CN107330875A - Based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images - Google Patents
Based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images Download PDFInfo
- Publication number
- CN107330875A CN107330875A CN201710396859.2A CN201710396859A CN107330875A CN 107330875 A CN107330875 A CN 107330875A CN 201710396859 A CN201710396859 A CN 201710396859A CN 107330875 A CN107330875 A CN 107330875A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing images
- water body
- change
- body surrounding
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Abstract
The present invention relates to based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images, it is uniform for the SLIC multi-scale division results contrasts of superimposed image, more uniform region (such as water body is distributed in characters of ground object, vegetation etc.) earth object of intimate square can be obtained, and will not destroy the boundary information of atural object in image when handling the region of atural object complex distribution, earth object will not be crushed excessively, relatively meet real atural object distribution；And by analyzing water body surrounding enviroment change test problems the characteristics of, organically combine spectrum, texture and index characteristic and build composite character space, the present invention has preferable effect in terms of the processing of " pseudo- change information ", illustrates that the present invention is preferable for the Detection results of change of water quality and vegetation Phenological change etc. " pseudo- change information ".
Description
Technical field
The present invention relates to based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images, belong to remote sensing
Technical field of image detection.
Background technology
Remote Sensing Imagery Change Detection is related to related the knowing of geographical science, information science and multiple subjects such as computer science
Know, be the hot research direction of current remote Sensing Image Analysis.Remote Sensing Imagery Change Detection refers to utilize image procossing, statistical analysis
The remote sensing image data of same geographic area but different phases is compared etc. technology, wrapped by analyzing in multi-temporal image
The change information contained points out the band of position that type of ground objects changes.
Research currently for Remote Sensing Imagery Change Detection technology is concentrated mainly in two phase Remote Sensing Imagery Change Detections,
To determine the position of region of variation as main purpose, and then carry out precision evaluation.Change detection for remote sensing images has been carried
A large amount of methods are gone out, pixel level, feature level and object level can be divided into according to the elementary cell rank of change detection process.
Change detecting method based on pixel has algebraic operation method, image converter technique and image classification method etc., with single picture
Member is detection unit, and mutual registering Multitemporal Remote Sensing Images are carried out into the comparison by pixel, the difference set up between correspondence pixel
Information, is then changed detection using different information.But such method is higher to the correction of multi-temporal image and with alignment request,
And it only considered the feature between single pixel, it is impossible to the good spatial information for utilizing surrounding adjacent to pixel, to noise more
It is sensitive.In order to overcome the change detecting method based on pixel easily by image irradiation difference, registration error and influence of noise
The shortcomings of, increase the robustness and accuracy of change detection algorithm, scholars propose the change detecting method using feature based：
Characters of ground object is extracted from multi-temporal image, analysis difference therein is carried out to characters of ground object, change information is detected.It is generally used for
The change detection of atural object (such as farmland, building) with particular edge feature or provincial characteristics.Object-based change inspection
Survey method is based on image segmentation and sorting technique, the spatially and spectrally characteristic around comprehensive utilization pixel, with homogeneity
Property pixel be combined composition earth object, then by relatively it is different when phase images in corresponding earth object attributive character
It is changed detection.Object oriented analysis method based on multi-scale division technology more meets the actual atural object distribution in image,
The spatial information of atural object can be made full use of, the generation of " spiced salt effect " is prevented effectively from, by the favor of scholars.
Water body periphery complicated topographical conditions, due to the reason such as orographic effect is obvious, special heterogeneity is high, build be applied to it is multiple
The change detecting method of miscellaneous orographic condition is always the difficult point of remote sensing application research.High-resolution remote sensing image can provide abundant
Atural object detailed information, but the atural object distribution situation of water body surrounding enviroment is complicated, and " the different spectrum of jljl ", " same object different images " phenomenon are tight
Weight, produces many interference informations.Simultaneously as the reason such as seasonal water-level fluctuation change, change of water quality and vegetation Phenological change,
Exist in multi-temporal image a lot " pseudo- change informations ", the change test problems of water body surrounding enviroment are faced with new challenges.It is existing
Method for detecting change of remote sensing image still suffer from many difficulties and problems have it is to be overcome and solve, how to effectively utilize remote sensing figure
The atural object detailed information enriched as in, distinguishes feature changes information interested and uninterested interference information, " puppet change is believed
Breath ", forms the change testing result of relatively accurate and robust, is the important research content of current change detection.
The content of the invention
The technical problems to be solved by the invention are to provide based on the forward and reverse heterogeneous water body surrounding enviroment of remote sensing images
Change detecting method, results in relatively accurate and robust change testing result, improves the detection efficiency of water body surrounding enviroment.
In order to solve the above-mentioned technical problem the present invention uses following technical scheme：The present invention is devised based on remote sensing images just
Reversely heterogeneous water body surrounding enviroment change detecting method, the change information of target water body surrounding environment is obtained for detecting,
Comprise the following steps：
Step 1. obtains t1 phases remote sensing images, the t2 phase remote sensing images corresponding to target water body, and for two phases
Remote sensing images carry out unified registration, subsequently into step 2；Wherein, t1 phases are located at before t2 phases；
Step 2. is obtained corresponding to target water body surrounding environment according to t1 phases remote sensing images and t2 phase remote sensing images
Each earth object, subsequently into step 3；
Step 3. is directed to t1 phases remote sensing images and t2 phase remote sensing images respectively, obtains the LBP corresponding to remote sensing images
Uniform pattern, normalization water body index NDWI and soil regulation vegetation index, and referred to according to LBP uniform patterns, normalization water body
Number NDWI, soil regulation vegetation index SAVI, build the composite character space corresponding to the remote sensing images, are derived from each phase
Remote sensing images distinguish corresponding composite character space, subsequently into step 4；
Each earth object of step 4. according to corresponding to target water body surrounding environment, and each phase remote sensing images difference
Corresponding composite character space, by t1 phases remote sensing images to t2 phase remote sensing images, forward direction builds each earth object
Forward is heterogeneous；Meanwhile, each earth object according to corresponding to target water body surrounding environment, and each phase remote sensing figure
As the corresponding composite character space of difference, by t2 phases remote sensing images to t1 phase remote sensing images, each atural object pair is reversely built
The Backward of elephant is heterogeneous, subsequently into step 5；
The Forward heterogeneities and Backward of step 5. each earth object according to corresponding to target water body surrounding environment
Heterogeneity, using maximal mathematical expectation algorithm, for each earth object, is classified by change and the class of non-changing two, is obtained
The change information of each change earth object, i.e. target water body surrounding environment corresponding to target water body surrounding environment.
It is used as a preferred technical solution of the present invention：In the step 3, obtaining each phase remote sensing images, institute is right respectively
While answering composite character space, also obtain t1 phases remote sensing images and the corresponding normalization of t2 phases remote sensing images difference is planted
By index NDVI；
Also include step 6 as follows, after execution of step 5, into step 6；
Step 6. normalized differential vegetation index corresponding according to t1 phases remote sensing images and t2 phases remote sensing images difference
NDVI, obtains the vegetation coverage information of each corresponding change earth object of target water body surrounding environment, and deletes mesh with this
Mark the pseudo- change information of vegetation in each corresponding change earth object of water body surrounding environment, more fresh target water body surrounding environment institute
Corresponding each change earth object, i.e. more change information of fresh target water body surrounding environment.
It is used as a preferred technical solution of the present invention：The step 2 comprises the following steps：
Step 2-1. carries out equal proportion for the data of the same band in t1 phases remote sensing images and t2 phase remote sensing images
It is added, superimposed image is obtained, subsequently into step 2-2；
Superimposed image is transformed into the feature space being made up of CIE Lab color spaces and position coordinates by step 2-2.
In, superimposed image is updated, subsequently into step 2-3；
Step 2-3. uses super-pixel generating algorithm, and multi-scale division is carried out for superimposed image, obtains segmentation result, so
Enter step 2-4 afterwards；
Segmentation result is linked on t1 phase remote sensing images with t2 phase remote sensing images by step 2-4. respectively, obtains mesh
Mark each earth object corresponding to water body surrounding environment.
It is used as a preferred technical solution of the present invention：In the step 4, according to corresponding to target water body surrounding environment
Each earth object, and each phase remote sensing images distinguish corresponding composite character space, during by t1 phases remote sensing images to t2
Phase remote sensing images, compare characteristic similarity of the earth object in composite character space by Distance conformability degree, and forward direction builds each
The Forward of individual earth object is heterogeneous.
It is used as a preferred technical solution of the present invention：The Distance conformability degree is card side's distance.
It is used as a preferred technical solution of the present invention：In the step 4, according to corresponding to target water body surrounding environment
Each earth object, and each phase remote sensing images distinguish corresponding composite character space, during by t2 phases remote sensing images to t1
Phase remote sensing images, by feature histogram of each earth object in composite character space, reversely build each earth object
Backward it is heterogeneous.
It is of the present invention used based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images more than
Technical scheme compared with prior art, with following technique effect：It is of the invention designed based on the forward and reverse heterogeneity of remote sensing images
Water body surrounding enviroment change detecting method, wherein, it is uniform for the SLIC multi-scale divisions results contrast of superimposed image, on ground
The more uniform region (such as water body, vegetation etc.) of thing feature distribution can obtain the earth object of intimate square, and on processing ground
The boundary information of atural object in image will not be destroyed during the region of thing complex distribution, earth object will not be crushed excessively, relatively meet true
Real atural object distribution；And the characteristics of change test problems by analyzing water body surrounding enviroment, organically combine and spectrum, texture and refer to
Number feature construction composite character space, the present invention has preferable effect in terms of the processing of " pseudo- change information ", illustrates this hair
The bright Detection results for change of water quality and vegetation Phenological change etc. " pseudo- change information " are preferable.
Brief description of the drawings
Fig. 1 is that the present invention is designed based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images
Schematic diagram；
Fig. 2 a and Fig. 2 b when corresponding to the two of embodiment one to remote sensing images respectively；
Fig. 3 a and Fig. 3 b when corresponding to the two of embodiment two to remote sensing images respectively；
Fig. 4 a and Fig. 4 b when corresponding to the two of embodiment three to remote sensing images respectively；
Fig. 5 a are the partial enlarged drawings of interference region in embodiment one；
Fig. 5 b are the partial enlarged drawings of interference region in embodiment three；
Fig. 6 a are in embodiment one typical " pseudo- change information "；
Fig. 6 b are in embodiment two typical " pseudo- change information "；
Fig. 6 c are in embodiment three typical " pseudo- change information "；
Fig. 7 a-Fig. 7 f are that various algorithms carry out multi-scale division to embodiment one and the qualitative results of change detection are contrasted；
Fig. 8 a-Fig. 8 f are that various algorithms carry out multi-scale division to embodiment two and the qualitative results of change detection are contrasted；
Fig. 9 a-Fig. 9 f are that various algorithms carry out multi-scale division to embodiment three and the qualitative results of change detection are contrasted.
Embodiment
The embodiment of the present invention is described in further detail with reference to Figure of description.
It is of the invention designed based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images, in spectrum
Water body index feature and vegetation index feature are merged with the basis of textural characteristics, constructs to change for water body surrounding enviroment and examines
The composite character space S TWV (Spectrum-Texture-NDWI-SAVI) of survey, then builds remote sensing in composite character space
Forward and reverse heterogeneity of image earth object is changed detection.Super-pixel generating algorithm SLIC (Simple are used first
Linear Iterative Cluster) superimposed images of two phases of processing obtains earth object, checks earth object from phase 1
Forward to phase 2 is heterogeneous, then it is heterogeneous to the Backward of phase 1 from phase 2 for inverse detection, and synthesis is forward and reverse
Heterogeneous information build earth object forward and reverse heterogeneity, then using EM (Expectation Maximization) algorithm
The change object of two phases is extracted with Bayes minimum error probability theory, the pseudo- change information of vegetation is finally excluded, formation is aligned
The change testing result of true and robust.
It is of the invention designed based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images, for examining
Survey in the change information for obtaining target water body surrounding environment, practical application, as shown in figure 1, specifically including following steps：
Step 1. obtains t1 phases remote sensing images, the t2 phase remote sensing images corresponding to target water body, and for two phases
Remote sensing images carry out unified registration, subsequently into step 2；Wherein, t1 phases are located at before t2 phases.
Step 2. is obtained corresponding to target water body surrounding environment according to t1 phases remote sensing images and t2 phase remote sensing images
Each earth object, subsequently into step 3.
Above-mentioned steps 2 specifically include following steps：
Step 2-1. carries out equal proportion for the data of the same band in t1 phases remote sensing images and t2 phase remote sensing images
It is added, superimposed image is obtained, subsequently into step 2-2.
Superimposed image is transformed into the feature space being made up of CIE Lab color spaces and position coordinates by step 2-2.
In, superimposed image is updated, subsequently into step 2-3.
Step 2-3. uses super-pixel generating algorithm (SLIC), carries out multi-scale division for superimposed image, is split
As a result, subsequently into step 2-4.
Segmentation result is linked on t1 phase remote sensing images with t2 phase remote sensing images by step 2-4. respectively, obtains mesh
Mark each earth object corresponding to water body surrounding environment.
Wherein, super-pixel generating algorithm (SLIC) is described as follows：
1) image is transformed into CIELab color spaces, then pixel xiLab color values and xy coordinates constitute 5 Wei Te
Levy vectorial Vi=[l, a, b, x, y], i=1,2 ..., N；
2) according to given super-pixel number, initial seed point is equably set, is k identical sizes by image pre-segmentation
Subregion, the distance of neighboring seeds point is approximatelyThe size of subregion is S*S；
3) seed point is reselected：, may in order to avoid seed point falls at the larger profile border of gradient or noise spot
The follow-up segmentation effect of influence, the minimum pixel of (n typically takes 3) Grad is set in initial seed point n*n contiguous ranges
New seed point；
4) the difference measurement D ' between pixel is defined, by color distance dcWith space length dsComposition：
Wherein, NsFor maximum space distance in class, N is defined ass=S, NcFor color distance maximum in class, according to picture
Contrast and cluster category setting are constant m (span [Isosorbide-5-Nitrae 0]), then D ' is turned to：
5) pixel searched for since cluster centre seed point in 2S*2S neighborhoods, it is therefore an objective to which accelerating algorithm restrains.Image
In each pixel may be searched by multiple seed points, the difference measurement of it and seed point is calculated respectively, difference value is taken most
Then small seed point gathers cluster centre identical pixel for a class as the cluster centre of the pixel；
6) iteration is clustered, each the class cluster divided to the last time, obtains its characteristic vector average VcIt is used as new cluster
Central seed point, is then re-flagged according to step 5.Iteration is to error convergence, i.e., the corresponding cluster centre of each pixel is not
When changing again, iteration ends generate preliminary super-pixel segmentation image；
Enhancing is connective, and preliminary super-pixel segmentation image may be present：Super-pixel is undersized, single
Super-pixel is cut into multiple discontinuous super-pixel etc..It is by algorithm of region growing that undersized or space is discontinuous super
Pixel in pixel is reassigned to the super-pixel of its spatial neighbor, makes the super-pixel space of generation continuous without excessively dividing
Dissipate.
Single features as the spectral signature of earth object is only considered in change detection process are difficult to reach higher
Precision, it is the focus direction studied at present that fusion various features, which build more stable and accurate change detection model,.Remote sensing images
The multiple features fusion method of change detection generally considers to merge the spectrum and Texture eigenvalue of earth object, in change detection
The processing of interference information have preferable effect.In the change test problems of water body surrounding enviroment, due to seasonal water quality,
The problems such as " pseudo- change information " that the change such as vegetation is produced, is still difficult to solve, but the index characteristic (water of these " pseudo- change informations "
Body index, vegetation index etc.) it is extremely obvious.Invariable rotary is had based on LBP (Local Binary Pattern) uniform pattern
Property and the advantages of gray scale consistency and normalization water body index feature NDWI (Normalized Difference Water
Index the outstanding performance) having during identifying water boy and soil regulation vegetation index SAVI (Soil-Adjusted
Vegetation Index) it can reduce the characteristics of Soil Background influences in vegetation pattern identification, fusion atural object is considered herein
Spectral signature, textural characteristics LBP uniform patterns and normalization water body index NDWI, the soil regulation vegetation index SAVI of object come
Construction feature space S TWV (Spectrum-Texture-NDWI-SAVI).To make full use of each dimensional feature pair in feature space
Feature changes situation in remote sensing images is described, and the quantification gradation of each dimensional feature in feature space is disposed as 256.
Step 3. is directed to t1 phases remote sensing images and t2 phase remote sensing images respectively, obtains the LBP corresponding to remote sensing images
Uniform pattern, normalization water body index NDWI and soil regulation vegetation index, and referred to according to LBP uniform patterns, normalization water body
Number NDWI, soil regulation vegetation index SAVI, build the composite character space corresponding to the remote sensing images, are derived from each phase
Remote sensing images distinguish corresponding composite character space S TWV；T1 phases remote sensing images and t2 phases remote sensing images point are also obtained simultaneously
Not corresponding normalized differential vegetation index NDVI, subsequently into step 4.
Wherein, LBP uniform patterns are based on original LBP features, and original LBP calculation formula is as follows：
Wherein, P is represented centered on pixel i, and radius is the pixel number in R local window, giAnd gpExpression innings respectively
The gray value of portion's window center pixel and neighborhood pixel.
Uniform pattern (UniformPattern) is defined as：(head and the tail element is included from 0 to 1 or from 1 to 0 transition times
Saltus step) be no more than 2 binary sequence.Research is found, the ratio of uniform pattern is met in original LBP 8 binary sequences
Example can reach more than 90%, therefore LBP uniform patterns are classified to original LBP binary sequence, by each uniform pattern
A class individually is classified as, then the other binary sequences for being unsatisfactory for uniform pattern are grouped into a class, calculation formula is as follows：
In formula (6), subscript riu2 represents that transition times maximum is 2, in P neighborhoods, meets U (LBPP, R)≤2 it is uniform
Number of modes is P* (P-1)+2.
In order to further reduce influence of the non-water body atural object to remote sensing images identifying water boy effect, scholars are poor in normalization
Divide proposition normalization difference water on the basis of vegetation index NDVI (Normalize Difference Vegetation Index)
Body index NDWI, is normalized difference processing to highlight the Water-Body Information in remote sensing images, it calculates public for specific band
Formula is as follows：
Wherein, Green represents green band, and NIR represents near infrared band.In general, vegetation is near infrared band
Reflectivity is maximum and water body then almost areflexia near infrared wavelength region, thus NDWI use based on green band with it is closely red
The ratio method of wave section builds Water indices model, can suppress the vegetation information and prominent Water-Body Information in image.Need
Illustrate, after the stretched processing of NDWI calculation formula, its result value scope be [- 1,1] so that different data sources or
Remotely-sensed data under different image-forming conditions results in the NDWI water body index features with close statistical property, in order to water body
Extraction model is analyzed and compared, so as to extract Water-Body Information therein.
Soil adjusts vegetation index SAVI to reduce influence of the Soil Background to vegetation index, in normalized differential vegetation index
Soil adjusted factor is added in NDVI, has the stronger ability for eliminating Influence To Soil and expression vegetation information, it calculates public
Formula is as follows：
Wherein, Red represents red band, and L is Soil adjusted factor, and span is that [0,1] (vegetation coverage gets over Gao Yue
Close to 0, otherwise closer to 1), it is to explain the optical signature change of Soil Background and adjust NDVI to the quick of Soil Background that it, which is acted on,
Sensitivity.As L=0, SAVI=NDVI represents that vegetation coverage is very high, can ignore the influence of Soil Background.
Each earth object of step 4. according to corresponding to target water body surrounding environment, and each phase remote sensing images difference
Corresponding composite character space S TWV, by t1 phases remote sensing images to t2 phase remote sensing images, by Distance conformability degree comparatively
Characteristic similarity of the thing object in composite character space, the Forward that forward direction builds each earth object is heterogeneous；
Conventional Distance conformability degree has absolute value distance, Euclidean distance and Ka Fang distances etc., wherein absolute value distance and Europe
Family name's distance in multiple feature spaces various features information for etc. power processing, fail reasonably to embody different characteristic structural differences shadow
Picture and the ability for characterizing change information, and card side conversion CST (Chi Square Transformation) can be according to two phases
Variance of the error image in different characteristic, considers the weighted value of various features, passes through the atural object pair of card side's transition structure
As heterogeneity is more objective and complete.
To ensure the uniformity of data, the various features mixed in multiple feature spaces STWV are normalized to [0,1] first
In the range of, for the earth object in two phase remote sensing images, the characteristics of mean vector representation of its interior pel is：Q is characterized vector dimension, then different by l-th of earth object Forward of card side's transformation calculations
Matter
Wherein,For the standard deviation of two phase earth object q dimensional feature differences.
Meanwhile, each earth object according to corresponding to target water body surrounding environment, and each phase remote sensing images difference
Corresponding composite character space S TWV, by t2 phases remote sensing images to t1 phase remote sensing images, by each earth object mixed
The feature histogram in feature space is closed, the Backward for reversely building each earth object is heterogeneous, subsequently into step 5.
Although analyzing the feature histogram of earth object it can be found that the interior pel feature distribution of earth object is different,
But its performance characteristic is consistent.The characteristic mean letter of earth object is only considered during in view of calculating Forward heterogeneities
Cease and fail to consider the feature distribution information of atural object, missing inspection problem may be caused, and the feature histogram of earth object is not only wrapped
Contain the characteristic mean information of atural object, further comprises the feature distribution information of atural object, it is considered to utilize the feature Nogata of earth object
Figure is heterogeneous to build its Backward so that change testing result is more accurate.
The log likelihood ratio statistic (Log-Likelihood Ratio Statistic), also known as G statistics, are a kind of non-
Parametric statistical methods, it is not necessary to the distribution situation of stochastic variable set is made in advance any hypothesis can be used to weigh two it is random
Similarity between variables collection, histogram similarity of the two phase earth objects on q dimensional features is weighed using G statistics
(q=1,2 .., Q), is calculated as follows：
Wherein, Ht1And Ht2For the feature histogram of two phase remote sensing images earth objects, fqFor the q Wei Te of earth object
Probability density function is levied, L is the quantification gradation of this feature.For single earth object, its histogram cumulative probability value is
1, then have：
Then above formula is turned to：
To make full use of the information that all features are included, l-th of earth object is calculated in mixing multiple feature spaces STWV
The weighted average of the feature histogram similarity of each characteristic dimension is heterogeneous as its Backward
Wherein, phase images can be weighed in this feature dimension in the entropy of the differential image of q dimensional features when E (q) is two
Differential image included in information content, information content it is bigger explanation earth object this feature dimension in difference it is bigger, its is right
The weights omega answeredqAlso it is bigger.
The Forward heterogeneities for obtaining l-th of earth object are calculated in mixing multiple feature spaces STWVWith
Backward is heterogeneousAfterwards, you can obtain forward and reverse heterogeneity of the earth object
The Forward heterogeneities and Backward of step 5. each earth object according to corresponding to target water body surrounding environment
Heterogeneity, using maximal mathematical expectation algorithm (EM), for each earth object, is classified by change and the class of non-changing two,
The change information of each change earth object, i.e. target water body surrounding environment corresponding to target water body surrounding environment is obtained, so
Enter step 6 afterwards.
Calculate forward and reverse heterogeneous set of earth object：Wherein n is the number of earth object.
In two phase remote sensing images, the earth object changed heterogeneous larger and the earth object that does not change is heterogeneous
Smaller, then the element in D can be divided into change and not change two major classes.Assuming that D meets the mixing height being made up of two Gaussian components
This distribution GMM (Gaussian Mixture Model), its density function is represented by：
Wherein lcAnd luIt is to change classification and do not change class label, is designated as l ∈ { lc,lu, p (l) is l dvielements in D
Shared ratio, meets p (lc)+p(lu)=1,For probability density function, Gaussian distributed：
Maximal mathematical expectation EM (Expectation Maximization) algorithm is a kind of solution probabilistic model parameter
The parameter p (l) and mean μ of mixture gaussian modelling in maximum Likelihood, above-mentioned hypothesisl, standard deviation sigmalIt can use
EM algorithms are estimated.D is gathered for two classes with K-means algorithms first, renewal is then iterated as the following formula, until convergence：
Calculated by Bayesian formulaBelong to classification l posterior probability：
It is according to minimal error rate principleDistribute class label lcOr lu, obtain the change information of correspondence earth object.
In the change test problems research process of water body surrounding enviroment, the real change that only type of ground objects changes
Belong to change information interested.Due to reasons such as season, weathers, same vegetative coverage region (especially farmland region) is not
Vegetation characteristics with phase may change, and produce " pseudo- change information ", belong to uninterested change information.
Step 6. normalized differential vegetation index corresponding according to t1 phases remote sensing images and t2 phases remote sensing images difference
NDVI, obtains the vegetation coverage information of each corresponding change earth object of target water body surrounding environment, and deletes mesh with this
Mark the pseudo- change information of vegetation in each corresponding change earth object of water body surrounding environment, more fresh target water body surrounding environment institute
Corresponding each change earth object, i.e. more change information of fresh target water body surrounding environment.
Specifically, calculating the vegetation coverage of different phase pixels by normalized differential vegetation index NDVI, then estimation becomes
Change the ratio of vegetation pixel in class earth object, by vegetation pixel scale all the higher person in two phase earth objects, it is believed that the change
It is caused by the Seasonal dynamics change of vegetation, to belong to the pseudo- change information of vegetation, comprise the following steps that to change earth object：
1. calculate pixel i vegetation coverage Fc：
Fc=(NDVIi-NDVImin)/(NDVImax-NDVImin) (20)
Wherein, NDVIminFor the NDVI minimum values of pixel in remote sensing images, NDVImaxFor maximum.
2. calculate the ratio N of vegetation pixel in change objectf：
Wherein, FctFor the threshold value of the vegetation coverage of vegetation pixel, 0.5 is set to, the vegetation coverage of pixel is more than should
Threshold value shows that the pixel is vegetation pixel, and n is the pixel count for changing object.
If 3. changed in the earth object of corresponding two phases of object, the ratio N of vegetation pixelfAll it is higher than 0.65,
Then think the change object pseudo- change as caused by vegetation Phenological change.
, based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images, it will be applied designed by the present invention
To among actual, experimental data of the invention applies the stone Lianghe reservoir neighboring area (position that center is provided for China Satecom's resource
In Jiangsu Province northeast) the 1A grades of GF-1 PMS remote sensing images of Level in April, 2015 and in April, 2016, including spatial discrimination
Rate is 2m full-colored data and 8m multispectral data (spectral region is 0.45 μm~0.89 μm), is pre-processed by ENVI5.1
After the image co-registration resampling converted based on GS, the spatial resolution of fused images is 2m.Ground species in reservoir surrounding enviroment
Type is various and distribution situation is complicated, and the influence of interference information and " pseudo- change information " to change detection is larger, and choosing three groups should
The view data of region embodiment is changed the validity that test experience carrys out overall merit context of methods.Wherein, Fig. 2 a, Fig. 3 a
It is respectively the t1 phase remote sensing images in each embodiment corresponding 18 days April in 2015 with Fig. 4 a, Fig. 2 b, Fig. 3 b and Fig. 4 b are respectively
Each embodiment correspondence 2 i.e. t2 phase remote sensing images on April 21st, 2016, wherein embodiment one corresponds to two phase remote sensing respectively
The size of image is 1166*881, and the size that embodiment two corresponds to two phase remote sensing images respectively is 597*452, three points of embodiment
Not Dui Ying the sizes of two phase remote sensing images be 747*564；The change information of Experimental Area mainly includes dam, house, paddy field
Build and transform Deng atural object, and the vegetative coverage region such as farmland, forest Phenological change and change of water quality etc. " puppet change ".
Some interference informations caused due to factors such as image-forming condition difference, illumination variations are contained in experimental data, such as
Fig. 5 a are the partial enlarged drawings of interference region in embodiment one, and type of ground objects is bare area；Fig. 5 b are interference regions in embodiment three
Partial enlarged drawing, type of ground objects is building.
The change detection influence of " pseudo- change information " on water body surrounding enviroment in Experimental Area is very big, and Fig. 6 a are embodiments
In one typical " pseudo- change information ", the brightness change in dam region；Fig. 6 b are typical " pseudo- change information ", water in embodiment two
Qualitative change；Fig. 6 c are vegetation puppet changes in embodiment three typical " pseudo- change information ".
Embodiment one and embodiment two change information such as are building and transformed comprising reservoir borderline region artificial dam
Meanwhile, the change information produced by reasons such as water-level fluctuations is also contains, " pseudo- change information " is mainly derived from reservoir water body
The change of water quality of itself and the brightness change of dam region；Main change information in embodiment three is Land-use
Change, including house and paddy field are built, vegetation Phenological change and change of water quality be its main " pseudo- change information " source.For
Effectiveness of the invention is verified, detection is changed to experimental data using many algorithms below, then testing result is carried out
Contrast and analysis.
To verify the validity of this paper algorithms, detection is changed to experimental data using many algorithms：Fig. 7 a-Fig. 7 f
It is that various algorithms carry out multi-scale division to embodiment one and the qualitative results of change detection are contrasted；Fig. 8 a-Fig. 8 f are various calculations
Method carries out multi-scale division to embodiment two and the qualitative results of change detection are contrasted；Fig. 9 a-Fig. 9 f are various algorithms to implementation
Example three carries out the qualitative results contrast of multi-scale division and change detection；Wherein, a refers to after two phase imaging importings using SLIC's
The result of segmentation；B refers to the reference testing result for interpreting obtain by visual observation；C refers to the Change vector Analysis method (CVA) of object-oriented
Testing result；D refers to the change testing result that (CST) is converted based on card side；E refers to is changed detection based on histogram (HIST)
Result；F refers to white portion in the testing result of this paper algorithms, figure and represents region of variation, and black region represents non-region of variation.
Preferably to verify robustness and validity that this paper algorithms change in detection in water body surrounding enviroment, below to upper
State change testing result and carry out Comparative and Quantitative Analysis.The precision of change testing result is evaluated with pixel quantitative error criterion, its
Middle empty inspection rate is that flase drop is that the number for changing pixel accounts in testing result the ratio for changing pixel sum, and loss is the change of missing inspection
Change the ratio that pixel number accounts for unchanged pixel sum in testing result, accuracy is that the pixel number correctly detected accounts for picture in image
The accuracy assessment of 4 kinds of change detection scheme experimental results is as shown in table 1 below in the ratio of first sum, upper section：
Table 1
Consolidated statement 1 and Fig. 7 a-Fig. 7 f, Fig. 8 a-Fig. 8 f and Fig. 9 a-Fig. 9 f changes detection Detection results figure it can be found that
The missing inspection situation of several algorithm process results is all less, by loss control in tolerance interval, but in flase drop situation and mistake
Then show different in terms of inspection rate：The Change vector Analysis method (CVA) of object-oriented can utilize the various features of multiple feature spaces
Information, detects most of region of variation in phase images when two, but it uses Euclidean distance in multiple feature spaces by each spy
Reference breath for etc. power processing, fail reasonably to embody the ability that different characteristic characterizes change information, the situation of flase drop is more tight
Weight；Flase drop situation greatly reduces for the change testing result for converting (CST) based on card side compares Change vector Analysis method；It is based on
The presence of the change testing result of histogram (HIST) more serious miss detection, reason be because when two phase images middlely
The feature distribution information of thing is disturbed even more serious by " the different spectrum of jljl " phenomenon, so as to influence to change testing result；In several groups of data
Change test experience in, this paper algorithms obtain preferable experiment effect and higher accuracy rate, while by false drop rate control
In relatively low scope, illustrate that this paper algorithms have preferable detectability for the change detection of water body surrounding enviroment.
Embodiments of the present invention are explained in detail above in conjunction with accompanying drawing, but the present invention is not limited to above-mentioned implementation
Mode, can also be on the premise of present inventive concept not be departed from the knowledge that those of ordinary skill in the art possess
Make a variety of changes.
Claims (6)
1. based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images, target water body is obtained for detecting
The change information of surrounding environment, it is characterised in that comprise the following steps：
Step 1. obtains t1 phases remote sensing images, the t2 phase remote sensing images corresponding to target water body, and for two phase remote sensing
Image carries out unified registration, subsequently into step 2；Wherein, t1 phases are located at before t2 phases；
Step 2. obtains each corresponding to target water body surrounding environment according to t1 phases remote sensing images and t2 phase remote sensing images
Earth object, subsequently into step 3；
Step 3. is directed to t1 phases remote sensing images and t2 phase remote sensing images respectively, and the LBP obtained corresponding to remote sensing images is uniform
Pattern, normalization water body index NDWI and soil regulation vegetation index, and according to LBP uniform patterns, normalization water body index
NDWI, soil regulation vegetation index SAVI, build the composite character space corresponding to the remote sensing images, are derived from each phase distant
Feel image and distinguish corresponding composite character space, subsequently into step 4；
Each earth object of step 4. according to corresponding to target water body surrounding environment, and each phase remote sensing images are corresponded to respectively
Composite character space, by t1 phases remote sensing images to t2 phase remote sensing images, forward direction builds the Forward of each earth object
It is heterogeneous；Meanwhile, each earth object according to corresponding to target water body surrounding environment, and each phase remote sensing images are right respectively
The composite character space answered, by t2 phases remote sensing images to t1 phase remote sensing images, reversely builds each earth object
Backward is heterogeneous, subsequently into step 5；
The Forward of step 5. each earth object according to corresponding to target water body surrounding environment is heterogeneous and Backward is heterogeneous
Property, using maximal mathematical expectation algorithm, for each earth object, classified by change and the class of non-changing two, obtain target
The change information of each change earth object, i.e. target water body surrounding environment corresponding to water body surrounding environment.
2. according to claim 1 based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images, its
It is characterised by：In the step 3, while each phase remote sensing images corresponding composite character space respectively is obtained, also obtain
T1 phases remote sensing images and the corresponding normalized differential vegetation index NDVI of t2 phases remote sensing images difference；
Also include step 6 as follows, after execution of step 5, into step 6；
The step 6. normalized differential vegetation index NDVI corresponding according to t1 phases remote sensing images and t2 phases remote sensing images difference, is obtained
The vegetation coverage information of each corresponding change earth object of target water body surrounding environment is obtained, and with this delete target water body week
The pseudo- change information of vegetation in each corresponding change earth object of collarette border, it is each corresponding to more fresh target water body surrounding environment
Individual change earth object, the i.e. more change information of fresh target water body surrounding environment.
3. it is according to claim 1 or claim 2 based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images,
Characterized in that, the step 2 comprises the following steps：
Step 2-1. carries out equal proportion with the data of the same band in t2 phase remote sensing images for t1 phases remote sensing images and is added,
Superimposed image is obtained, subsequently into step 2-2；
Superimposed image is transformed into the feature space being made up of CIE Lab color spaces and position coordinates by step 2-2., more
New superimposed image, subsequently into step 2-3；
Step 2-3. uses super-pixel generating algorithm, and multi-scale division is carried out for superimposed image, obtains segmentation result, Ran Houjin
Enter step 2-4；
Segmentation result is linked on t1 phase remote sensing images with t2 phase remote sensing images by step 2-4. respectively, obtains target water
Each earth object corresponding to body surrounding environment.
4. it is according to claim 1 or claim 2 based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images,
Characterized in that, in the step 4, each earth object according to corresponding to target water body surrounding environment, and each phase are distant
Feel image and distinguish corresponding composite character space, by t1 phases remote sensing images to t2 phase remote sensing images, pass through Distance conformability degree
Compare characteristic similarity of the earth object in composite character space, the Forward that forward direction builds each earth object is heterogeneous.
5. according to claim 4 based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images, its
It is characterised by, the Distance conformability degree is card side's distance.
6. it is according to claim 1 or claim 2 based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images,
Characterized in that, in the step 4, each earth object according to corresponding to target water body surrounding environment, and each phase are distant
Feel image and distinguish corresponding composite character space, by t2 phases remote sensing images to t1 phase remote sensing images, pass through each atural object pair
As the feature histogram in composite character space, the Backward for reversely building each earth object is heterogeneous.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710396859.2A CN107330875B (en) | 2017-05-31 | 2017-05-31 | Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710396859.2A CN107330875B (en) | 2017-05-31 | 2017-05-31 | Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107330875A true CN107330875A (en) | 2017-11-07 |
| CN107330875B CN107330875B (en) | 2020-04-21 |
Family
ID=60192896
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710396859.2A Active CN107330875B (en) | 2017-05-31 | 2017-05-31 | Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107330875B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108257160A (en) * | 2018-01-22 | 2018-07-06 | 西安理工大学 | Remote sensing image variation detection post-processing approach based on multi-scale division-greatest hope |
| CN108564002A (en) * | 2018-03-22 | 2018-09-21 | 中国科学院遥感与数字地球研究所 | A kind of remote sensing image time series variation detection method and system |
| CN109657598A (en) * | 2018-12-13 | 2019-04-19 | 宁波大学 | Seashore wetland Classification in Remote Sensing Image method based on Stratified Strategy |
| CN110070545A (en) * | 2019-03-20 | 2019-07-30 | 重庆邮电大学 | A kind of method that textural characteristics density in cities and towns automatically extracts cities and towns built-up areas |
| CN110135492A (en) * | 2019-05-13 | 2019-08-16 | 山东大学 | Equipment fault diagnosis and method for detecting abnormality and system based on more Gauss models |
| CN110309781A (en) * | 2019-07-01 | 2019-10-08 | 中国科学院遥感与数字地球研究所 | Damage remote sensing recognition method in house based on the fusion of multi-scale spectrum texture self-adaption |
| CN112417935A (en) * | 2019-08-23 | 2021-02-26 | 经纬航太科技股份有限公司 | Environment inspection system and method |
| CN113125358A (en) * | 2021-04-26 | 2021-07-16 | 山东深蓝智谱数字科技有限公司 | Hyperspectrum-based food pesticide residue detection method, device and medium |
| CN113160183A (en) * | 2021-04-26 | 2021-07-23 | 山东深蓝智谱数字科技有限公司 | Hyperspectral data processing method, device and medium |
| CN113553967A (en) * | 2021-07-28 | 2021-10-26 | 广东工业大学 | Vegetation and water body change detection method and system based on remote sensing image |
| CN117312973A (en) * | 2023-09-26 | 2023-12-29 | 中国科学院空天信息创新研究院 | Inland water body optical classification method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103489193A (en) * | 2013-09-30 | 2014-01-01 | 河海大学 | High-resolution remote-sensing image change detection method facing targets and based on integrating strategy |
| US20160055305A1 (en) * | 2014-08-19 | 2016-02-25 | eagleyemed, Inc. | Video enhancements for live sharing of medical images |
-
2017
- 2017-05-31 CN CN201710396859.2A patent/CN107330875B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103489193A (en) * | 2013-09-30 | 2014-01-01 | 河海大学 | High-resolution remote-sensing image change detection method facing targets and based on integrating strategy |
| US20160055305A1 (en) * | 2014-08-19 | 2016-02-25 | eagleyemed, Inc. | Video enhancements for live sharing of medical images |
Non-Patent Citations (1)
| Title |
|---|
| 张明哲等: "基于超像素分割和多方法融合的SAR图像变化检测方法", 《遥感技术与应用》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108257160A (en) * | 2018-01-22 | 2018-07-06 | 西安理工大学 | Remote sensing image variation detection post-processing approach based on multi-scale division-greatest hope |
| CN108257160B (en) * | 2018-01-22 | 2021-10-19 | 西安理工大学 | Remote sensing image change detection post-processing method based on multi-scale segmentation-maximum expectation |
| CN108564002A (en) * | 2018-03-22 | 2018-09-21 | 中国科学院遥感与数字地球研究所 | A kind of remote sensing image time series variation detection method and system |
| CN109657598A (en) * | 2018-12-13 | 2019-04-19 | 宁波大学 | Seashore wetland Classification in Remote Sensing Image method based on Stratified Strategy |
| CN109657598B (en) * | 2018-12-13 | 2020-05-05 | 宁波大学 | Coastal wetland remote sensing classification method based on layering strategy |
| CN110070545A (en) * | 2019-03-20 | 2019-07-30 | 重庆邮电大学 | A kind of method that textural characteristics density in cities and towns automatically extracts cities and towns built-up areas |
| CN110070545B (en) * | 2019-03-20 | 2023-05-26 | 重庆邮电大学 | Method for automatically extracting urban built-up area by urban texture feature density |
| CN110135492A (en) * | 2019-05-13 | 2019-08-16 | 山东大学 | Equipment fault diagnosis and method for detecting abnormality and system based on more Gauss models |
| CN110309781B (en) * | 2019-07-01 | 2021-03-02 | 中国科学院空天信息创新研究院 | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion |
| CN110309781A (en) * | 2019-07-01 | 2019-10-08 | 中国科学院遥感与数字地球研究所 | Damage remote sensing recognition method in house based on the fusion of multi-scale spectrum texture self-adaption |
| CN112417935A (en) * | 2019-08-23 | 2021-02-26 | 经纬航太科技股份有限公司 | Environment inspection system and method |
| CN113125358A (en) * | 2021-04-26 | 2021-07-16 | 山东深蓝智谱数字科技有限公司 | Hyperspectrum-based food pesticide residue detection method, device and medium |
| CN113160183A (en) * | 2021-04-26 | 2021-07-23 | 山东深蓝智谱数字科技有限公司 | Hyperspectral data processing method, device and medium |
| CN113553967A (en) * | 2021-07-28 | 2021-10-26 | 广东工业大学 | Vegetation and water body change detection method and system based on remote sensing image |
| CN117312973A (en) * | 2023-09-26 | 2023-12-29 | 中国科学院空天信息创新研究院 | Inland water body optical classification method and system |
| CN117312973B (en) * | 2023-09-26 | 2024-05-03 | 中国科学院空天信息创新研究院 | Inland water body optical classification method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107330875B (en) | 2020-04-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107330875A (en) | Based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images | |
| Shendryk et al. | Deep learning for multi-modal classification of cloud, shadow and land cover scenes in PlanetScope and Sentinel-2 imagery | |
| CN109063754B (en) | Remote sensing image multi-feature joint classification method based on OpenStreetMap | |
| CN108830870B (en) | Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning | |
| CN103971115B (en) | Automatic extraction method for newly-increased construction land image spots based on NDVI and PanTex index | |
| CN101840581B (en) | Method for extracting profile of building from satellite remote sensing image | |
| CN108846832A (en) | A kind of change detecting method and system based on multi-temporal remote sensing image and GIS data | |
| CN109034233B (en) | High-resolution remote sensing image multi-classifier joint classification method combined with OpenStreetMap | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| CN107229917A (en) | A kind of several remote sensing image general character well-marked target detection methods clustered based on iteration | |
| CN104282008B (en) | The method and apparatus that Texture Segmentation is carried out to image | |
| CN110309780A (en) | High resolution image houseclearing based on BFD-IGA-SVM model quickly supervises identification | |
| CN110598564B (en) | OpenStreetMap-based high-spatial-resolution remote sensing image transfer learning classification method | |
| CN110097101A (en) | A kind of remote sensing image fusion and seashore method of tape sorting based on improvement reliability factor | |
| Peng et al. | Object-based change detection from satellite imagery by segmentation optimization and multi-features fusion | |
| Jamil et al. | Tree species extraction and land use/cover classification from high-resolution digital orthophoto maps | |
| CN113963222B (en) | High-resolution remote sensing image change detection method based on multi-strategy combination | |
| Truax | Comparing spectral and object based approaches for classification and transportation feature extraction from high resolution multispectral imagery | |
| CN108492288B (en) | Random forest based multi-scale layered sampling high-resolution satellite image change detection method | |
| CN107992856A (en) | High score remote sensing building effects detection method under City scenarios | |
| CN110473205A (en) | Remote sensing image information extracting method and system based on arrow bar phantom | |
| Chen et al. | Heterogeneous images change detection based on iterative joint global–local translation | |
| CN111882573A (en) | Cultivated land plot extraction method and system based on high-resolution image data |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |