification of a Landsat time-series | |
| Wang et al. | Mobile smart device-based vegetable disease and insect pest recognition method | |
| CN108399355B (en) | Hyperspectral image classification method based on spatial information adaptive fusion | |
| CN109187552B (en) | Wheat scab damage grade determination method based on cloud model | |
| CN106682675B (en) | A kind of empty spectrum union feature extracting method towards high spectrum image | |
| CN107871132B (en) | Hyperspectral image classification method for spatial feature adaptive optimization | |
| Zhang et al. | Extraction of tree crowns damaged by Dendrolimus tabulaeformis Tsai et Liu via spectral-spatial classification using UAV-based hyperspectral images | |
| CN105512622B (en) | A kind of visible remote sensing image sea land dividing method based on figure segmentation and supervised learning | |
| Imani | Anomaly detection using morphology-based collaborative representation in hyperspectral imagery | |
| CN109977899B (en) | Training, reasoning and new variety adding method and system for article identification | |
| CN103793709A (en) | Cell recognition method and device, and urine analyzer | |
| Xiao et al. | Citrus greening disease recognition algorithm based on classification network using TRL-GAN | |
| CN109598284A (en) | A kind of hyperspectral image classification method based on large-spacing distribution and space characteristics | |
| Roth et al. | Automated detection of individual clove trees for yield quantification in northeastern Madagascar based on multi-spectral satellite data | |
| da Silva et al. | Modeling and detection of invasive trees using UAV image and machine learning in a subtropical forest in Brazil | |
| Biswas et al. | Delineation of tree patches in a mangrove-marsh transition zone by watershed segmentation of aerial photographs | |
| Bejnordi et al. | Novel chromatin texture features for the classification of pap smears | |
| CN107832693A (en) | A kind of high spectrum image vegetation classification method based on spatial autocorrelation information | |
| Liu et al. | DS-MENet for the classification of citrus disease | |
| Kent et al. | Early symptom detection of basal stem rot disease in oil palm trees using a deep learning approach on UAV images | |
| Alli et al. | Animal identification based on footprint recognition | |
| Desai et al. | A Novel technique for detecting crop diseases with efficient feature extraction |
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: 20180330 |
|
| RJ01 | Rejection of invention patent application after publication |