CN107330875B - Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image - Google Patents
Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image Download PDFInfo
- Publication number
- CN107330875B CN107330875B CN201710396859.2A CN201710396859A CN107330875B CN 107330875 B CN107330875 B CN 107330875B CN 201710396859 A CN201710396859 A CN 201710396859A CN 107330875 B CN107330875 B CN 107330875B
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- water body
- surrounding environment
- ground object
- sensing image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 230000008859 change Effects 0.000 title claims abstract description 118
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 104
- 238000001514 detection method Methods 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 claims abstract description 30
- 230000011218 segmentation Effects 0.000 claims abstract description 22
- 238000004422 calculation algorithm Methods 0.000 claims description 29
- 239000002689 soil Substances 0.000 claims description 15
- 208000027066 STING-associated vasculopathy with onset in infancy Diseases 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 12
- 238000012545 processing Methods 0.000 abstract description 10
- 238000001228 spectrum Methods 0.000 abstract description 7
- 238000011160 research Methods 0.000 description 7
- 230000009466 transformation Effects 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000003595 spectral effect Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 230000001932 seasonal effect Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000013139 quantization Methods 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 235000002566 Capsicum Nutrition 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 239000006002 Pepper Substances 0.000 description 1
- 235000016761 Piper aduncum Nutrition 0.000 description 1
- 235000017804 Piper guineense Nutrition 0.000 description 1
- 244000203593 Piper nigrum Species 0.000 description 1
- 235000008184 Piper nigrum Nutrition 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
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
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing images, aiming at the fact that SLIC multi-scale segmentation results of superposed images are uniform, nearly square ground object objects can be obtained in areas (such as water bodies, vegetations and the like) with uniform ground object feature distribution, boundary information of ground objects in the images cannot be damaged when the areas with complex ground object distribution are processed, and the ground object objects cannot be excessively crushed and are more in line with real ground object distribution; and through analyzing the characteristics of the detection problem of the change of the surrounding environment of the water body, the mixed characteristic space is constructed by organically combining the spectrum, the texture and the index characteristics, the method has a better effect on the aspect of processing the 'pseudo change information', and the detection effect of the method on the 'pseudo change information' such as the water quality change, the vegetation climate change and the like is better.
Description
Technical Field
The invention relates to a method for detecting changes of surrounding environment of a water body based on forward and reverse heterogeneity of a remote sensing image, and belongs to the technical field of remote sensing image detection.
Background
The remote sensing image change detection relates to the related knowledge of multiple disciplines such as geographic science, information science, computer science and the like, and is a hotspot research direction of remote sensing image analysis at present. The remote sensing image change detection is to compare remote sensing image data of the same geographical area but different time phases by using the technologies of image processing, statistical analysis and the like, and to indicate a position area where the type of a ground object changes by analyzing change information contained in a multi-time-phase image.
At present, research aiming at the remote sensing image change detection technology mainly focuses on two-time phase remote sensing image change detection, and the method mainly aims at determining the position of a change area so as to perform precision evaluation. A number of methods have been proposed for change detection of remote sensing images, which can be classified into pixel, feature and object levels according to the basic unit level of the change detection process.
The pixel-based change detection method comprises an algebraic operation method, an image transformation method, an image classification method and the like, a single pixel is used as a detection unit, multi-temporal remote sensing images which are mutually registered are compared pixel by pixel, difference information among corresponding pixels is established, and then change detection is carried out by utilizing the difference information. However, the method has high requirements on the correction and registration of multi-temporal images, only considers the characteristics among single pixels, cannot well utilize the spatial information of surrounding adjacent pixels, and is sensitive to noise. In order to overcome the defects that the pixel-based change detection method is easily influenced by image illumination difference, registration error, noise and the like, and increase the robustness and accuracy of the change detection algorithm, scholars propose to adopt a feature-based change detection method: and extracting feature features from the multi-temporal image, analyzing differences among the feature features, and detecting change information. The method is generally used for detecting the change of the ground features (such as farmlands, buildings and the like) with special edge features or area features. The object-based change detection method is based on image segmentation and classification technology, comprehensively utilizes the spatial and spectral characteristics around pixels, combines the pixels with homogeneity to form a ground object, and then carries out change detection by comparing corresponding ground object attribute characteristics in images of different time phases. The object-oriented analysis method based on the multi-scale segmentation technology better conforms to the actual ground feature distribution in the image, can fully utilize the spatial information of the ground features, effectively avoids the generation of the salt and pepper effect, and is favored by students.
The method has the advantages that the terrain conditions around the water body are complex, and due to the reasons of obvious terrain effect, high spatial heterogeneity and the like, the difficulty of remote sensing application research is always to construct a change detection method suitable for the complex terrain conditions. The high-resolution remote sensing image can provide abundant ground feature detail information, but the ground feature distribution condition of the surrounding environment of the water body is complex, and phenomena of 'same-object different-spectrum' and 'same-spectrum foreign matter' are serious, so that a lot of interference information is generated. Meanwhile, due to seasonal water level fluctuation changes, water quality changes, vegetation phenological changes and the like, a plurality of 'pseudo-change information' exist in multi-temporal images, and the problem of change detection of the surrounding environment of the water body faces new challenges. The existing remote sensing image change detection method still has a lot of difficulties and problems to be overcome and solved, how to effectively utilize abundant ground feature detail information in the remote sensing image to distinguish interested ground feature change information from uninteresting interference information and pseudo change information to form a relatively accurate and robust change detection result, and the method is an important research content of current change detection.
Disclosure of Invention
The invention aims to solve the technical problem of providing a water body surrounding environment change detection method based on the forward and reverse heterogeneity of a remote sensing image, so that a relatively accurate and robust change detection result can be obtained, and the detection efficiency of the water body surrounding environment is improved.
The invention adopts the following technical scheme for solving the technical problems: the invention designs a water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing images, which is used for detecting and obtaining change information of the surrounding environment of a target water body, and comprises the following steps:
step 2, obtaining each ground object corresponding to the surrounding environment of the target water body according to the t1 time phase remote sensing image and the t2 time phase remote sensing image, and then entering step 3;
step 3, aiming at the t1 time phase remote sensing image and the t2 time phase remote sensing image respectively, obtaining LBP uniform mode image characteristics, normalized water body indexes NDWI and soil regulation vegetation indexes corresponding to the remote sensing images, constructing a mixed characteristic space corresponding to the remote sensing images according to the LBP uniform mode image characteristics, the normalized water body indexes NDWI and the soil regulation vegetation indexes SAVI, thus obtaining mixed characteristic spaces corresponding to the time phase remote sensing images respectively, and then entering step 4;
and 5, classifying the various ground object objects according to variation and non-variation by adopting a maximum mathematical expectation algorithm according to the Forward heterogeneity and the Backward heterogeneity of the various ground object objects corresponding to the surrounding environment of the target water body to obtain the variation information of the various varied ground object objects corresponding to the surrounding environment of the target water body, namely the variation information of the surrounding environment of the target water body.
As a preferred technical scheme of the invention: in the step 3, while the mixed feature space corresponding to each time phase remote sensing image is obtained, normalized vegetation indexes NDVI corresponding to the t1 time phase remote sensing image and the t2 time phase remote sensing image are also obtained;
step 6, after step 5, entering step 6;
and 6, acquiring vegetation coverage information of each variable land feature object corresponding to the surrounding environment of the target water body according to the normalized vegetation indexes NDVI corresponding to the t1 time-phase remote sensing image and the t2 time-phase remote sensing image respectively, deleting vegetation pseudo-variation information in each variable land feature object corresponding to the surrounding environment of the target water body, and updating each variable land feature object corresponding to the surrounding environment of the target water body, namely updating variation information of the surrounding environment of the target water body.
As a preferred technical solution of the present invention, the step 2 includes the steps of:
step 2-1, carrying out equal proportion addition on data of the same wave band in the t1 time phase remote sensing image and the t2 time phase remote sensing image to obtain a superposed image, and then entering step 2-2;
2-2, converting the superposed image into a feature space formed by a CIE Lab color space and position coordinates, updating the superposed image, and then entering the step 2-3;
step 2-3, performing multi-scale segmentation on the superposed image by adopting a super-pixel generation algorithm to obtain a segmentation result, and then entering step 2-4;
and 2-4, respectively sleeving the segmentation result on the t1 time-phase remote sensing image and the t2 time-phase remote sensing image to obtain each ground object corresponding to the surrounding environment of the target water body.
As a preferred technical scheme of the invention: the distance similarity is a chi-square distance.
Compared with the prior art, the method for detecting the change of the surrounding environment of the water body based on the forward and reverse heterogeneity of the remote sensing image has the following technical effects: the invention designs a water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing images, wherein SLIC multi-scale segmentation results aiming at superimposed images are relatively uniform, nearly square ground object objects can be obtained in areas (such as water bodies, vegetation and the like) with relatively uniform ground object feature distribution, boundary information of ground objects in the images cannot be damaged when the areas with complex ground object distribution are processed, and the ground object objects cannot be excessively crushed and are relatively in line with real ground object distribution; and through analyzing the characteristics of the detection problem of the change of the surrounding environment of the water body, the mixed characteristic space is constructed by organically combining the spectrum, the texture and the index characteristics, the method has a better effect on the aspect of processing the 'pseudo change information', and the detection effect of the method on the 'pseudo change information' such as the water quality change, the vegetation climate change and the like is better.
Drawings
FIG. 1 is a schematic diagram of a method for detecting changes in the surrounding environment of a water body based on forward and reverse heterogeneity of remote sensing images, which is disclosed by the invention;
fig. 2a and fig. 2b respectively correspond to two time-direction remote sensing images of the first embodiment;
fig. 3a and 3b respectively correspond to two time-direction remote sensing images of the second embodiment;
fig. 4a and 4b respectively correspond to the two-time remote sensing image of the third embodiment;
FIG. 5a is a partial enlarged view of the interference area in the first embodiment;
FIG. 5b is a partial enlarged view of the interference area in the third embodiment;
FIG. 6a is a typical "pseudo-variation information" in the first embodiment;
fig. 6b is typical "pseudo-variation information" in the second embodiment;
fig. 6c is typical "pseudo-variation information" in the third embodiment;
7 a-7 f are qualitative result comparisons of multi-scale segmentation and change detection performed by various algorithms for embodiment one;
8 a-8 f are qualitative result comparisons of multi-scale segmentation and change detection performed by various algorithms for example two;
fig. 9 a-9 f are qualitative result comparisons of multi-scale segmentation and change detection performed by various algorithms on example three.
Detailed Description
The following description will explain embodiments of the present invention in further detail with reference to the accompanying drawings.
The invention designs a water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing images, which is characterized in that water body index features and vegetation index features are fused on the basis of Spectrum and Texture features to construct a mixed feature space STWV (Spectrum-Texture-NDWI-SAVI) for water body surrounding environment change detection, and then forward and reverse heterogeneity of ground object objects of the remote sensing images is constructed in the mixed feature space for change detection. Firstly, processing a superposed image of two time phases by adopting a super pixel generation algorithm SLIC (simple linear Iterative Cluster) to obtain a ground object, checking Forward heterogeneity of the ground object from the time phase 1 to the time phase 2, reversely detecting back heterogeneity of the ground object from the time phase 2 to the time phase 1, integrating Forward and reverse heterogeneous information to construct Forward and reverse heterogeneity of the ground object, then extracting a variable object of the two time phases by using an EM (expectation maximization) algorithm and a Bayesian minimum error rate theory, and finally excluding vegetation pseudo variable information to form a relatively accurate and robust variable detection result.
The invention relates to a water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing images, which is used for detecting and obtaining change information of the surrounding environment of a target water body, and in practical application, as shown in figure 1, the method specifically comprises the following steps:
And 2, obtaining each ground object corresponding to the surrounding environment of the target water body according to the t1 time phase remote sensing image and the t2 time phase remote sensing image, and then entering the step 3.
The step 2 specifically comprises the following steps:
and 2-1, adding the data of the same wave band in the t1 time phase remote sensing image and the t2 time phase remote sensing image in an equal proportion to obtain a superposed image, and then entering the step 2-2.
And 2-2, converting the superposed image into a feature space formed by a CIE Lab color space and position coordinates, updating the superposed image, and then entering the step 2-3.
And 2-3, performing multi-scale segmentation on the superposed image by adopting a super-pixel generation algorithm (SLIC) to obtain a segmentation result, and then entering the step 2-4.
And 2-4, respectively sleeving the segmentation result on the t1 time-phase remote sensing image and the t2 time-phase remote sensing image to obtain each ground object corresponding to the surrounding environment of the target water body.
Wherein the superpixel generation algorithm (SLIC) is described as follows:
1) convert the image to CIE Lab color space, then pixel xiThe Lab color value and the xy coordinate form a 5-dimensional feature vector Vi＝[l，a，b，x，y]，i＝1，2，...，N；
2) Setting initial seed points uniformly according to the given number of super pixels, pre-dividing the image into k subregions with the same size, and approximating the distance between adjacent seed points to
The size of the subarea is S;
3) and (3) reselecting a seed point: in order to avoid that the seed points fall on the contour boundary with larger gradient or the noise points and possibly influence the subsequent segmentation effect, the pixel points with the minimum gradient value in the n x n neighborhood range (n is generally 3) of the initial seed points are set as new seed points;
4) defining a difference measure D' between picture elements, measured by a color distance DcAnd a spatial distance dsConsists of the following components:
wherein N issIs the maximum space distance in the class and is defined as Ns＝S，NcFor the maximum color distance within a class, a constant m (value range [1, 40 ]) is set according to the picture contrast and the cluster class]) Then D' is:
5) and searching pixel points in the 2S-2S neighborhood from the seed point of the cluster center, so as to accelerate the convergence of the algorithm. Each pixel in the image can be searched by a plurality of seed points, the difference measurement between the pixel and the seed points is respectively calculated, the seed point with the minimum difference value is taken as the clustering center of the pixel, and then the pixels with the same clustering center are clustered into a class;
6) iterative clustering, for each cluster divided last time, calculating the characteristic vector mean value VcAs a new cluster center seed point and then relabeled as per step 5. Iteration is carried out until the error is converged, namely when the clustering center corresponding to each pixel point is not changed any more, the iteration is terminated, and a preliminary super-pixel segmentation image is generated;
to enhance connectivity, the preliminary superpixel segmentation image may have some problems: an excessively small superpixel size, a single superpixel being cut into a plurality of discontinuous superpixels, etc. The pixels in the super-pixels with undersize or discontinuous space are redistributed to the super-pixels with adjacent space by the region growing algorithm, so that the generated super-pixels are continuous space without being dispersed too much.
In the change detection process, only a single characteristic such as the spectral characteristic of a ground object is considered, so that higher precision is difficult to achieve, and the construction of a more stable and accurate change detection model by combining multiple characteristics is a hotspot direction of current research. The multi-feature fusion method for remote sensing image change detection generally considers the features of the fused surface object such as spectrum, texture and the like, and has a good effect on processing interference information in change detection. In the problem of detecting changes in the water surrounding environment, problems such as "pseudo-change information" due to seasonal changes in water quality, vegetation, and the like are still difficult to solve, but the index features (such as water body index and vegetation index) of the "pseudo-change information" are extremely obvious. Based on the characteristics that an LBP (local Binary Pattern) uniform mode has the advantages of rotation invariance, gray scale invariance and the like and the characteristics that a normalized Water body Index characteristic NDWI (normalized Difference WaterIndex) has excellent performance in a water body identification process and a Soil-adjusted vegetation Index SAVI (Soil-adjusted Vegetation Index) can reduce the influence of Soil background in vegetation type identification, a feature space STWV (Spectrum-Texture-NDWI-SAVI) is constructed by considering the spectral feature, the Texture feature LBP uniform mode image feature, the normalized Water body Index NDWI and the Soil-adjusted vegetation Index SAVI of a fused ground object. In order to fully utilize all the dimensional features in the feature space to describe the change situation of the ground features in the remote sensing image, the quantization levels of all the dimensional features in the feature space are set to be 256.
Step 3, aiming at the t1 time phase remote sensing image and the t2 time phase remote sensing image respectively, obtaining LBP uniform mode image characteristics, normalized water body indexes NDWI and soil regulation vegetation indexes corresponding to the remote sensing images, and constructing a mixed characteristic space corresponding to the remote sensing images according to the LBP uniform mode image characteristics, the normalized water body indexes NDWI and the soil regulation vegetation indexes SAVI, so as to obtain a mixed characteristic space STWV corresponding to each time phase remote sensing image; meanwhile, normalized vegetation indexes NDVI corresponding to the t1 time phase remote sensing image and the t2 time phase remote sensing image are obtained respectively, and then the step 4 is carried out.
The LBP uniform mode image characteristic is based on the original LBP characteristic, and the calculation formula of the original LBP is as follows:
wherein, P represents the number of pixels in a local window with a radius R and a pixel i as the center, giAnd gPAnd respectively representing the gray values of the central pixel and the neighborhood pixel of the local window.
The Uniform Pattern (Uniform Pattern) is defined as: a binary sequence having a number of transitions from 0 to 1 or from 1 to 0 (including transitions of leading and trailing elements) of not more than 2. Research finds that the proportion of the 8-bit binary sequence of the original LBP which meets the uniform pattern can reach more than 90%, so that the LBP uniform pattern image characteristics classify the binary sequence of the original LBP, each uniform pattern is classified into one type independently, and then other binary sequences which do not meet the uniform pattern are classified into one type, and the calculation formula is as follows:
in equation (6), the superscript riu2 indicates that the maximum hop count is 2, and that U (LBP) is satisfied in the P-neighborhoodP，R) The number of the uniform modes less than or equal to 2 is P (P-1) + 2.
In order to further reduce the influence of non-water body ground objects on the water body identification effect of the remote sensing image, scholars provide a normalized differential water body index NDWI on the basis of a normalized differential Vegetation index NDVI (normalized Difference vector index), and perform normalized Difference processing on a specific wave band to highlight the water body information in the remote sensing image, wherein the calculation formula is as follows:
wherein Green represents a Green band, and NIR represents a near infrared band. Generally speaking, the reflectivity of vegetation is the largest in a near infrared band, and the water body has almost no reflection in a near infrared wavelength range, so the NDWI adopts a ratio method based on a green band and the near infrared band to construct a water body index model, so that the vegetation information in an image can be restrained, and the water body information is highlighted. It should be noted that, after the NDWI calculation formula is subjected to stretching processing, the result value range is [ -1, 1], so that remote sensing data under different data sources or different imaging conditions can obtain NDWI water body index features with similar statistical characteristics, so that the water body extraction model can analyze and compare the NDWI water body index features, and water body information in the water body extraction model is extracted.
In order to reduce the influence of the soil background on the vegetation index, the soil regulation parameters are added into the normalized vegetation index NDVI, so that the soil regulation vegetation index SAVI has stronger capability of eliminating the soil influence and expressing vegetation information, and the calculation formula is as follows:
wherein Red represents a Red wave band, L is a soil adjusting parameter, the value range is [0, 1] (the higher the vegetation coverage is, the closer the vegetation coverage is to 0, and vice versa, the more the vegetation coverage is, the closer the vegetation coverage is to 1), and the function of the method is to explain the optical characteristic change of the soil background and adjust the sensitivity of NDVI to the soil background. When L is 0, SAVI NDVI indicates that vegetation coverage is very high, and the effect of the soil background can be ignored.
the common distance similarity includes an absolute value distance, a Euclidean distance, a chi-Square distance and the like, wherein various feature information of the absolute value distance and the Euclidean distance in a multi-feature space is subjected to equal-weight processing, the capability of constructing difference images and representing change information by different features cannot be reasonably reflected, and chi Square transformation CST (chi Square transformation) can comprehensively consider weight values of various features according to variances of two time-phase difference images on different features, so that the heterogeneity of a ground object constructed by the chi-Square transformation is more objective and complete.
To ensure data consistency, various features in the mixed multi-feature space STWV are first normalized to [0, 1]]Within the range, for the ground object in the two-time phase remote sensing image, the mean characteristic vector of the internal pixel is expressed as:
q is the dimension of the feature vector, and then the Forward heterogeneity of the first ground object is calculated through chi-square transformation
Wherein the content of the first and second substances,
the standard deviation of the difference value of the q-dimension characteristic of the ground object in two phases is adopted.
Meanwhile, according to various ground object objects corresponding to the surrounding environment of the target water body and mixed feature spaces STWV corresponding to various time phase remote sensing images, Backward heterogeneity of the various ground object objects is reversely constructed according to the t2 time phase remote sensing images to the t1 time phase remote sensing images through feature histograms of the various ground object objects in the mixed feature spaces, and then the step 5 is carried out.
Analyzing the feature histogram of the surface feature object can find that the appearance features of the surface feature object are consistent although the internal pixel feature distribution of the surface feature object is different. Considering that only the feature mean information of the ground object is considered and the feature distribution information of the ground object is not considered when the Forward heterogeneity is calculated, the problem of missing detection is possibly caused, the feature histogram of the ground object not only comprises the feature mean information of the ground object, but also comprises the feature distribution information of the ground object, and considering that the change detection result is more accurate by constructing the backsward heterogeneity by using the feature histogram of the ground object.
Log-Likelihood Ratio statistical (Log-likehood Ratio statistical), also called G statistical, is a non-parametric statistical method, which can be used to measure the similarity between two random variable sets without making any assumption on the distribution of the random variable sets, and the G statistical is used to measure the histogram similarity (Q is 1,2,., Q) of two-phase ground object on the Q-th dimensional feature, and is calculated as follows:
wherein Ht1And Ht2Characteristic histogram of ground object in two-time phase remote sensing image fqIs the q-dimension characteristic probability density function of the ground object, and L is the quantization level of the characteristic. For a single ground object, the cumulative probability value of the histogram is 1, and then:
the above equation then applies:
in order to fully utilize the information contained in all the features, the weighted average of the feature histogram similarity of each feature dimension of the ith ground object is calculated in the mixed multi-feature space STWV as the Backward heterogeneity
Wherein e (q) is the entropy of the difference image of the q-dimension feature of the two-time phase image, and can measure the information content contained in the difference image in the feature dimension, and the larger the information content is, the larger the difference of the ground object in the feature dimension is, the corresponding weight ω isqThe larger.
Calculating and obtaining Forward heterogeneity of the first ground object in the mixed multi-feature space STWV
And Backward heterogeneity
Then, the forward and reverse heterogeneity of the ground object can be obtained
And 5, classifying the ground object according to variation and non-variation according to the Forward heterogeneity and Backward heterogeneity of the ground object corresponding to the surrounding environment of the target water body by adopting a maximum mathematical expectation algorithm (EM), so as to obtain the varied ground object corresponding to the surrounding environment of the target water body, namely the variation information of the surrounding environment of the target water body, and then entering the step 6.
Calculating a forward and backward heterogeneity set of the ground object:
wherein n is the number of the ground object objects. In the two-time phase remote sensing image, the heterogeneity of the ground object which changes is large, and the heterogeneity of the ground object which does not change is small, so that the elements in D can be divided into two types, namely changed and unchanged. Assuming that D satisfies a Mixture Gaussian distribution GMM (Gaussian Mixture model) consisting of two Gaussian components, the density function can be expressed as:
wherein lcAnd luIs a change class and a not change class label, and is marked as l e { lc,luP (l) is the proportion of l-type elements in D, and p (l) is satisfiedc)+p(lu)＝1，
As a function of probability density, obey a gaussian distribution:
the maximum mathematical expectation EM (expectation maximization) algorithm is a maximum likelihood estimation method for solving probability model parameters, wherein the parameters p (l) and the mean value mu of a mixed Gaussian distribution model in the assumptionlStandard deviation σlThe estimation can be done with the EM algorithm. Firstly, clustering D into two classes by using a K-means algorithm, and then carrying out iterative updating according to the following formula until convergence:
calculated by Bayes' formula
Posterior probability of belonging to class l:
according to the principle of minimum error rate
Assign class label lcOr luAnd obtaining the change information of the corresponding ground object.
In the research process of the change detection problem of the surrounding environment of the water body, only the real change of the ground object type belongs to interested change information. Due to seasons, climate and the like, vegetation characteristics of the same vegetation coverage area (especially a farmland area) at different time phases may change, and 'pseudo change information' is generated and belongs to change information which is not interested in.
And 6, acquiring vegetation coverage information of each variable land feature object corresponding to the surrounding environment of the target water body according to the normalized vegetation indexes NDVI corresponding to the t1 time-phase remote sensing image and the t2 time-phase remote sensing image respectively, deleting vegetation pseudo-variation information in each variable land feature object corresponding to the surrounding environment of the target water body, and updating each variable land feature object corresponding to the surrounding environment of the target water body, namely updating variation information of the surrounding environment of the target water body.
Specifically, the vegetation coverage of different time phase pixels is calculated through a normalized vegetation index NDVI, then the proportion of the vegetation pixels in the variable-class ground object is estimated, the proportion of the vegetation pixels in the two time phase ground object is higher, the variable-class ground object is considered to be caused by the seasonal phase change of vegetation and belongs to vegetation pseudo-change information, and the method specifically comprises the following steps:
1. calculating the vegetation coverage F of the pixel ic：
Fc＝(NDVIi-NDVImin)/(NDVImax-NDVImin) (20)
Wherein NDVIminIs the minimum value of NDVI of the pixels in the remote sensing imagemaxIs the maximum value.
2. Calculating the proportion N of vegetation pixels in the changed objectf：
Wherein, FctAnd setting the threshold of the vegetation coverage of the vegetation pixel as 0.5, wherein the vegetation coverage of the pixel is greater than the threshold, which indicates that the pixel is the vegetation pixel, and n is the pixel number of the variable object.
3. If the ratio N of the vegetation pixel in the ground object of the two time phases corresponding to the change objectfAll above 0.65, the change object is considered to be composed ofPseudo-variation due to vegetation phenological changes.
The method for detecting the change of the surrounding environment of the water body based on the forward and reverse heterogeneity of the remote sensing image is applied to the practice, the experimental data of the method are the Level 1A-grade GF-1 PMS remote sensing images of the stone beam river water reservoir surrounding area (located in the northeast of Jiangsu province) in 2015 4 month and 2016 4 month provided by the China satellite resource application center, the panchromatic data with the spatial resolution of 2m and the multispectral data (the spectral range is 0.45 mu m-0.89 mu m) with the spatial resolution of 8m are included, and after the image fusion resampling based on ENVI 5.1 preprocessing and GS transformation, the spatial resolution of the fusion image is 2 m. The types of ground objects in the surrounding environment of the reservoir are various and the distribution condition is complex, the influence of interference information and 'pseudo change information' on change detection is large, and three groups of image data of the regional embodiment are selected to carry out a change detection experiment to comprehensively evaluate the effectiveness of the method. Wherein, fig. 2a, fig. 3a and fig. 4a are t1 time-phase remote sensing images corresponding to 2015 year 4, month 18 and day respectively for each embodiment, fig. 2b, fig. 3b and fig. 4b are t2 time-phase remote sensing images corresponding to 2016 year 4, month 4 and day 21 respectively for each embodiment, wherein the size of the two time-phase remote sensing images corresponding to one embodiment is 1166 × 881, the size of the two time-phase remote sensing images corresponding to two embodiments is 597 × 452, and the size of the two time-phase remote sensing images corresponding to three embodiments is 747 × 564 respectively; the change information of the experimental area mainly comprises the construction and the reconstruction of ground objects such as dams, houses and paddy fields, and the 'pseudo-change' such as the phenological change and the water quality change of vegetation coverage areas such as farmlands and forests.
The experimental data includes some interference information caused by the difference of imaging conditions, the change of illumination, etc., for example, fig. 5a is a partial enlarged view of an interference area in the first embodiment, and the ground object type is bare land; fig. 5b is a partially enlarged view of the interference area in the third embodiment, and the type of the ground object is a building.
The 'pseudo change information' in the experimental area has a great influence on the detection of the change of the water body surrounding environment, and fig. 6a is typical 'pseudo change information' in the first embodiment, and the brightness of the dam area changes; FIG. 6b is a typical "pseudo variation information" of the second embodiment, showing the variation of water quality; fig. 6c is typical "pseudo-variation information" in the third embodiment, and the vegetation is pseudo-varied.
The first embodiment and the second embodiment comprise the change information of the construction, the reconstruction and the like of the artificial dam in the boundary area of the reservoir, and also comprise the change information generated due to the reasons of water level fluctuation and the like, wherein the 'pseudo change information' mainly comes from the water quality change of the water body of the reservoir and the brightness change of the dam area; the main change information in the third embodiment is the change of land utilization information, including the construction of houses and paddy fields, and vegetation climate change and water quality change are main sources of 'pseudo change information'. In order to verify the effectiveness of the invention, a plurality of algorithms are adopted to detect the change of experimental data, and then the detection results are compared and analyzed.
To verify the validity of the algorithm, various algorithms are used to perform change detection on experimental data: 7 a-7 f are qualitative result comparisons of multi-scale segmentation and change detection performed by various algorithms for embodiment one; 8 a-8 f are qualitative result comparisons of multi-scale segmentation and change detection performed by various algorithms for example two; FIGS. 9 a-9 f are qualitative result comparisons of multi-scale segmentation and change detection performed by various algorithms on the third embodiment; wherein, a refers to a segmentation result of the two time phase images which are superposed and then are segmented by using SLIC; b refers to a reference detection result obtained by visual interpretation; c denotes a detection result of a Change Vector Analysis (CVA) facing the object; d denotes a variation detection result based on chi-square transformation (CST); e refers to the result of change detection based on Histogram (HIST); f refers to the detection result of the algorithm herein, and the white area in the figure represents the changed area and the black area represents the unchanged area.
In order to better verify the robustness and effectiveness of the algorithm in the detection of the change of the water body surrounding environment, the change detection result is subjected to quantitative comparative analysis. Evaluating the precision of the change detection result by using an error criterion of the number of pixels, wherein the false detection rate is the ratio of the number of the change pixels to the total number of the change pixels in the detection result, the missed detection rate is the ratio of the number of the change pixels which are missed to the total number of the unchanged pixels in the detection result, the correct rate is the ratio of the number of the pixels which are correctly detected to the total number of the pixels in the image, and the precision evaluation of the experimental results of the 4 change detection schemes in the upper section is shown in the following table 1:
TABLE 1
By combining table 1 with the graphs of fig. 7a to 7f, fig. 8a to 8f, and fig. 9a to 9f, it can be found that the missing detection conditions of the processing results of several algorithms are few, and the missing detection rate is controlled within an acceptable range, but the false detection conditions and the false detection rates are different: the object-oriented Change Vector Analysis (CVA) can detect most of change areas in two time-phase images by using various feature information of a multi-feature space, but the Euclidean distance is used for carrying out equal-weight processing on the feature information in the multi-feature space, so that the capability of representing change information by different features cannot be reasonably embodied, and the condition of false detection is serious; the change detection result based on chi-square transformation (CST) is greatly reduced in false detection compared with a change vector analysis method; the change detection result based on the Histogram (HIST) has a serious false detection phenomenon because the feature distribution information of the ground features in the two-time phase image is more seriously interfered by the phenomenon of 'same object different spectrum', so that the change detection result is influenced; in several groups of data of change detection experiments, the algorithm obtains better experiment effect and higher accuracy, and meanwhile, the false detection rate is controlled in a lower range, which shows that the algorithm has better detection capability for the change detection of the water body surrounding environment.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.
Claims (4)
1. A water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing images is used for detecting and obtaining change information of a target water body surrounding environment, and is characterized by comprising the following steps:
step 1, obtaining a t1 time phase remote sensing image and a t2 time phase remote sensing image corresponding to a target water body, carrying out unified registration on the two time phase remote sensing images, and then entering step 2; wherein phase t1 precedes phase t 2;
step 2, obtaining each ground object corresponding to the surrounding environment of the target water body according to the t1 time phase remote sensing image and the t2 time phase remote sensing image, and then entering step 3;
step 3, aiming at the t1 time phase remote sensing image and the t2 time phase remote sensing image respectively, obtaining LBP uniform mode image characteristics, normalized water body indexes NDWI and soil regulation vegetation indexes corresponding to the remote sensing images, constructing a mixed characteristic space corresponding to the remote sensing images according to the LBP uniform mode image characteristics, the normalized water body indexes NDWI and the soil regulation vegetation indexes SAVI, thus obtaining mixed characteristic spaces corresponding to the time phase remote sensing images respectively, and then entering step 4;
step 4, according to various ground object objects corresponding to the surrounding environment of the target water body and mixed feature spaces corresponding to various time phase remote sensing images respectively, comparing feature similarity of the ground object objects in the mixed feature spaces according to the t1 time phase remote sensing images to t2 time phase remote sensing images through distance similarity, and Forward constructing Forward heterogeneity of the ground object objects; meanwhile, reversely constructing backup heterogeneity of each ground object according to each ground object corresponding to the surrounding environment of the target water body and a mixed feature space corresponding to each time-phase remote sensing image from the t2 time-phase remote sensing image to the t1 time-phase remote sensing image through a feature histogram of each ground object in the mixed feature space, and then entering the step 5;
and 5, classifying the various ground object objects according to variation and non-variation by adopting a maximum mathematical expectation algorithm according to the Forward heterogeneity and the Backward heterogeneity of the various ground object objects corresponding to the surrounding environment of the target water body to obtain the variation information of the various varied ground object objects corresponding to the surrounding environment of the target water body, namely the variation information of the surrounding environment of the target water body.
2. The method for detecting changes in the surrounding environment of the water body based on forward and reverse heterogeneity of remote sensing images as claimed in claim 1, wherein: in the step 3, while the mixed feature space corresponding to each time phase remote sensing image is obtained, normalized vegetation indexes NDVI corresponding to the t1 time phase remote sensing image and the t2 time phase remote sensing image are also obtained;
step 6, after step 5, entering step 6;
and 6, acquiring vegetation coverage information of each variable land feature object corresponding to the surrounding environment of the target water body according to the normalized vegetation indexes NDVI corresponding to the t1 time-phase remote sensing image and the t2 time-phase remote sensing image respectively, deleting vegetation pseudo-variation information in each variable land feature object corresponding to the surrounding environment of the target water body, and updating each variable land feature object corresponding to the surrounding environment of the target water body, namely updating variation information of the surrounding environment of the target water body.
3. The method for detecting changes in the surrounding environment of the water body based on the forward and reverse heterogeneity of remote sensing images according to claim 1 or 2, wherein the step 2 comprises the following steps:
step 2-1, carrying out equal proportion addition on data of the same wave band in the t1 time phase remote sensing image and the t2 time phase remote sensing image to obtain a superposed image, and then entering step 2-2;
2-2, converting the superposed image into a feature space formed by a CIE Lab color space and position coordinates, updating the superposed image, and then entering the step 2-3;
step 2-3, performing multi-scale segmentation on the superposed image by adopting a super-pixel generation algorithm to obtain a segmentation result, and then entering step 2-4;
and 2-4, respectively sleeving the segmentation result on the t1 time-phase remote sensing image and the t2 time-phase remote sensing image to obtain each ground object corresponding to the surrounding environment of the target water body.
4. The method for detecting changes in the surrounding environment of the water body based on the forward and reverse heterogeneity of remote sensing images as claimed in claim 1, wherein the distance similarity is chi-square distance.
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 CN107330875A (en) | 2017-11-07 |
| CN107330875B true CN107330875B (en) | 2020-04-21 |
Family
ID=60192896
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710396859.2A Expired - Fee Related 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) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN108564002B (en) * | 2018-03-22 | 2020-08-21 | 中国科学院遥感与数字地球研究所 | Method and system for detecting time sequence change of remote sensing image |
| CN109657598B (en) * | 2018-12-13 | 2020-05-05 | 宁波大学 | Coastal wetland remote sensing classification method based on layering strategy |
| 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 |
| CN110135492B (en) * | 2019-05-13 | 2020-12-22 | 山东大学 | Equipment fault diagnosis and abnormality detection method and system based on multiple Gaussian 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 |
| TWI726396B (en) * | 2019-08-23 | 2021-05-01 | 經緯航太科技股份有限公司 | Environmental inspection system and method |
| CN113160183B (en) * | 2021-04-26 | 2022-06-17 | 山东深蓝智谱数字科技有限公司 | Hyperspectral data processing method, device and medium |
| CN113125358B (en) * | 2021-04-26 | 2022-03-25 | 山东深蓝智谱数字科技有限公司 | Hyperspectrum-based food pesticide residue detection 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 |
| 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 |
Citations (1)
| 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 |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 not_active Expired - Fee Related
Patent Citations (1)
| 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 |
Non-Patent Citations (1)
| Title |
|---|
| 基于超像素分割和多方法融合的SAR图像变化检测方法;张明哲等;《遥感技术与应用》;20160630;第31卷(第3期);481-487 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107330875A (en) | 2017-11-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107330875B (en) | Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image | |
| CN103971115B (en) | Automatic extraction method for newly-increased construction land image spots based on NDVI and PanTex index | |
| Zhang et al. | Object-oriented shadow detection and removal from urban high-resolution remote sensing images | |
| Li et al. | A review of remote sensing image classification techniques: The role of spatio-contextual information | |
| CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
| CN108830870B (en) | Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| Hoonhout et al. | An automated method for semantic classification of regions in coastal images | |
| CN103077515B (en) | Multi-spectral image building change detection method | |
| Su | Scale-variable region-merging for high resolution remote sensing image segmentation | |
| CN112529910B (en) | SAR image rapid superpixel merging and image segmentation method | |
| CN110309780A (en) | High resolution image houseclearing based on BFD-IGA-SVM model quickly supervises identification | |
| CN103839267A (en) | Building extracting method based on morphological building indexes | |
| CN115690086A (en) | Object-based high-resolution remote sensing image change detection method and system | |
| Devereux et al. | An efficient image segmentation algorithm for landscape analysis | |
| CN107992856A (en) | High score remote sensing building effects detection method under City scenarios | |
| Hao et al. | A hierarchical region-merging algorithm for 3-D segmentation of individual trees using UAV-LiDAR point clouds | |
| Dong et al. | Cloud detection method for high resolution remote sensing imagery based on the spectrum and texture of superpixels | |
| Robb et al. | A semi-automated method for mapping glacial geomorphology tested at Breiðamerkurjökull, Iceland | |
| CN116721121A (en) | Plant phenotype color image feature extraction method | |
| Zhou et al. | Shadow Detection on High-Resolution Digital Orthophoto Map (DOM) using Semantic Matching | |
| 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 | |
| Di et al. | Multi-scale segmentation of high resolution remote sensing images by integrating multiple features | |
| CN115984689A (en) | Multi-scale earth surface complexity feature extraction and land utilization segmentation method |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200421 |