CN105550695A - Object-oriented single-class classification method for remote sensing image - Google Patents
Object-oriented single-class classification method for remote sensing image Download PDFInfo
- Publication number
- CN105550695A CN105550695A CN201510880838.9A CN201510880838A CN105550695A CN 105550695 A CN105550695 A CN 105550695A CN 201510880838 A CN201510880838 A CN 201510880838A CN 105550695 A CN105550695 A CN 105550695A
- Authority
- CN
- China
- Prior art keywords
- interest
- category
- class
- image
- remote sensing
- 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 37
- 238000012549 training Methods 0.000 claims abstract description 21
- 239000000284 extract Substances 0.000 claims description 11
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 6
- 238000005070 sampling Methods 0.000 abstract description 7
- 230000008569 process Effects 0.000 abstract description 6
- 239000000523 sample Substances 0.000 description 39
- 238000009826 distribution Methods 0.000 description 15
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 13
- 238000000605 extraction Methods 0.000 description 7
- 230000011218 segmentation Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000007812 deficiency Effects 0.000 description 3
- 230000003595 spectral effect Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012732 spatial analysis Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24317—Piecewise classification, i.e. whereby each classification requires several discriminant rules
Abstract
The invention discloses an object-oriented single-class classification method for a remote sensing image and relates to the technical field of remote sensing image classification. The method comprises following steps: (1) a class space is divided into an interest class and a non-interest class; (2) training samples of interest class and training samples of the non-interest class are selected according to the image space proximity; (3) partial training samples of interest class and partial training samples of the non-interest class are selected according to the image space proximity; (4) on the basis of samplings of the two classes, nearest-neighbor classifying is performed to extract an interest class. By use of the method, sampling selection process is simplified, and remote sensing image single-class information is effectively extracted.
Description
Technical field
What the present invention relates to is classification of remote-sensing images technical field, is specifically related to a kind of OO remote sensing image list class sorting technique.
Background technology
In recent years, multi-class classification method emerges in an endless stream, and utilizes the category of interest in one-class classifier extraction remote sensing image, and relevant research is also less.A conventional one-class classifier is one-class support vector machines (one-classsupport-vectormachine, OCSVM).OCSVM method, in higher dimensional space, finds the lineoid with largest interval that category of interest is separated by an energy, and its shortcoming is the selection difficulty of free parameter.In structural classification device process, except category of interest sample, unmarked sample also provides useful information, such as TransductiveSVM (TSVM) method, good classification performance can be obtained by utilizing unmarked sample, TSVM method needs to set iterations, and needs the positive sample in marker samples and negative sample two aspects.Also has the method (supportvectordatadescription described based on support region, SVDD), a spheroid little as far as possible comprising target data is utilized to differentiate, just can obtain good classifying quality by small sample training, the major defect of SVDD method is also optimum configurations more complicated.Lietal. propose PUL (positiveandunlabeledlearning) algorithm and carry out single class classification, test in high-resolution remote sensing image, be extracted each the single class in image respectively, comprise urban district, trees, meadow, water body, bare area.The MAXENT method of the remote sensing image list class classification Lietal. proposed, only require positive sample in the training process, experimental result shows that its single class classifying quality is better than OCSVM method.MAXENT method choice has the Data distribution8 of distribution form as category of interest of maximum entropy, and be a kind of parameterized method, for irregular category of interest distribution form, effect is affected.
Summary of the invention
For the deficiency that prior art exists, the present invention seeks to be to provide a kind of OO remote sensing image list class sorting technique, simplify samples selection process, effectively can realize the extraction of remote sensing image list category information.
To achieve these goals, the present invention realizes by the following technical solutions: OO remote sensing image list class sorting technique, comprises the following steps: classification spatial division is category of interest and non-category of interest two class by (1); (2) category of interest and non-category of interest training sample is selected according to the propinquity of image space; (3) the part training sample of category of interest and non-category of interest is selected according to the propinquity of feature space; (4) on the basis of two class samples, carry out arest neighbors classification, extract category of interest.
The present invention is basic operating unit with the image object obtained after Iamge Segmentation, instead of based on single pixel operation.The generation of image object has considered spectrum and the spatial information of neighborhood pixels, and neighborhood pixels is merged into homogeneous patch, overcomes the deficiency based on pixels approach.And, the image object obtained by Iamge Segmentation has certain attribute, not only contain spectral information, also comprise the additional information that texture, size, shape, compactness, context etc. extract from image, the foundation of classification is added when not increasing external information, thus improve the precision of classification, make classification results more close to the result of visual identification.After remote sensing image is split into object, the topological relation between object can also be set up, thus likely realize the spatial analysis in Geographic Information System.
Beneficial effect of the present invention: based on arest neighbors method, extracts the category of interest in image.Object-oriented method combines spatial information and the spectral information of image in the segmentation stage, adds classification foundation.On the other hand, object-oriented method is optimized former image data collection, reduces variance within clusters, adds inter-class separability, and one-class classifier nicety of grading is improved.According to the feature of arest neighbors method, be category of interest and non-category of interest by category division, and select the part training sample of non-category of interest according to image space and feature space propinquity, simplify assorting process.
Accompanying drawing explanation
The present invention is described in detail below in conjunction with the drawings and specific embodiments;
TM remote sensing image experimental data 1 figure that Fig. 1 (a) is embodiments of the invention 1;
TM remote sensing image experimental data 2 figure that Fig. 1 (b) is embodiments of the invention 1;
The image space samples selection figure that Fig. 2 (a) extracts for impermeable surface of the present invention;
The extraction result figure that Fig. 2 (b) extracts for impermeable surface of the present invention;
The image space samples selection figure that Fig. 3 (a) is Clean water withdraw of the present invention;
The extraction result figure that Fig. 3 (b) is Clean water withdraw of the present invention;
The image that Fig. 4 (a) is Clean water withdraw of the present invention and feature space samples selection;
The extraction result figure that Fig. 4 (b) is Clean water withdraw of the present invention.
Embodiment
The technological means realized for making the present invention, creation characteristic, reaching object and effect is easy to understand, below in conjunction with embodiment, setting forth the present invention further.
This embodiment is by the following technical solutions: OO remote sensing image list class sorting technique, comprises the following steps: classification spatial division is category of interest and non-category of interest two class by (1); (2) category of interest and non-category of interest training sample is selected according to the propinquity of image space; (3) the part training sample of category of interest and non-category of interest is selected according to the propinquity of feature space; (4) on the basis of two class samples, carry out arest neighbors classification, extract category of interest.
In the object-oriented classification of this embodiment, pixel average after segmentation in each imaged object is as data point, and the parameter that cutting procedure uses can control the mean size splitting rear imaged object, the similar sample totally extracting some in certain distribution, then calculates the distribution of sample average.Suppose certain data set (certain distribution overall), the expectation and variance of this distribution known, a part (m) data are extracted out from this is overall, form a sample, calculate a sample mean, have the selecting sample many times put back to like this, numerous average of samples will be produced, and these average of samples have the distribution form of oneself.When sampling fraction is very little, sampling without replacement is identical with the error of sampling with replacement substantially, the error calculation formula of sampling with replacement can be utilized replace without situation about putting back to, calculate so the data set distribution form after Image Segmentation can be similar.
To the overall X of Arbitrary distribution, expect that, for EX, variance is DX, puts back to selecting sample, capacity is m, if sample average is stochastic variable y, then
y＝(x
1+x
2+…+x
m)/m
Wherein, x
1, x
2... x
mfor an overall m sampling with replacement, so the expectation of y is
Due to Dy=Ey
2-(Ey)
2, ask Ey
2?
So
Can obtain thus, the expectation of the sample average of extraction with totally expect equal, variance is the 1/m of population variance.And statistical theory shows, no matter overall distribution, as long as sample size m enough large (being greater than 30), the distribution of sample average always trends towards normal distribution.In object-oriented classification, generally can ensure that the imaged object of actual segmentation is enough large, can comprise from tens to up to a hundred pixels, therefore, the distribution form of sample average data can well approximate normal distribution.Object-oriented method is classified to the data set that all image patches form, because expectation value of all categories is constant, variance within clusters is the 1/m of population variance, and Normal Distribution, therefore, the between class distance between each classification is constant, and variance within clusters reduces, thus classification separability increases, be conducive to the selection of threshold parameter in the classification of single class.
Extract in training sample selection at single category information, first the representative region of category of interest is selected, and to non-category of interest, then select the part sample training contiguous with category of interest, object finds the border of category of interest and non-category of interest, distinguishes category of interest and non-category of interest by border.By the spatial proximity effects of remote sensing image, mutually contiguous pixel characteristic is often close.Therefore, while selecting category of interest sample, in image space, select the non-category of interest sample contiguous with category of interest, and do not need to select completely non-category of interest representative point, make samples selection process become simple.In addition, in actual applications, the propinquity of image space and feature space being combined the selection carrying out training sample, when without the need to understanding all categories division in image, realizing effective extraction of classification interested.
In image space, select mutually contiguous category of interest and non-category of interest image object, improve learning efficiency and reliability.In feature space, because the image object splitting rear formation has property value, select the non-category of interest image object close with category of interest as training sample according to property value.Two kinds of methods do not consider the category division situation in image, and only carry out samples selection according to space and feature propinquity, the two combines the effective differentiation realizing category of interest and non-category of interest.
Embodiment 1: experimental image is two width TM remote sensing images, as shown in Figure 1, wherein Fig. 1 (a) comprises impermeable surface, meadow, forest land and water body four classifications, and Fig. 1 (b) also comprises four classifications: impermeable surface, water body, arable land and village.
Adopt arest neighbors method to extract single category information, around category of interest sample, the non-category of interest sample of selection, this approach reduces the workload of samples selection.Image is tested to the TM in Fig. 1 (a), take impervious surface as category of interest, the selection of training sample is as shown in Fig. 2 (a), and wherein white portion is category of interest sample, black region is around non-category of interest sample, and Fig. 2 (b) is classification results.The precision evaluation index that single category information extracts is generally production precision and user's precision, production precision refers to that the correct number of categories of category of interest accounts for the ratio of this classification pixel sum in reference data, and user's precision refers to that the correct number of categories of category of interest accounts for the ratio being divided into such pixel sum.The production precision that in experimental data 1, impervious surface list class is extracted and user's precision are respectively 85.3% and 90.9%.
And for some situation, category of interest close for feature and non-category of interest can not well be separated by the system of selection of this training sample.Such as in the experiment image shown in Fig. 1 (b), extract water body classification, the training sample that Fig. 3 (a) selects for water body classification and surrounding thereof, white portion is water body, and black region is other classifications.Fig. 3 (b) is classification results, can find out, because the feature of shade is similar to water body, some shadow region is divided into water body by mistake.
In order to avoid the mistake producing feature phase close-target based on the samples selection of image space is divided, image space and feature space are combined.Due in object-oriented method, after Iamge Segmentation, each object has respective attributive character value, thus can select the non-category of interest training sample similar to category of interest spectral signature.As shown in Fig. 4 (a), the shadow region below increase image is as non-category of interest sample, and Fig. 4 (b) is classification results.Can find out, the method that this feature space and image space samples selection combine can overcome the deficiency of single method, wrong subregion is obviously reduced, and single class of this Water In The Experiment body extracts production precision and user's precision is respectively 99.0% and 98.0%, good classification effect.
More than show and describe ultimate principle of the present invention and principal character and advantage of the present invention.The technician of the industry should understand; the present invention is not restricted to the described embodiments; what describe in above-described embodiment and instructions just illustrates principle of the present invention; without departing from the spirit and scope of the present invention; the present invention also has various changes and modifications, and these changes and improvements all fall in the claimed scope of the invention.Application claims protection domain is defined by appending claims and equivalent thereof.
Claims (1)
1. OO remote sensing image list class sorting technique, is characterized in that, comprise the following steps: classification spatial division is category of interest and non-category of interest two class by (1); (2) category of interest and non-category of interest training sample is selected according to the propinquity of image space; (3) the part training sample of category of interest and non-category of interest is selected according to the propinquity of feature space; (4) on the basis of two class samples, carry out arest neighbors classification, extract category of interest.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510880838.9A CN105550695A (en) | 2015-11-27 | 2015-11-27 | Object-oriented single-class classification method for remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510880838.9A CN105550695A (en) | 2015-11-27 | 2015-11-27 | Object-oriented single-class classification method for remote sensing image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105550695A true CN105550695A (en) | 2016-05-04 |
Family
ID=55829878
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510880838.9A Pending CN105550695A (en) | 2015-11-27 | 2015-11-27 | Object-oriented single-class classification method for remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105550695A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106485239A (en) * | 2016-11-02 | 2017-03-08 | 郑州航空工业管理学院 | One kind is using one-class support vector machines detection river mesh calibration method |
| CN106530170A (en) * | 2016-09-21 | 2017-03-22 | 环境保护部南京环境科学研究所 | Method for boundary optimization of regional ecological protection red line |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101430763A (en) * | 2008-11-10 | 2009-05-13 | 西安电子科技大学 | Detection method for on-water bridge target in remote sensing image |
| US20090265297A1 (en) * | 2008-04-21 | 2009-10-22 | Archan Misra | data reduction method to adaptively scale down bandwidth and computation for classification problems |
| CN103544502A (en) * | 2013-10-29 | 2014-01-29 | 上海市城市建设设计研究总院 | High-resolution remote-sensing image ship extraction method based on SVM |
| CN104504399A (en) * | 2015-01-05 | 2015-04-08 | 哈尔滨工业大学 | Multi-spectrum data supervised classification method based on relevant linear information entropy |
| CN104573049A (en) * | 2015-01-20 | 2015-04-29 | 安徽科力信息产业有限责任公司 | KNN classifier training sample cutting method based on center vector |
-
2015
- 2015-11-27 CN CN201510880838.9A patent/CN105550695A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090265297A1 (en) * | 2008-04-21 | 2009-10-22 | Archan Misra | data reduction method to adaptively scale down bandwidth and computation for classification problems |
| CN101430763A (en) * | 2008-11-10 | 2009-05-13 | 西安电子科技大学 | Detection method for on-water bridge target in remote sensing image |
| CN103544502A (en) * | 2013-10-29 | 2014-01-29 | 上海市城市建设设计研究总院 | High-resolution remote-sensing image ship extraction method based on SVM |
| CN104504399A (en) * | 2015-01-05 | 2015-04-08 | 哈尔滨工业大学 | Multi-spectrum data supervised classification method based on relevant linear information entropy |
| CN104573049A (en) * | 2015-01-20 | 2015-04-29 | 安徽科力信息产业有限责任公司 | KNN classifier training sample cutting method based on center vector |
Non-Patent Citations (3)
| Title |
|---|
| 刘家锋 等: "《面向对象遥感影像信息提取》", 31 August 2014, 哈尔滨：哈尔滨工业大学出版社 * |
| 邬小青 等: "改进最近邻算法及其在雷达目标识别中的应用", 《航空学报》 * |
| 陈金丽: "中国优秀硕士学位论文全文数据库 基础科学辑", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106530170A (en) * | 2016-09-21 | 2017-03-22 | 环境保护部南京环境科学研究所 | Method for boundary optimization of regional ecological protection red line |
| CN106485239A (en) * | 2016-11-02 | 2017-03-08 | 郑州航空工业管理学院 | One kind is using one-class support vector machines detection river mesh calibration method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wäldchen et al. | Plant species identification using computer vision techniques: A systematic literature review | |
| CN107229917B (en) | A kind of several remote sensing image general character well-marked target detection methods based on iteration cluster | |
| Zhao et al. | Learning mid-level filters for person re-identification | |
| CN107909039B (en) | High-resolution remote sensing image earth surface coverage classification method based on parallel algorithm | |
| DeGol et al. | Geometry-informed material recognition | |
| Herdiyeni et al. | Combination of morphological, local binary pattern variance and color moments features for indonesian medicinal plants identification | |
| CN110399909A (en) | A kind of hyperspectral image classification method based on label constraint elastic network(s) graph model | |
| CN107392968B (en) | The image significance detection method of Fusion of Color comparison diagram and Color-spatial distribution figure | |
| CN109657610A (en) | A kind of land use change survey detection method of high-resolution multi-source Remote Sensing Images | |
| CN104881865A (en) | Forest disease and pest monitoring and early warning method and system based on unmanned plane image analysis | |
| CN109784192B (en) | Hyperspectral image classification method based on superpixel feature extraction neural network algorithm | |
| CN104408467B (en) | Classification of Polarimetric SAR Image method based on pyramid sampling and support vector machine | |
| CN105138970A (en) | Spatial information-based polarization SAR image classification method | |
| CN104182765A (en) | Internet image driven automatic selection method of optimal view of three-dimensional model | |
| CN102254326A (en) | Image segmentation method by using nucleus transmission | |
| CN103984963B (en) | Method for classifying high-resolution remote sensing image scenes | |
| Sun et al. | Object-oriented land cover classification using HJ-1 remote sensing imagery | |
| CN105913090B (en) | SAR image objective classification method based on SDAE-SVM | |
| CN105976376A (en) | High resolution SAR image target detection method based on part model | |
| CN106096651B (en) | Polarization SAR terrain classification method based on super-pixel and metric learning | |
| CN109447111A (en) | A kind of remote sensing supervised classification method based on subclass training sample | |
| CN102542285B (en) | Image collection scene sorting method and image collection scene sorting device based on spectrogram analysis | |
| CN106650811A (en) | Hyperspectral mixed pixel classification method based on neighbor cooperation enhancement | |
| Jayanth et al. | Land-use/land-cover classification using elephant herding algorithm | |
| CN105550695A (en) | Object-oriented single-class classification method for remote sensing image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160504 |
|
| RJ01 | Rejection of invention patent application after publication |