CN104881677B - Method is determined for the optimum segmentation yardstick of remote sensing image ground mulching classification - Google Patents
Method is determined for the optimum segmentation yardstick of remote sensing image ground mulching classification Download PDFInfo
- Publication number
- CN104881677B CN104881677B CN201510232524.8A CN201510232524A CN104881677B CN 104881677 B CN104881677 B CN 104881677B CN 201510232524 A CN201510232524 A CN 201510232524A CN 104881677 B CN104881677 B CN 104881677B
- Authority
- CN
- China
- Prior art keywords
- segmentation
- yardstick
- classification
- 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.)
- Expired - Fee Related
Links
Abstract
The invention provides a kind of optimum segmentation yardstick for the classification of remote sensing image ground mulching to determine method.Methods described includes the multi-scale division and classification, the choice of optimal scale step based on entropy information of remote sensing image.This method is based on pixel and two kinds of sorting techniques of object-oriented by fusion, and makes full use of sample information to carry out classification of remote-sensing images.This method effectively overcomes the problem of traditional pixel-oriented method produces a large amount of ' pepper salt ' noises, while realizes automatically selecting for object optimal scale, and a kind of effective method is provided for the drawing of ground mulching.
Description
Technical field
The present invention relates to the processing method of remote sensing image, especially for the optimum segmentation of remote sensing image ground mulching classification
Yardstick determines method, belongs to image processing field.
Background technology
The synthesis for all key elements of earth's surface that land cover pattern refers to nature Inaedificatio and artificial structure is covered, including earth's surface are planted
Quilt, soil, lake, Marsh Wetland and various buildings (such as road, house).Land cover pattern is the important of global environmental change
Forcing factors, the concern of more and more researchers was received in the last few years.
With the development of remote sensing science and technology, the resolution ratio more and more higher of remote sensing image, this is a variety of space scale Shangdis
The drawing of table covering provides feasibility.The classification of remote sensing image is the important step in ground mulching drawing, determines earth's surface
Cover the quality of drawing.At present, the method for carrying out classification of remote-sensing images is broadly divided into two major classes:(1) method of pixel-oriented,
(2) Object--oriented method.
Pixel is the elementary cell of remote sensing image, and the classification that remote sensing image is carried out using the statistical information of pixel is most simple
Clean, effective method.However, in the remote sensing image of middle and high yardstick, floor area is smaller corresponding to single pixel.Therefore by
The influence of ground complexity, substantial amounts of ' pepper salt ' noise can be produced by carrying out classification of remote-sensing images using the method based on pixel.
This greatly reduces the precision of classification of remote-sensing images.
In order to overcome based on ' pepper salt ' noise caused by the method for pixel, a kind of combination atural object texture and spatial information
Sorting technique is quietly risen --- the classification of object-oriented.Different from the method based on pixel, Object--oriented method first will
The standalone object that Remote Sensing Image Segmentation is spectrum homogeneous, space is continuous, then carries out classification processing by object.Can so have
Removal ' pepper salt ' noise of effect, improve nicety of grading.However, the size for the object that Image Segmentation obtains depends on point of image
Cut scale parameter.Larger segmentation yardstick can cause atural object by low segmentation, on the contrary, less segmentation yardstick can cause atural object by mistake
Segmentation.Obviously turn out, the over-segmentation of image and it is low split can all cause the reduction of nicety of grading, referring to Liu D, Xia
F.Assessing object-based classification:advantages and limitations[J].Remote
Sensing Letters,2010,1(4):187-194..Unfortunately, the determination of optimal scale is generally required to different points
Cut yardstick to be tested, and judge optimum segmentation yardstick by experience.This not only needs to expend substantial amounts of manpower, and is difficult to protect
Demonstrate,prove the accuracy of the optimal scale obtained.
In the last few years, numerous researchers were devoted to the select permeability for solving Image Segmentation optimal scale.Lucian Drǎ
The it is proposeds such as gut (2009) automatically determine the optimum segmentation yardstick of image using local variance, referring toL,Tiede D,
Levick S R.ESP:a tool to estimate scale parameter formultiresolution image
segmentation of remotely sensed data[J].InternationalJournal of Geographical
Information Science,2010,24(6):859-871. this method for object all on image using it is same most
Excellent yardstick.But in most cases, the object size in a width image is different, by same segmentation scale dimension applications in different big
Small object is split clearly irrational.In addition, T.Esch (2008) etc., which proposes one kind, automatically determines different objects
The method of respective optimum segmentation yardstick, referring to Esch T, Thiel M, Bock M, et al.Improvement of image
segmentation accuracy based on multiscale optimization procedure[J]
.Geoscience and Remote Sensing Letters,IEEE,2008,5(3):463-467. still substantial amounts of parameter
It is used for the determination of optimal scale, this causes whole algorithm to become suitable complexity.Meanwhile for different application purposes and
Speech, even if there is also different optimum segmentation yardsticks with piece image.For example needed when the object that house is extracted as needs
Want a relatively large yardstick, but when using automobile as needing a relatively small yardstick when needing extracting object.Cause
This, there is an urgent need to a kind of method that can automatically determine different object optimum segmentation yardsticks in image for we.
The content of the invention
For this, the invention provides the optimum segmentation yardstick classified for remote sensing image ground mulching to determine method, this method
The problem of being formerly mentioned can be reduced or avoided.
To solve the above problems, the optimum segmentation yardstick provided by the invention for the classification of remote sensing image ground mulching determines
Method, it comprises the following steps:
Step A, the multi-scale division of remote sensing image and classification;
The segmentation software that the step utilization can generate multi-scale division result obtains multiple dimensioned segmentation result, and merges
The sample information of Pixel-level calculates the averaged spectrum of each imaged object after segmentation, and averaged spectrum is classified, and obtains not
With the classification results and posterior probability vector of each object under segmentation yardstick;
Step B, the choice of optimal scale based on entropy information;
The multiple dimensioned posterior probability vector that the step obtains according to step A, each object is incremented by by segmentation yardstick
Order calculate the entropy of posterior probability respectively, the calculation formula of entropy is as follows:
Wherein PiRepresent that object belongs to the probability of the i-th class, n represents classification number, selects the minimum segmentation yardstick of entropy to make
For the optimum segmentation yardstick of the object, and the classification results of object under optimum segmentation yardstick are made as the final classification of object.
Further, the concrete methods of realizing of the step A is:
Carry out the segmentation of multiple yardsticks to remote sensing image first, the selection of multi-split yardstick is according to the DN values of image or reflects
The scope of rate determined, during segmentation, form factor and degree of the compacting factor are set according to characteristics of image;
Secondly, rule set is created in segmentation software according to the segmentation range scale of selection, and according to yardstick from big to small
Or order from small to large is split step by step;
Then, the object of multi-scale division result is formed into numbering file, exports as Raster Images successively, in Raster Images
Value corresponding to each pixel is exactly the numbering of place object;
Subsequently, according to the numbering file of object after the Image Segmentation of acquisition, and original multiband image, count respectively
The averaged spectrum of each object under each segmentation yardstick is calculated, obtains multiple dimensioned average spectral data;
Finally, training sample is chosen from original multiband image to be averaged to the object under different scale obtained above
Spectrum image is classified respectively, and the requirement of the training sample of selection is typical, randomness;
While classification, the posterior probability vector of the object under multiple yardsticks is obtained, on a certain segmentation yardstick
For one object, its posterior probability vector represents as follows:
P=(P1, P2..., Pi..., Pn)
Wherein PiRepresent that object belongs to the probability of the i-th class, n represents classification number.
Wherein preferably, the form factor is arranged to 0.2, degree of the compacting factor is arranged to 0.5.
Wherein preferably, the selection of segmentation yardstick, is the increase according to yardstick, the principle that yardstick interval also increases therewith is entered
OK.
The inventive method is based on two methods of pixel and object-oriented by fusion, and it is distant to make full use of sample information to carry out
Feel image classification.This method effectively overcomes the problem of traditional pixel-oriented method produces a large amount of ' pepper salt ' noises, simultaneously
Automatically selecting for object optimal scale is realized, a kind of effective method is provided for the drawing of ground mulching.
Other features and advantages of the present invention will illustrate in the following description, also, partial become from specification
Obtain it is clear that or being understood by implementing the present invention.The purpose of the present invention and other advantages can be by the explanations write
Specifically noted structure is realized and obtained in book, claims and accompanying drawing.
Brief description of the drawings
The following drawings is only intended to, in doing schematic illustration and explanation to the present invention, not delimit the scope of the invention.Wherein,
Fig. 1 is the optimum segmentation chi classified for remote sensing image ground mulching according to the specific embodiment of the present invention
Spend the schematic flow sheet of selection algorithm;
Fig. 2 is the principle schematic of method provided by the invention;
Fig. 3 is to determine that method is entered according to the optimum segmentation yardstick provided by the invention classified for remote sensing image ground mulching
Row classification, and the totality using the result for being classified to obtain based on pixels approach, different segmentation yardstick object-oriented methods
Precision statisticses figure;
Fig. 4 is to determine that method is entered according to the optimum segmentation yardstick provided by the invention classified for remote sensing image ground mulching
Row classification, and the result for being classified to obtain using the method based on pixel, optimal single scale Object--oriented method are total
Body precision changes statistical chart with sample size.
Embodiment
In order to which technical characteristic, purpose and the effect of the present invention is more clearly understood, now illustrate that the present invention's is specific
Embodiment.But it will be appreciated by those skilled in the art that following examples are not to the unique of technical solution of the present invention work
Limit, every any equivalents done under technical solution of the present invention Spirit Essence or change, be regarded as belonging to this hair
Bright protection domain.
Fig. 1 is the optimum segmentation chi classified for remote sensing image ground mulching according to the specific embodiment of the present invention
Spend the schematic flow sheet of determination method;Shown in reference picture 1, the following detailed description of according to provided by the invention for remote sensing image
The optimum segmentation yardstick of table cover classification determines the principle of method, and methods described includes following two big steps:
Step A, the multi-scale division of remote sensing image and classification;
Step B, the choice of optimal scale based on entropy information.
For remote sensing image, before processing procedure of the present invention is carried out, it is necessary first to Atmospheric Correction is carried out to image,
Remove atmospheric effect.For Hyperspectral imaging, in order to reduce amount of calculation, it is proposed that first with principal component point
Analyse (PCA) and dimension-reduction treatment is carried out to Hyperspectral imaging.Principal component analysis (PCA) is known method.
Step A and step B are discussed in detail separately below:
Step A, remote sensing image multiple dimensioned segmentation and classification
The present invention has been merged based on two methods of pixel and object-oriented, for Object--oriented method, is needed first
Dividing processing is carried out to remote sensing image.Therefore, the present embodiment is entered to remote sensing image first using the softwares of eCognition 8.9
The segmentation of the multiple yardsticks of row, actually eCognition all versions can, if other partitioning algorithm softwares can give birth to
Into the segmentation result of multiple yardsticks, this method is equally applicable to.Empirically determined, during segmentation, form factor is arranged to 0.2, compacts
It is preferable that the degree factor is arranged to 0.5.The selection of multi-stage division yardstick can be according to the DN values of image or the scope determination of reflectivity, this
Example selection multi-stage division yardstick be 100,120,140,160,180,200,240,280,320,360,400,460,520,
580th, 640,700,780,860,940,1020,1100 these.It can be seen that yardstick interval increases with the increase of yardstick
Greatly, because with the increase of yardstick, the area of object constantly increases, and the heterogeneous sex differernce between object is also increasing, less
Yardstick interval will not change segmentation result substantially, therefore this example have selected the segmentation interval scale constantly increased.
Secondly, MRS is utilized in the softwares of eCognition 8.9 according to the segmentation range scale of selection
(multiresolution segmentation) algorithm creates rule set, and (establishment of rule set is the base of eCognition softwares
This function, refers specifically to the setting of different partitioning parameters (form factor, degree of the compacting factor), and according to yardstick from big to small or from
Small order is arrived greatly step by step to split.
Then, the object of multi-scale division result is formed into numbering file, exports as Raster Images, therefore grid shadow successively
The numbering of object where value corresponding to each pixel is exactly as in.Formed numbering file process be eCognition softwares in itself
Function, and known method.
Subsequently, according to the numbering file of object after the Image Segmentation of above-mentioned steps acquisition, and original multiband shadow
Picture, the averaged spectrum of each object under each segmentation yardstick is calculated respectively, obtains multiple dimensioned average spectral data.Average light
The calculating of spectrum is exactly that the spectrum of all pixels inside object is averaged, and this is known process.
Finally, training sample is chosen from original multiband image to be averaged to the object under different scale obtained above
Spectrum image is classified respectively.The selection of training sample require the sample of selection be typical, randomness can.Train sample
This selection all exists in all supervised classifications, and a known process.
Maximum likelihood classifier (SVM) or support vector machine classifier (MLC) may be used to above-mentioned assorting process.It is right
In Hyperspectral imaging, it is proposed that using SVM classifier, preferably differentiated because SVM classifier has to high-spectral data
Ability;For the intermediate-resolution image such as Landsat, it is proposed that using MLC graders, because MLC has classification faster
Speed, be advantageous to improve whole efficiency.
In the posterior probability vector that can obtain object simultaneously classified using above-mentioned grader, (object is for difference
The ownership probability of class categories).For an object of a certain segmentation scale layer, its posterior probability vector can represent such as
Under:
P=(P1, P2..., Pi..., Pn)
Wherein PiRepresent that object belongs to the probability of the i-th class, n represents classification number.To all objects of different segmentation yardsticks
Above-mentioned sorting technique is used respectively, you can obtains multiple dimensioned classification results and multiple dimensioned posterior probability vector.
Step B, the choice of optimal scale based on entropy information
It can be obtained according to the posterior probability vector that flow described in step A is tried to achieve:(split yardstick when object is in over-segmentation
When smaller), the pixel included in object is less, is influenceed by ' pepper salt ' noise in object, and object is not true in classification
Qualitative increase.When object is in low segmentation (when segmentation yardstick is larger), the pixel included in object is more.Although ' pepper
Salt ' influence that brings of noise can be ignored, but inside object the mixing of different atural objects equally increase object classify it is not true
It is qualitative.
Based on above thought, the present invention proposes by the use of entropy information and is used as the standard for weighing object classification inaccuracy.Root
The multilayer posterior probability tried to achieve according to step A, ask for the entropy of object posterior probability vector respectively on different segmentation yardsticks.Entropy
Calculation formula is with reference to as follows:
Wherein PiRepresent that object belongs to the probability of the i-th class, n represents classification number.When object is in over-segmentation state,
The uncertain increase that can cause entropy of classification caused by ' pepper salt ' noise;Equally, when object is in low segmentation, a variety of atural objects
Mixing caused by the uncertain increase for also resulting in entropy of classification.Its rule is as shown in Figure 2.It can thus be concluded that:When object
When the posterior probability vector of a certain segmentation yardstick has minimum entropy, its stability of classifying is highest.Therefore selection entropy
Optimum segmentation yardstick of the minimum segmentation yardstick as object where the pixel.Made simultaneously with the object classification results under the yardstick
For the final classification result of object.As described above, above procedure is implemented to all objects, you can obtain the most optimal sorting of all objects
Cut yardstick and final classification results.
In order to better illustrate the technique effect of the present invention, for a panel height spectrum image, proposition of the present invention is utilized respectively
The optimum segmentation yardstick for the classification of remote sensing image ground mulching determine method and single based on pixels approach and optimal
Single scale object-oriented method has carried out the classification of image, and then sorted result is compared.All assorting processes are adopted
Completed with SVM classifier, precision highest chi in the classification results of the multiple yardsticks of object oriented classification selection of optimal single scale
Degree.The classification based training sample and the classification based training sample of method provided by the invention that above-mentioned control methods is selected are completely the same.Together
When, in order to verify the stability of this method, the present invention has carried out test of many times under different sample sizes.Population sample size with
1000 increase to 10000 pixels for interval from 1000 pixels.Ten experiments are carried out to each sample size simultaneously, every time experiment
Sample from reference to randomly selecting in classification chart.It is first before the processing of method provided by the invention is carried out to reduce amount of calculation
First with Principal Component Analysis Algorithm (PCA) by image dimension-reduction treatment, and 8 wave bands are used for method provided by the invention before extraction
Processing procedure.
Compared to the method based on pixel, it is directed to what remote sensing image ground mulching was classified by using according to provided by the invention
Optimum segmentation scale selection method carries out the result that the classification of remote sensing image obtains, and effectively eliminates ' pepper salt ' noise and brings
Error in classification;Compared with the classification results of optimal single scale object-oriented, method provided by the invention, it can extract more
Detailed information, improve the precision of classification.
Fig. 3 is to determine that method is entered using the optimum segmentation yardstick provided by the invention for the classification of remote sensing image ground mulching
Row classification is classified with the method based on pixel, and is classified what is obtained to Hyperspectral imaging using different segmentation yardsticks
The statistical chart of classification overall accuracy, it can be seen that method provided by the invention, compared to the method and list of pixel-oriented
Yardstick object oriented classification all has higher precision.
Fig. 4 is to determine that method is entered using the optimum segmentation yardstick provided by the invention for the classification of remote sensing image ground mulching
The experiment of the different sample sizes of row, from the experiment using the method based on pixel and the different sample sizes of optimal single scale method progress
Obtained overall classification accuracy statistical chart.From fig. 4, it can be seen that compared with the method based on pixel, the method for this research offer
It is overall all higher than the method based on pixel under conditions of different sample sizes;Compared with optimal single scale, it is less than in sample size
In the case of 4000, method overall accuracy proposed by the present invention is slightly below the method for optimal single scale, but when sample size is more than
When 4000, the method overall accuracy originally researched and proposed is higher than optimal single scale method.This explanation, in the condition of sample size deficiency
Under, the method robustness of offer of the invention reduces;Under conditions of sample size abundance, the method originally researched and proposed is to carry out shadow
As the effective ways of classification.
It can be seen that by above-mentioned figure example, according to the optimum segmentation provided by the invention classified for remote sensing image ground mulching
Yardstick determines effectively carry out the classification of remote sensing image.It is provided by the present invention to be directed to what remote sensing image ground mulching was classified
Optimum segmentation yardstick determines that method has been effectively combined based on pixel and two kinds of sorting techniques of object-oriented, effectively overcomes face
The problem of a large amount of ' pepper salt ' noises being produced to pixel, while realize the automatic selection of object optimal scale.For ground mulching system
Figure provides a kind of effective method.
It will be appreciated by those skilled in the art that although the present invention is described in the way of multiple embodiments, together
The Shi Jinhang across comparisons of several method, but not each embodiment only includes an independent technical scheme.Specification
In so narration just for the sake of for the sake of clear, the skilled in the art should refer to the specification as a whole is understood,
And it technical scheme involved in each embodiment is regarded as can be mutually combined into the modes of different embodiments to understand this hair
Bright protection domain.
Claims (4)
1. a kind of optimum segmentation yardstick for the classification of remote sensing image ground mulching determines method, it is characterised in that methods described
Comprise the following steps:
Step A, the multi-scale division of remote sensing image and classification;
The segmentation software that the step utilization can generate multi-scale division result obtains multiple dimensioned segmentation result, and merges pixel
The sample information of level calculates the averaged spectrum of each imaged object after segmentation, and averaged spectrum is classified, and obtains different points
Cut the classification results and posterior probability vector of each object under yardstick;
Step B, the choice of optimal scale based on entropy information;
The multiple dimensioned posterior probability vector that the step obtains according to step A, to each object by incremental suitable of segmentation yardstick
Sequence calculates the entropy of posterior probability respectively, and the calculation formula of entropy is as follows:
<mrow>
<mi>E</mi>
<mo>=</mo>
<mo>-</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>P</mi>
<mi>i</mi>
</msub>
<mo>&times;</mo>
<msub>
<mi>log</mi>
<mn>2</mn>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>P</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
Wherein PiRepresent that object belongs to the probability of the i-th class, n represents classification number, selects the minimum segmentation yardstick of entropy right as this
The optimum segmentation yardstick of elephant, and the classification results of object under optimum segmentation yardstick are made as the final classification of object;
Wherein, for an object on a certain segmentation yardstick, its posterior probability vector represents as follows:
P=(P1, P2..., Pi..., Pn)
Wherein PiRepresent that object belongs to the probability of the i-th class, n represents classification number.
2. according to the method for claim 1, it is characterised in that the concrete methods of realizing of the step A is:
Carry out the segmentation of multiple yardsticks to remote sensing image first, the selection of multi-split yardstick is according to the DN values of image or reflectivity
Scope is determined, during segmentation, form factor and degree of the compacting factor are set according to characteristics of image;
Secondly, rule set is created in segmentation software according to the segmentation range scale of selection, and according to yardstick from big to small or from
It is small to split step by step to big order;
Then, the object of multi-scale division result is formed into numbering file, exports as Raster Images successively, in Raster Images each
The numbering of object where value is exactly corresponding to pixel;
Subsequently, according to the numbering file of object after the Image Segmentation of acquisition, and original multiband image, calculate respectively every
The averaged spectrum of each object, obtains multiple dimensioned average spectral data under one segmentation yardstick;
Finally, training sample is chosen from original multiband image to the object averaged spectrum under different scale obtained above
Image is classified respectively, and the requirement of the training sample of selection is typical, randomness;
While classification, the posterior probability vector of the object under multiple yardsticks is obtained, for one on a certain segmentation yardstick
For object, its posterior probability vector represents as follows:
P=(P1, P2..., Pi..., Pn)
Wherein PiRepresent that object belongs to the probability of the i-th class, n represents classification number.
3. according to the method for claim 2, it is characterised in that the form factor is arranged to 0.2, and degree of the compacting factor is set
For 0.5.
4. according to the method described in claim 1 or 2 or 3, it is characterised in that split the selection of yardstick, according to the increase of yardstick,
The principle that yardstick interval also increases therewith is carried out.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510232524.8A CN104881677B (en) | 2015-05-08 | 2015-05-08 | Method is determined for the optimum segmentation yardstick of remote sensing image ground mulching classification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510232524.8A CN104881677B (en) | 2015-05-08 | 2015-05-08 | Method is determined for the optimum segmentation yardstick of remote sensing image ground mulching classification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104881677A CN104881677A (en) | 2015-09-02 |
CN104881677B true CN104881677B (en) | 2017-12-05 |
Family
ID=53949164
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510232524.8A Expired - Fee Related CN104881677B (en) | 2015-05-08 | 2015-05-08 | Method is determined for the optimum segmentation yardstick of remote sensing image ground mulching classification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104881677B (en) |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105427309B (en) * | 2015-11-23 | 2018-10-23 | 中国地质大学(北京) | The multiple dimensioned delamination process of object-oriented high spatial resolution remote sense information extraction |
CN105608474B (en) * | 2015-12-30 | 2019-07-09 | 二十一世纪空间技术应用股份有限公司 | Region adaptivity plant extraction method based on high resolution image |
CN106557784B (en) * | 2016-11-23 | 2020-05-01 | 上海航天控制技术研究所 | Rapid target identification method and system based on compressed sensing |
CN106651865B (en) * | 2016-12-23 | 2019-10-08 | 湖北工业大学 | A kind of optimum segmentation scale automatic selecting method of new high-resolution remote sensing image |
CN106780503A (en) * | 2016-12-30 | 2017-05-31 | 北京师范大学 | Remote sensing images optimum segmentation yardstick based on posterior probability information entropy determines method |
CN107067405B (en) * | 2017-03-30 | 2020-04-03 | 河海大学 | Remote sensing image segmentation method based on scale optimization |
CN107194187B (en) * | 2017-06-06 | 2019-08-06 | 国家基础地理信息中心 | Take the ground mulching sample size calculation method of space composition and configuration into account |
CN107480634A (en) * | 2017-08-12 | 2017-12-15 | 天津市测绘院 | A kind of geographical national conditions ground mulching monitoring method based on multistage target classification |
CN108053406B (en) * | 2017-09-11 | 2021-05-11 | 同济大学 | Surface coverage mapping method and device based on multi-resolution remote sensing image |
CN109063577B (en) * | 2018-07-05 | 2021-08-31 | 浙江大学 | Satellite image segmentation optimal segmentation scale determination method based on information gain rate |
CN109740631B (en) * | 2018-12-07 | 2021-02-12 | 中国科学院东北地理与农业生态研究所 | OBIA-SVM-CNN remote sensing image classification method based on object |
CN109583516A (en) * | 2018-12-24 | 2019-04-05 | 天津珞雍空间信息研究院有限公司 | A kind of space and time continuous PM2.5 inversion method based on ground and moonscope |
CN110991463B (en) * | 2019-11-04 | 2023-05-02 | 同济大学 | Multi-scale guide filtering feature extraction method under guidance of super-pixel map |
CN113902759B (en) * | 2021-10-13 | 2022-04-22 | 自然资源部国土卫星遥感应用中心 | Space-spectrum information combined satellite-borne hyperspectral image segmentation and clustering method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102509297A (en) * | 2011-11-14 | 2012-06-20 | 西安电子科技大学 | Clonal selection-based method for detecting change of remote sensing image with optimal entropy threshold |
CN103198480A (en) * | 2013-04-02 | 2013-07-10 | 西安电子科技大学 | Remote sensing image change detection method based on area and Kmeans clustering |
CN103413146A (en) * | 2013-08-23 | 2013-11-27 | 西安电子科技大学 | Method for finely classifying polarized SAR images based on Freeman entropy and self-learning |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8379085B2 (en) * | 2009-08-18 | 2013-02-19 | Behavioral Recognition Systems, Inc. | Intra-trajectory anomaly detection using adaptive voting experts in a video surveillance system |
-
2015
- 2015-05-08 CN CN201510232524.8A patent/CN104881677B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102509297A (en) * | 2011-11-14 | 2012-06-20 | 西安电子科技大学 | Clonal selection-based method for detecting change of remote sensing image with optimal entropy threshold |
CN103198480A (en) * | 2013-04-02 | 2013-07-10 | 西安电子科技大学 | Remote sensing image change detection method based on area and Kmeans clustering |
CN103413146A (en) * | 2013-08-23 | 2013-11-27 | 西安电子科技大学 | Method for finely classifying polarized SAR images based on Freeman entropy and self-learning |
Also Published As
Publication number | Publication date |
---|---|
CN104881677A (en) | 2015-09-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104881677B (en) | Method is determined for the optimum segmentation yardstick of remote sensing image ground mulching classification | |
CN109919206B (en) | Remote sensing image earth surface coverage classification method based on full-cavity convolutional neural network | |
Ryherd et al. | Combining spectral and texture data in the segmentation of remotely sensed images | |
CN107358260B (en) | Multispectral image classification method based on surface wave CNN | |
CN103208001B (en) | In conjunction with shape-adaptive neighborhood and the remote sensing image processing method of texture feature extraction | |
CN100595782C (en) | Classification method for syncretizing optical spectrum information and multi-point simulation space information | |
CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
CN107392130A (en) | Classification of Multispectral Images method based on threshold adaptive and convolutional neural networks | |
CN101587189B (en) | Texture elementary feature extraction method for synthetizing aperture radar images | |
CN103824309A (en) | Automatic extracting method of urban built-up area border | |
CN103048329A (en) | Pavement crack detecting method based on active contour model | |
CN113160062B (en) | Infrared image target detection method, device, equipment and storage medium | |
CN103020649A (en) | Forest type identification method based on texture information | |
CN106780503A (en) | Remote sensing images optimum segmentation yardstick based on posterior probability information entropy determines method | |
CN102073867B (en) | Sorting method and device for remote sensing images | |
CN106875481B (en) | A kind of production method of three-dimensional visualization remote sensing image Surface classification model | |
CN104217436A (en) | SAR image segmentation method based on multiple feature united sparse graph | |
Salah et al. | Evaluation of the self‐organizing map classifier for building detection from lidar data and multispectral aerial images | |
CN107229910A (en) | A kind of remote sensing images icing lake detection method and its system | |
Tang et al. | A multiple-point spatially weighted k-NN method for object-based classification | |
Chen et al. | A mathematical morphology-based multi-level filter of LiDAR data for generating DTMs | |
CN102081799B (en) | Method for detecting change of SAR images based on neighborhood similarity and double-window filtering | |
CN102938069A (en) | Pure and mixed pixel automatic classification method based on information entropy | |
Han et al. | An enhanced image binarization method incorporating with Monte-Carlo simulation | |
CN105205485B (en) | Large scale image partitioning algorithm based on maximum variance algorithm between multiclass class |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
EXSB | Decision made by sipo to initiate substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171205 Termination date: 20180508 |