CN102842130A - Method for detecting buildings and extracting number information from synthetic aperture radar image - Google Patents
Method for detecting buildings and extracting number information from synthetic aperture radar image Download PDFInfo
- Publication number
- CN102842130A CN102842130A CN2012102280578A CN201210228057A CN102842130A CN 102842130 A CN102842130 A CN 102842130A CN 2012102280578 A CN2012102280578 A CN 2012102280578A CN 201210228057 A CN201210228057 A CN 201210228057A CN 102842130 A CN102842130 A CN 102842130A
- Authority
- CN
- China
- Prior art keywords
- area
- foreground
- buildings
- image
- result
- 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
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000001514 detection method Methods 0.000 claims abstract description 28
- 238000012417 linear regression Methods 0.000 claims description 31
- 230000009466 transformation Effects 0.000 claims description 18
- 230000000877 morphologic effect Effects 0.000 claims description 16
- 230000007797 corrosion Effects 0.000 claims description 13
- 238000005260 corrosion Methods 0.000 claims description 13
- 230000001419 dependent effect Effects 0.000 claims description 12
- 230000011218 segmentation Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000010219 correlation analysis Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 abstract description 7
- 230000003287 optical effect Effects 0.000 description 12
- 230000008569 process Effects 0.000 description 9
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 9
- 238000003708 edge detection Methods 0.000 description 4
- 239000003673 groundwater Substances 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000001134 F-test Methods 0.000 description 2
- 238000000692 Student's t-test Methods 0.000 description 2
- 230000001427 coherent effect Effects 0.000 description 2
- 230000003628 erosive effect Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000012353 t test Methods 0.000 description 2
- 235000005956 Cosmos caudatus Nutrition 0.000 description 1
- 238000006424 Flood reaction Methods 0.000 description 1
- 241000282342 Martes americana Species 0.000 description 1
- 230000002844 continuous effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000010339 dilation Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a method for detecting buildings and extracting number information from a synthetic aperture radar image. The method ensures that a building detection result has a smaller false alarm rate and achieves connection with an adjacent bright characteristic from the same building by high and low threshold values, morphology open and close operations and removing small-area regions to the maximum, and furthermore, the method can improve extraction precision of building number information in a region and well separate very closed or even mutually-overlapped bright characteristics from different buildings. Moreover, the method adopts single width high-resolution SAR (Synthetic Aperture Radar) data which is the easily acquired and can save data cost compared with the data in other forms. Each sub-algorithm in the invention is very simple and efficient (mainly a threshold value operation and a mathematical morphology operation).Thus, the method can be conveniently applied to conditions with a high data volume and a large area region.
Description
Technical Field
The invention relates to the field of remote sensing information extraction and application, in particular to a method for detecting buildings and extracting the number information of the buildings from a high-resolution synthetic aperture radar image.
Background
Synthetic Aperture Radar (SAR) has become a ground observation remote sensing technology generally regarded by countries in the world with high resolution, all-weather and all-day large-area data acquisition capability. Advanced satellite-borne radar systems, such as the German radar satellite TerrraSAR-X, the Canadian radar satellite RADARSAT-2, and the Italy COSMO/SkyMet, have been able to provide image data at resolution on the order of meters, while advanced airborne SAR systems have been able to reach the level on the order of decimeters. At such resolution, the geometric and detail information of the urban building is clearly visible, and the extraction of the urban building information can be completely carried out based on the high-resolution SAR data source. The number of buildings in one or more spatial zones and the variation in this number have important instructive roles in many aspects such as city planning, land management, allocation of environmental protection resources, assessment of residential environment and related policy making.
Building detection is a key step for further extracting the number of buildings. Ideally, if each independent detection area in the detection result corresponds to an independent building, the number information of the buildings in the designated area can be obtained by simply counting the detection areas. Due to the SAR-specific side-view coherent imaging principle, a single building often features a high-resolution SAR image as a cluster of highlighted points or lines (caused by the shadowing and secondary scattering phenomena). The feature is obviously different from most natural ground objects and is the main foundation of SAR image building detection at present. To obtain an accurate number of buildings in an area, building detection requires connecting discrete light features of the same building while separating light features from different buildings that are very close to or even overlap each other.
To date, much research has been done on high resolution SAR building detection.
For example, Thiele et al extract the building boundary with an edge detection operator based on multi-view SAR data, and then obtain altitude information in combination with InSAR to determine whether a building exists. And the Sportouche and the like extract bright linear features by using an edge detection operator based on a high-resolution optical image and SAR data, detect building shadows by using a threshold value, judge the combination of the bright features and the shadows meeting the fixed position and the neighbor relation as building candidates, and finally fuse the information extracted by the optical image to perform comprehensive judgment and three-dimensional information extraction. Simeneto and the like obtain candidate areas of bright features of the SAR image building only by utilizing pixel classification and edge detection aiming at wide and large square buildings based on stereopair SAR data, then detect the bright features of L type, T type and X type by utilizing hough transformation, and finally fuse the bright feature detection results on different images of the stereopair to further reduce the false alarm rate of building detection. Xu et al describe the statistical distribution of buildings in high resolution SAR using Wishart distribution based on multi-view SAR data, and then detect the parallel structure of the buildings based on constant false alarm edge detection and hough transform to locate the buildings. Michaelsen et al first detect the apparent bright spot and line features based on the SAR image with less than 1m spatial resolution accuracy, then perform grouping and optimization by using the geosult system, and finally detect the building. Wegner et al, based on InSAR data and optical orthographic imagery, classify buildings and non-buildings using optical image features of buildings and SAR data features (mainly highlight line features caused by secondary scattering) under a Conditional Random Field (CRF) framework.
These efforts have several problems for building number information extraction:
(1) these efforts are directed to further extracting the three-dimensional geometric information of the building. In order to be able to extract precise building geometry information, the highest possible resolution and the highest possible amount of data are used, for example SAR data with a spatial resolution of 1m and below, multi-view SAR data, SAR stereopair data, InSAR data, optical and SAR data, optical and InSAR data. For the extraction of the building number information, such a strict data requirement is not necessary, and unnecessary cost waste is caused.
(2) Only relatively discrete areas between individual buildings are considered, and when applied to densely built areas, the accuracy of the building number information is rapidly degraded. In other words, it is not considered how to deal with the situation of bright features from different buildings that are close to or even overlapping each other.
(3) Most work only detects buildings of a particular type, such as wide rectangular buildings, buildings with L-shaped light features.
Therefore, the existing methods for detecting buildings and extracting number information from synthetic aperture radar images are not ideal.
Disclosure of Invention
The invention aims to provide a method for detecting buildings and extracting number information of the buildings from a synthetic aperture radar image aiming at the defects of the existing high-resolution SAR image building detection method.
In order to achieve the purpose of the invention, the method for detecting buildings and extracting number information from synthetic aperture radar images comprises the steps of obtaining high-resolution images through a synthetic aperture radar sensor and processing the high-resolution images by using a computer, wherein the aperture radar images are also called SAR images, and when the SAR images are processed by using the computer, a specific value can be selected as a threshold value in a normal pixel value range of the SAR images, wherein the method comprises the following specific steps:
step 1, opening a high-resolution SAR image through a high-resolution SAR image processing program installed in a computer;
step 2, selecting two thresholds of high and low in a normal pixel value range of the SAR image, respectively carrying out binarization processing on the opened SAR image by using a computer based on the two selected thresholds of high and low, and setting a building area as a foreground;
step 3, setting the foreground area in the low-threshold value binarization result obtained in the step 2 as a starting point, expanding the range in the foreground area in the high-threshold value binarization result, merging the foreground areas with connected or overlapped positions in the two results into a new foreground area, and removing the foreground area without the high-threshold value binarization result and the foreground area in the low-threshold value binarization result overlapped or connected with the foreground area;
step 4, combining the adjacent and close foreground areas in the result of the step 3, and closing a small gap;
step 5, separating the weakly connected regions in the result of the step 4, and eliminating small discrete points and peaks;
step 6, removing a foreground area with a smaller area in the result of the step 5;
step 7, setting more than five foreground areas representing real building areas as building area samples;
step 8, calculating the image characteristics of the selected building area sample by using a computer, and acquiring the number of real buildings in the building area;
step 9, selecting the next sample, and repeating the step 8 until the calculation of all samples is completed;
step 10, performing correlation analysis on the image characteristics of the sample and the number of real buildings by using the data sets calculated in the steps 7 to 9, and screening out an image characteristic set with high correlation with the number of the real buildings;
step 11, performing multivariate linear regression analysis by taking the number of real buildings as a dependent variable and the sample image feature subset as an independent variable to construct a linear regression equation;
step 12, evaluating the effectiveness of the linear regression equation in a computer by using a formula;
step 13, selecting different sample image feature subsets, and repeating the steps 11 to 12 until a linear regression equation with the best performance is obtained;
step 14, preliminarily determining the number of buildings in each foreground area in the result of the step 6 by using a linear regression equation;
step 15, dividing each foreground region in the result of the step 6 into independent regions with the number estimated by the linear regression equation to obtain a final building detection result;
and step 16, calculating the number information of the buildings in the designated area by using the building detection result of the step 15 and taking the number of the independent foreground areas in the designated area as the number of the buildings in the area through a computer.
Further, in the method for detecting buildings and extracting number information from the synthetic aperture radar image, the following method may be adopted in the step 16: and (4) calculating the number information of the buildings in the designated area through a computer by using the building detection result in the step (6) and the linear regression equation obtained in the step (13). Specifically, assuming that m foreground regions are included in the designated region, the number y of buildings in the entire region is calculated by the following formula:
yi=β0+β1x1,i+...+βjxj,i+....+βkxk,i
wherein x isj，iFor the j image feature, β, calculated from the i foreground regioniI =0, 1.. k is the coefficient of the regression equation, yiThe number of buildings in the ith foreground region calculated by using the regression equation.
Further, the method for detecting buildings and extracting number information from the synthetic aperture radar image, wherein the step 15 of dividing each foreground region in the result of the step 6 into independent small regions with the number estimated by the linear regression equation, comprises the following sub-steps:
the step 15 of dividing each foreground region in the result of the step 6 into independent small regions with the number estimated by the linear regression equation comprises the following substeps:
step 15.1, copying a foreground region into a small rectangle containing the region by using an SAR image processing program installed in a computer to form a small binary image (the foreground region and the background region take different values); repeatedly carrying out morphological corrosion operation on the foreground area, and marking a limit corrosion area and a background area;
step 15.2, calculating a distance transformation graph of the binary image;
step 15.3, calculating significance indexes of all marked areas;
step 15.4, sequencing all the marked areas from high to low according to the significance indexes by using computer software, and selecting the first N (N is the building number in the area calculated by the regression equation) marked areas in sequence;
step 15.5, with the marked area selected in the step 15.4 as an initial overflow area, performing watershed transformation on the distance transformation graph obtained in the step 15.2 to obtain a segmentation result of the foreground area;
step 15.6, replacing the corresponding foreground area in the result of the step 6 with the segmentation result obtained in the step 15.5;
and step 15.7, repeating the steps 15.1 to 15.6 until all foreground areas in the result of the step 6 are processed.
Further, in the method for detecting buildings and extracting number information from a synthetic aperture radar image, the SAR image processing program refers to computer software that can input and output the SAR image or change the value thereof and extract information.
Still further, the method for detecting buildings and extracting number information from synthetic aperture radar images as described above, wherein the SAR image processing program is software written based on an IDL language.
The invention has the following advantages and positive effects:
(1) by means of high and low thresholds, morphological opening and closing operations and removal of small-area areas, a small false alarm rate of a building detection result is guaranteed, and meanwhile adjacent bright features from the same building are connected to the maximum extent;
(2) the regression model is used for modeling the complex relationship between the regional characteristics of the building and the number of the actual buildings in the region, so that the extraction precision of the information of the number of the buildings in the region is improved;
(3) the morphological watershed transform is utilized to segment the large detection area into small areas which accord with the estimation number of the regression model, and bright features from different buildings which are very close to or even mutually overlapped are well separated. The result can be further used for extracting other building information (such as building position, shape, height and the like);
(4) for single high-resolution SAR data, the data is the most easily acquired data form at present, and compared with data in other forms, the data cost can be saved;
(5) the sub-algorithms in the invention are very simple and efficient (mainly threshold operation and mathematical morphology operation). This makes it easy to apply to large data volume and large area situations.
Drawings
FIG. 1 is a SAR urban image of the present invention with a resolution of 3 meters;
FIG. 2 is an image of the building detection result obtained by removing the foreground region with a smaller area in step 6 of the present invention;
FIG. 3 is a process image of the present invention step 15 segmenting a large foreground region into 4 independent regions;
FIG. 4 is an optical image corresponding to a foreground region of the present invention;
FIG. 5 is the foreground region image corresponding to the scene of FIG. 4 detected at step 6 of the present invention;
FIG. 6 is a resulting image of the invention step 15 slicing FIG. 5.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Referring to fig. 1 and fig. 2, fig. 1 is an image of an urban SAR region with a resolution of 3m, wherein most of the highlighted regions are buildings, and fig. 2 is a result obtained by removing the foreground regions with smaller areas in the result of step 5 in step 6. Referring again to fig. 3, fig. 3 illustrates a process of dividing a large foreground region into 4 (the number estimated by the linear regression equation) independent regions in step 15, where the process includes five steps a, b, c, d, and e, where a is a foreground region containing a plurality of buildings; b is a distance transformation image; c is marked limit corrosion area and background area (white part is mark); d is the top 4 most significant independent mark regions and background regions selected based on the significance index; e is a segmentation result obtained by running a watershed transform on the distance transform image b based on the marker image d. Referring to fig. 4, 5 and 6, fig. 4 is an optical image corresponding to the foreground region of the present invention; FIG. 5 is the foreground region corresponding to the scene of FIG. 4 detected at step 6 of the present invention; fig. 6 shows the result of slicing fig. 5 according to step 15 of the present invention.
The concrete implementation mode of the method for detecting the buildings and extracting the number information of the buildings, which is provided by the invention, is detailed as follows according to the steps:
step 1, opening a high-resolution SAR image through a high-resolution SAR image processing program installed in a computer; the high-resolution SAR image is obtained by using a Synthetic Aperture radar (Synthetic Aperture radar) sensor, the sensor can be arranged on an airplane platform, a satellite platform or an airship platform, and the like, and when a computer is used for processing the SAR image, a specific value can be selected from a normal pixel value range of the SAR image as a threshold value.
And 2, selecting two thresholds of high and low values in the normal pixel value range of the SAR image, and respectively carrying out binarization processing on the opened SAR image by using a computer based on the two selected thresholds of high and low values to set a building area as a foreground.
Here, the binarization processing procedure is as follows:
let the value of an image pixel be x, which has a value range [ xmin, xmax ]. The threshold is a specific value selected within the range of values of x. An image with a pixel value range of [ xmin, xmax ] can be changed into a binary image with a pixel value range of {0,1} by using a threshold, which is specifically realized as follows: the specific value of each image pixel is compared with the threshold, and if the pixel value is greater than the threshold, the pixel value is assigned to 1 (or 0), otherwise, the pixel value is assigned to 0 (or 1).
In specific implementation, the high and low thresholds may be obtained according to experience or training sample data, or may be obtained by directly and interactively selecting a threshold on an image. The direct interaction threshold value obtaining mode is more consistent with most situations of practical application. Here, a high threshold value detects a part of building bright areas with high reliability, and the detection result has a small false alarm; the lower threshold detects most of the building bright areas, but the detection result contains a large number of false alarms.
And 3, setting the foreground area in the low-threshold binarization result obtained in the step 2 as a starting point, expanding the range in the foreground area in the high-threshold binarization result, merging the foreground areas with connected or overlapped positions in the two results into a new foreground area, and removing the foreground area without the high-threshold binarization result and the foreground area in the low-threshold binarization result overlapped or connected with the foreground area.
In specific implementation, this step can be performed by using a morphological reconstruction method (using the high-threshold binarization result as a labeled map and the low-threshold binarization result as a masked map). The morphological reconstruction method is a well-known technique in the field of images and will not be described herein.
And 4, combining the adjacent and close foreground areas in the result of the step 3, and closing the small gap.
In specific implementation, the type and size of the structural primitive can be selected according to requirements by using a morphological closing operator. For example, a morphological closing operation is performed using a rectangular structuring element of size 5 x 5, connecting foreground regions within a distance of 2 pixels, and filling gaps and holes smaller than the structuring element.
And 5, separating the weakly connected regions in the result of the step 4, and eliminating small discrete points and peaks.
In specific implementation, the type and size of the structural primitive can be selected according to requirements by using a morphological opening operator. For example, a rectangular structural unit with a size of 3 × 3 is used to perform a morphological opening operation, separate adjacent foreground regions where only a single pixel is connected, and remove discrete points and peaks smaller than the structural unit.
And 6, removing the foreground area with smaller area in the result of the step 5.
Due to the SAR coherent imaging principle and the complexity of the ground feature situation, the ground features such as bare land and grassland woodland can also generate partial bright spots which are relatively isolated and have small areas relative to the building bright areas. A large number of false alarm conditions can be removed by removing foreground regions that are smaller than a certain area threshold.
And 7, setting more than five foreground areas representing the real building areas as building area samples.
In particular, the number of the building area samples is usually a large number. The minimum number of samples required is related to the number of parameters of the regression equation in step 11, i.e. if the number of parameters of the regression equation is 10, the number of samples is at least 10, and if the number of samples is less than 10, the parameters of the regression equation are difficult to estimate efficiently. In fact, to make regression equation parameter estimation more accurate, the number of building area samples is much larger than the number of parameters of the regression equation.
And 8, calculating the image characteristics of the selected building area sample by using a computer, and acquiring the number of real buildings in the building area.
In specific implementation, the outline of the sample region can be firstly mapped onto a high-resolution optical image of the corresponding region, and the number of real buildings in the region can be obtained by identification and counting on the optical image. The image characteristics of the area sample have very many kinds, and the invention preferentially uses the characteristics of area, density, length of long and short axes of the area contour fitting ellipse and the like.
And 9, selecting the next sample, and repeating the step 8 until the calculation of all samples is completed.
And step 10, performing correlation analysis on the image characteristics of the sample and the number of real buildings by using the data sets calculated in the steps 7 to 9, and screening out an image characteristic set with high correlation with the number of the real buildings.
In order to ensure the effectiveness of the regression equation constructed in the following steps, the image features with high correlation with the number of real buildings are screened out as much as possible. For example, the correlation can be evaluated using the common Pearson product-Difference correlation coefficient (i.e., the covariance of two variables divided by the standard deviation of the two variables).
And 11, performing multivariate linear regression analysis by taking the number of the real buildings as dependent variables and the sample image feature subset as independent variables to construct a linear regression equation.
The number of real buildings is represented by y, xiRepresenting the ith image feature, the k-linear regression equation can be expressed as:
y=β0+β1x1+...+βixi+....+βkxkformula (1)
Wherein beta is0,β1,β2,..,βi,..,βkFor the regression coefficients, a least squares method may be used based on the sample data set { y, x1，...，xi，...，xk}j=1，...，NAnd (6) obtaining the estimation. The sample data set is composed of the real building number and the image characteristic data of each building sample region calculated in the steps 7 to 10.
And step 12, evaluating the effectiveness of the linear regression equation in a computer by using a formula.
An important step before using the linear regression equation is to test the goodness of fit of the model and the statistical significance of the estimated model parameters and equations. The goodness of fit of the test model may use a modified R-square coefficient:
Here, the
Is the overall mean of the dependent variable y sample data,
dependent variable value, y, estimated for the regression equation for the ith independent variable sampleiIs the ith dependent variable sample value, n is the number of sample data, and k is the number of independent variables. The closer this coefficient is to 1, the better the model fits.
The statistical significance of the parameters of the estimated regression equation can be tested by using a t-test to construct the following statistics:
Wherein, biIs to the ith regression equation parameter betaiAn estimated value of (c), SE (b)i) Is biN is the number of sample data, k is the number of independent variables, and α is the significance level. the purpose of the t-test is to detect whether the dependent variable is affected by each of the independent variables.
The statistical significance of the regression equation was tested using the F-test to construct the following statistics:
Here, the
Is the overall mean of the dependent variable y sample data,
the dependent variable value estimated for the ith independent variable sample for the regression equation is taken, n is the number of sample data, k is the number of independent variables, and alpha is the significance level. The purpose of the F-test is to detect the linear relationship of the dependent variable to the independent variable.
And step 13, selecting different sample image feature subsets, and repeating the steps 11 to 12 until a linear regression equation with the best performance is obtained.
And step 14, preliminarily determining the number of buildings in each foreground area in the result of the step 6 by using a linear regression equation.
In specific implementation, the following equation regression equation is used for calculating the number of buildings in each foreground region:
y=β0+β1x1+...+βixi+....+βkxk
here beta0,β1,β2，...,βi，...,βkFor the regression equation coefficient, x, obtained in step 13iThe ith image feature of the current foreground region.
And 15, dividing each foreground region in the result of the step 6 into independent regions with the number estimated by the linear regression equation to obtain a final building detection result.
And step 16, calculating the number information of the buildings in the designated area through a computer.
There are two scenarios:
in the first scheme, the number information of buildings in the specified area is calculated by using the building detection result in the step 6 and the linear regression equation obtained in the step 13. Specifically, assuming that m foreground regions are included in the designated region, the number y of buildings in the entire region is calculated by the following formula:
yi=β0+β1x1，i+β2x2，i+β3x3，i
wherein x isj，iFor the j image feature, β, calculated from the i foreground regioniWhere i is 0,1, 2, and 3 are regression equation coefficients (coefficients of the optimal regression equation obtained in step 13), and y isiThe number of buildings in the ith foreground region calculated by using the regression equation.
And step two, using the building detection result in the step 15, and taking the number of the independent foreground areas in the designated area as the building number in the area.
Wherein the first scheme is a preferred scheme.
The step 15 of dividing each foreground region in the result of the step 6 into independent small regions with the number estimated by the linear regression equation comprises the following substeps:
step 15.1, a foreground region is copied to a small rectangle containing the region by using an SAR image processing program installed in a computer to form a small binary image (the foreground region and the background region take different values). Repeatedly carrying out morphological corrosion operation on the foreground area, and marking a limit corrosion area and a background area;
in specific implementation, a structural element B with the radius r is adopted to carry out corrosion operation on the foreground area A. From the mathematical morphology theory, the erosion of the region A by the structural element B is expressed as
The corrosion has the effect of shrinking the image area, for a given foreground area A, the structural element B is used for repeatedly carrying out corrosion operation, a layer with the thickness of r is continuously stripped, disconnected areas can be continuously generated along with the continuous effect of the corrosion, and meanwhile, certain areas can gradually disappear. The last step before disappearance of one connected component is called the final connected component, while the union of all the final connected components is called the limit erosion zone of relative radius r. This step marks all the final connected components within the small rectangle, as well as the background area.
Step 15.2, calculating a distance transformation graph of the binary image;
the distance transformation image in the step is obtained by adopting a method that I represents the current binary image, and an initial distance transformation image d (I) is obtained by calculating the distance between each pixel in the binary image and the nearest background pixel, wherein the distance transformation image finally used in the step is L-d (I), and L is a constant value. If the distance transformation value is not negative, L needs to select a value greater than or equal to max (d (i)). The distance measure used for calculating the distance can be the commonly used Euclidean distance, chessboard distance, urban area distance, Mahalanobis distance and the like.
Step 15.3, calculating significance indexes of all marked areas;
considering the distance transformed map in step 15.2 as a three-dimensional terrain (the x and y coordinates of the pixels mark their geographical location and the value of the pixels represents the height of the terrain), the different regions marked in step 15.1 can be considered as different basins in the terrain. There are different ways to assess the significance of the marked area. The simplest way of evaluating is according to the height of the basin bottom of the basin corresponding to the region, namely, the lower the height of the basin bottom, the more remarkable the basin is. The evaluation does not consider the relative relationship among the basins, so that the evaluation values of the significance of the basins in the low-terrain areas are high, and the evaluation values of the significance of the basins in the high-terrain areas are low. In the invention, the significance of the basin is recommended to be evaluated by using a kinetic index.
Dynamic definition of basin M: and selecting a path with the lowest highest point height from all paths from the bottom of the basin M to the bottom of any lower basin, and subtracting the basin bottom height of the basin M from the highest point height of the path, wherein the difference is the dynamic value of the basin M. It can be seen by definition that the dynamics are not a local significance criterion and may be related to the nature of the basins (not just the adjacent basins) in a large range around it. The invention proposes to calculate the dynamic values of the basins when the flows of the two basins intersect during the overflow. First, the overflow process is explained, and the process that the groundwater level rises continuously until the whole terrain is submerged is called the overflow process, assuming that the lowest points of basins corresponding to all the marked areas of the terrain are all drilled through and connected with the same groundwater source. The method for calculating the basin dynamics value when the water of two different basins intersects in the overflow process is as follows: considering that the basin with high terrain is swallowed by the basin with low terrain, and simultaneously calculating the dynamic value of the basin with higher terrain, which is equal to the height h of the current water level minus the height h (M) of the bottom of the basin, namely dy (M) = h-h (M). Thus, the global minimum basin will not obtain the basin dynamics value, and can be made to be a maximum value. In addition, when the water of two basins with the same terrain intersects, the above algorithm cannot judge which basin the dynamic value should be calculated for, and one basin can be selected to calculate the dynamic value.
Step 15.4, sequencing all the marked areas from high to low according to the significance indexes by using computer software, and selecting the first N (N is the building number in the area calculated by the regression equation) marked areas in sequence;
step 15.5, with the marked area selected in the step 15.4 as an initial overflow area, performing watershed transformation on the distance transformation graph obtained in the step 15.2 to obtain a segmentation result of the foreground area;
considering the distance transformed map in step 15.2 as a three-dimensional terrain (the x and y coordinates of the pixels mark its geographical position and the value of the pixels represents the height of the terrain), the marked areas selected in step 15.4 are considered as different basins in the terrain and it is assumed that only these marked areas are the initial overflow areas (i.e. the lowest points of the corresponding basins are perforated and connected to the same groundwater supply). Basin change is based on an overflow process performed on this three-dimensional terrain, and as groundwater gradually rises, water gradually floods the basin and rises toward the ridge. Since the water in all basins rises synchronously, the intersection of the water flows of different basins is just the ridge point. And building a dam at each intersection to prevent intersection of different water flows. Thus, the water level is continuously increased until only the dam is on the water surface. It can be seen that the dike is closed and the entire terrain area is divided into different patch areas. The number of segmentation regions is equal to the number of initial marking regions. The divided areas are the division results obtained in the step;
step 15.6, replacing the corresponding foreground area in the result of the step 6 with the segmentation result obtained in the step 15.5;
and step 15.7, repeating the steps 15.1 to 15.6 until all foreground areas in the result of the step 6 are processed.
The following examples are further provided in connection with the present disclosure:
in this embodiment, the technical scheme provided by the invention is applied to the terrasaar-X intensity data based on the spatial resolution of 1.25m × 1.25m to extract the building number information in the region of the marten city, north of Hu. The whole process is as follows:
step (1) opens a high resolution SAR image with size 20000 × 20000 pixels. The image was taken by terrasaar-X at a location in wuhan city, china, and had an image spatial resolution of 1.25m by 1.25m (obtained by oversampling at a spatial resolution of 3m by 3 m).
Step (2), determining that the two thresholds are 700 and 400 respectively by utilizing a mode of interactively selecting the threshold and observing on the image, and respectively carrying out binarization processing on the SAR image, wherein a region larger than the threshold is a foreground region;
and (3) fusing two thresholding results by using a morphological reconstruction method, namely performing morphological dilation reconstruction by taking the low-threshold (400) binarization result image as a mask image and the high-threshold (700) binarization result image as a marked image. Finally, foreground regions with connected or overlapped positions in the two results are merged into a new foreground region, and the foreground regions in the low-threshold binarization results without overlapping or connecting the foreground regions in the high-threshold binarization results are removed;
and (4) performing morphological closing operation on the result of the step (3) by adopting 5-by-5 rectangular structural elements, merging adjacent and close foreground regions, and closing small gaps.
And (5) performing morphological opening operation on the result of the step (4) by adopting a 3-by-3 rectangular structural element, separating weakly connected regions, and eliminating small discrete points and peaks.
Removing the foreground area with the area smaller than 25 pixels in the result of the step (5);
step (7), comparing the optical reference image, and selecting 48 independent foreground areas from the result of the step (6) as a building area sample;
step (8), 5 image characteristics (area, mass density, length of a short axis of a fitting ellipse, length of a long axis of the fitting ellipse and perimeter) of the selected building area sample are calculated; mapping the sample area to a corresponding position of an optical reference map, and obtaining the number of real buildings in the area through manual identification and counting;
and (9) selecting the next sample, and repeating the step (8) until the calculation of all 48 samples is completed. Store all data results as a data set { y, x1，...，xi，...，x5}j＝1，...，48Wherein y is the number of real buildings in the sample of the building area, xiIs the ith image feature.
And (10) respectively calculating a Pearson product-difference correlation coefficient (namely, the covariance of the two variables is divided by the standard deviation of the two variables) of each image feature and the number of real buildings by using the data set in the step (9). The absolute value of the correlation coefficient between 3 characteristics (area, density, fitting ellipse minor axis length) and the number of real buildings is found to be more than 0.8, and the absolute value of the correlation coefficient between other characteristics and the number of real buildings is found to be less than 0.6. Therefore, the characteristics of 'area', 'density', 'fitting ellipse minor axis length' are selected as image feature sets with high correlation with the number of real buildings, and the sample data sets are updated to be { y, x1,x2,x3}j＝1，...，48。
Step (11), based on the sample data set in step (10), constructing the following 3-variable linear regression equation:
y=β0+β1x1+β2x2+β3x3
wherein y is the number of buildings in a certain area, x1,x2,x3Respectively being a regionThe characteristics of area, density and fitting ellipse minor axis length. Estimating the coefficient of the regression equation to be beta by using least square algorithm0=0.4276424，β1=0.0004744，β2=0.0067597，β3=0.0197668。
Step (12), calculating F statistic equal to 71.47016 according to formula (4), and searching F distribution table to obtain F0.05(3,44) = 2.82. Since 71.47016 is much larger than F0.05(3,44) it can be judged that the overall linear relationship is effective and significant; calculating relevant t statistics of 3 coefficients of the regression equation according to the formula (3) respectively to be t1=4.148, t2=3.022 and t3=2.466, and searching a t distribution table to obtain t0.025(44) = 2.01. Since the t statistic of each coefficient is greater than t0.025(44) Then, it can be determined that each independent variable and dependent variable have significant linear relationship; the corrected R square coefficient calculated according to the formula (2) is 0.8297, which is relatively close to 1, and shows that the constructed autoregressive equation can be used for largely fitting the relationship between the independent variable and the dependent variable.
Step (13), then constructing y = β, respectively0+β1x1+β2x2，y=β0+β2x2+β3x3，y=β0+β1x1+β3x3，y=β0+β1x1，y=β0+β2x2，y=β0+β3x3The autoregressive equation was evaluated in six cases, and the performance was evaluated. Comparing the performances of the autoregressive equations in seven different cases (plus one of the cases constructed in the previous step) shows that the regression equation constructed in step (11) has the best performance, and therefore, the regression equation is selected for subsequent application.
Step (14), calculating three characteristics of 'area', 'compactness' and 'fitting ellipse minor axis length' of each independent foreground region in the result of the step (6), and substituting the characteristics into the linear regression equation obtained in the step (13) to obtain the estimated building number in the foreground region;
and (15) carrying out the following operation on each independent foreground area in the result of the step (6):
and step 15.1), copying the foreground region into a small rectangle containing the region to form a small binary image (the foreground region takes 1, and the background region takes 0). Repeatedly carrying out morphological corrosion operation on the foreground region by using the 3 x 3 rectangular structural element, and marking a limit corrosion region and a background region;
step 15.2), calculating a distance transformation graph of the binary image in the step 15.1), wherein the distance measure adopts Euclidean distance;
step 15.3), calculating the basin dynamics indexes of all the marked areas in the step 15.1);
step 15.4), sorting the marked regions from high to low according to the basin dynamics indexes, and selecting the top N (N is the nearest integer value of the building number in the region calculated in the step (14) by the regression equation) marked regions in order;
step 15.5), taking the marked area selected in the step 15.4) as an initial overflow area, and operating watershed transformation on the distance transformation graph obtained in the step 15.2) to obtain a segmentation result of the foreground area;
step 15.6), replacing the corresponding foreground area in the result of the step 6 with the segmentation result obtained in the step 15.5);
and (16) circling 8 irregular areas on the opened high-resolution SAR image (a place in Wuhan City in China shot by Terras SAR-X), and calculating the number information of buildings in the designated area by adopting two different schemes.
And (3) calculating the number information of the buildings in the specified area by using the building detection result in the step (6) and the linear regression equation obtained in the step (13). Specifically, assuming that m foreground regions are included in the designated region, the number y of buildings in the entire region is calculated by the following formula:
yi=β0+β1x1，i+β2x2，i+β3x3，i
wherein x isj，iFor the j image feature, β, calculated from the i foreground regioniWhere i is 0,1, 2, and 3 are regression equation coefficients (coefficients of the optimal regression equation obtained in step (13)), and y isiThe number of buildings in the ith foreground region calculated by using the regression equation.
And (II) taking the number of independent foreground areas in the specified area as the number of buildings in the area by using the building detection result in the step (15).
Wherein the first scheme is a preferred scheme.
Table 1 shows the result of extracting the building data information in the circled 8 irregular regions. The average value of the absolute errors of the extracted building number information of 8 areas in the first scheme of the step (16) is 0.925. When the second scheme in the step (16) is adopted to extract the number information of the buildings in the area, the performance indexes of the absolute errors are different because the number of the buildings estimated by the regression equation needs to be rounded, and the average value of the absolute errors is 1.125.
Table 1: detection result of building number information in 8 test areas
Of course, the above is only a specific application example of the invention, and other embodiments of the invention are also possible, and all technical solutions formed by using equivalent substitutions or equivalent changes are within the scope of the invention as claimed.
Claims (5)
1. A method for detecting buildings and extracting number information from synthetic aperture radar images comprises the steps of obtaining SAR images through a synthetic aperture radar sensor and processing the SAR images by using a computer, and selecting specific values as threshold values in a normal pixel value range of the SAR images when the SAR images are processed by using the computer, and is characterized in that: the method comprises the following specific steps:
step 1, opening a high-resolution SAR image through a high-resolution SAR image processing program installed in a computer;
step 2, selecting two thresholds of high and low in a normal pixel value range of the SAR image, respectively carrying out binarization processing on the opened SAR image by using a computer based on the two selected thresholds of high and low, and setting a building area as a foreground;
step 3, setting the foreground area in the low-threshold value binarization result obtained in the step 2 as a starting point, expanding the range in the foreground area in the high-threshold value binarization result, merging the foreground areas with connected or overlapped positions in the two results into a new foreground area, and removing the foreground area without the high-threshold value binarization result and the foreground area in the low-threshold value binarization result overlapped or connected with the foreground area;
step 4, combining the adjacent and close foreground areas in the result of the step 3, and closing a small gap;
step 5, separating the weakly connected regions in the result of the step 4, and eliminating small discrete points and peaks;
step 6, removing a foreground area with a smaller area in the result of the step 5;
step 7, setting more than five foreground areas representing real building areas as building area samples;
step 8, calculating the image characteristics of the selected building area sample by using a computer, and acquiring the number of real buildings in the building area;
step 9, selecting the next sample, and repeating the step 8 until the calculation of all samples is completed;
step 10, performing correlation analysis on the image characteristics of the sample and the number of real buildings by using the data sets calculated in the steps 7 to 9, and screening out an image characteristic set with high correlation with the number of the real buildings;
step 11, performing multivariate linear regression analysis by taking the number of real buildings as a dependent variable and the sample image feature subset as an independent variable to construct a linear regression equation;
step 12, evaluating the effectiveness of the linear regression equation in a computer by using a formula;
step 13, selecting different sample image feature subsets, and repeating the steps 11 to 12 until a linear regression equation with the best performance is obtained;
step 14, preliminarily determining the number of buildings in each foreground area in the result of the step 6 by using a linear regression equation;
step 15, dividing each foreground region in the result of the step 6 into independent regions with the number estimated by the linear regression equation to obtain a final building detection result;
and step 16, calculating the number information of the buildings in the designated area by using the building detection result of the step 15 and taking the number of the independent foreground areas in the designated area as the number of the buildings in the area through a computer.
2. The method for detecting buildings and extracting number information from synthetic aperture radar images according to claim 1, wherein: the step 16 may also adopt the following method:
calculating the number information of the buildings in the designated area by the computer by using the building detection result of the step 6 and the linear regression equation obtained in the step 13, wherein the number information of the buildings in the designated area is assumed to be included in the designated area as followsmThe number of buildings in the whole foreground areayCalculated from the following formula:
wherein,x j,i is composed ofiCalculated in the individual foreground region j The characteristics of the image are determined by the characteristics of the image,β i ,i=0,1,...,kin order to be the coefficients of the regression equation,y i is calculated as the first using a regression equation iThe number of buildings in the individual foreground areas.
3. Method for detecting buildings and extracting number information from synthetic aperture radar images according to claim 1 or 2, characterized in that: the step 15 of dividing each foreground region in the result of the step 6 into independent small regions with the number estimated by the linear regression equation comprises the following substeps:
step 15.1, copying a foreground region into a small rectangle containing the region by using an SAR image processing program installed in a computer to form a small binary image (the foreground region and the background region take different values); repeatedly carrying out morphological corrosion operation on the foreground area, and marking a limit corrosion area and a background area;
step 15.2, calculating a distance transformation graph of the binary image;
step 15.3, calculating significance indexes of all marked areas;
step 15.4, sequencing all the marked areas from high to low according to the significance indexes by using computer software, and selecting the first N (N is the building number in the area calculated by the regression equation) marked areas in sequence;
step 15.5, with the marked area selected in the step 15.4 as an initial overflow area, performing watershed transformation on the distance transformation graph obtained in the step 15.2 to obtain a segmentation result of the foreground area;
step 15.6, replacing the corresponding foreground area in the result of the step 6 with the segmentation result obtained in the step 15.5;
and step 15.7, repeating the steps 15.1 to 15.6 until all foreground areas in the result of the step 6 are processed.
4. The method of detecting buildings and extracting number information from synthetic aperture radar images according to claim 3, wherein: the SAR image processing program refers to computer software which can carry out input and output on the SAR image or change the value of the SAR image and extract information.
5. Method for detecting buildings and extracting number information from synthetic aperture radar images according to claim 4, characterized in that: the SAR image processing program is software written based on IDL language.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210228057.8A CN102842130B (en) | 2012-07-04 | 2012-07-04 | Method for detecting buildings and extracting number information from synthetic aperture radar image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210228057.8A CN102842130B (en) | 2012-07-04 | 2012-07-04 | Method for detecting buildings and extracting number information from synthetic aperture radar image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102842130A true CN102842130A (en) | 2012-12-26 |
| CN102842130B CN102842130B (en) | 2015-07-08 |
Family
ID=47369451
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210228057.8A Expired - Fee Related CN102842130B (en) | 2012-07-04 | 2012-07-04 | Method for detecting buildings and extracting number information from synthetic aperture radar image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102842130B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105023254A (en) * | 2015-07-22 | 2015-11-04 | 北京航空航天大学 | Height reconstruction method of synthetic aperture radar image |
| CN105740826A (en) * | 2016-02-02 | 2016-07-06 | 大连楼兰科技股份有限公司 | Lane mark binaryzation detection method based on dual scales |
| CN107122728A (en) * | 2017-04-21 | 2017-09-01 | 西安电子科技大学 | Building method is detected in SAR image based on evidence theory algorithm |
| CN107632001A (en) * | 2017-08-08 | 2018-01-26 | 西安电子科技大学 | Based on non-homogeneous pulse modulated offshore spilled oil monitoring method |
| CN108376407A (en) * | 2018-02-05 | 2018-08-07 | 李刚毅 | Hot-zone object aggregation detection method and system |
| CN108388845A (en) * | 2018-02-05 | 2018-08-10 | 李刚毅 | Method for checking object and system |
| CN110688923A (en) * | 2019-09-19 | 2020-01-14 | 中国电子科技集团公司第二十九研究所 | Sentinel 1A SAR data-based urban inland inundation risk area extraction method |
| CN110930449A (en) * | 2019-11-27 | 2020-03-27 | 上海眼控科技股份有限公司 | Prediction result evaluation method and device, computer equipment and storage medium |
| CN110942425A (en) * | 2019-11-26 | 2020-03-31 | 贵州师范学院 | Reconstruction method and reconstruction system of super-resolution image and electronic equipment |
| CN115089839A (en) * | 2022-08-25 | 2022-09-23 | 柏斯速眠科技(深圳)有限公司 | Head detection method and system and control method and system of sleep-assisting device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100246890A1 (en) * | 2009-03-26 | 2010-09-30 | Microsoft Corporation | Detection of objects in images |
| CN102129559A (en) * | 2011-04-22 | 2011-07-20 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image object detection method based on Primal Sketch algorithm |
-
2012
- 2012-07-04 CN CN201210228057.8A patent/CN102842130B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100246890A1 (en) * | 2009-03-26 | 2010-09-30 | Microsoft Corporation | Detection of objects in images |
| CN102129559A (en) * | 2011-04-22 | 2011-07-20 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image object detection method based on Primal Sketch algorithm |
Non-Patent Citations (2)
| Title |
|---|
| SONKE MULLER等: "ROBUST BUILDING DETECTION IN AERIAL IMAGES", 《2012 FOURTH INTERNATIONAL CONFERENCE ON COMPUTATIONAL AND INFORMATION SCIENCES》, 30 August 2005 (2005-08-30), pages 143 - 148 * |
| 崔建军等: "基于边缘检测算法的LiDAR数据建筑物提取", 《测绘科学技术学报》, vol. 25, no. 2, 15 April 2008 (2008-04-15), pages 98 - 100 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105023254A (en) * | 2015-07-22 | 2015-11-04 | 北京航空航天大学 | Height reconstruction method of synthetic aperture radar image |
| CN105740826A (en) * | 2016-02-02 | 2016-07-06 | 大连楼兰科技股份有限公司 | Lane mark binaryzation detection method based on dual scales |
| CN107122728A (en) * | 2017-04-21 | 2017-09-01 | 西安电子科技大学 | Building method is detected in SAR image based on evidence theory algorithm |
| CN107122728B (en) * | 2017-04-21 | 2019-10-11 | 西安电子科技大学 | Building method is detected in SAR image based on evidence theory algorithm |
| CN107632001A (en) * | 2017-08-08 | 2018-01-26 | 西安电子科技大学 | Based on non-homogeneous pulse modulated offshore spilled oil monitoring method |
| CN108376407A (en) * | 2018-02-05 | 2018-08-07 | 李刚毅 | Hot-zone object aggregation detection method and system |
| CN108388845A (en) * | 2018-02-05 | 2018-08-10 | 李刚毅 | Method for checking object and system |
| CN110688923A (en) * | 2019-09-19 | 2020-01-14 | 中国电子科技集团公司第二十九研究所 | Sentinel 1A SAR data-based urban inland inundation risk area extraction method |
| CN110942425A (en) * | 2019-11-26 | 2020-03-31 | 贵州师范学院 | Reconstruction method and reconstruction system of super-resolution image and electronic equipment |
| CN110930449A (en) * | 2019-11-27 | 2020-03-27 | 上海眼控科技股份有限公司 | Prediction result evaluation method and device, computer equipment and storage medium |
| CN115089839A (en) * | 2022-08-25 | 2022-09-23 | 柏斯速眠科技(深圳)有限公司 | Head detection method and system and control method and system of sleep-assisting device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102842130B (en) | 2015-07-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102842130A (en) | Method for detecting buildings and extracting number information from synthetic aperture radar image | |
| Van Den Eeckhaut et al. | Object-oriented identification of forested landslides with derivatives of single pulse LiDAR data | |
| Tian et al. | Building change detection based on satellite stereo imagery and digital surface models | |
| Kuffer et al. | The development of a morphological unplanned settlement index using very-high-resolution (VHR) imagery | |
| Kuffer et al. | Urban morphology of unplanned settlements: The use of spatial metrics in VHR remotely sensed images | |
| CN106650812B (en) | A kind of urban water-body extracting method of satellite remote-sensing image | |
| Gao et al. | Building extraction from RGB VHR images using shifted shadow algorithm | |
| CN111028255A (en) | Farmland area pre-screening method and device based on prior information and deep learning | |
| CN101975940A (en) | Segmentation combination-based adaptive constant false alarm rate target detection method for SAR image | |
| O'Neil et al. | Deep learning using physically-informed input data for wetland identification | |
| CN105160355A (en) | Remote sensing image change detection method based on region correlation and visual words | |
| Li et al. | Road detection from airborne LiDAR point clouds adaptive for variability of intensity data | |
| Bamdadinejad et al. | Shoreline extraction using image processing of satellite imageries | |
| CN110889840A (en) | Effectiveness detection method of high-resolution 6 # remote sensing satellite data for ground object target | |
| Uzar | Automatic building extraction with multi-sensor data using rule-based classification | |
| CN104951765A (en) | Remote sensing image target division method based on shape priori information and vision contrast ratio | |
| Sunny et al. | Performance study of imageries from MODIS, Landsat 8 and Sentinel-2 on measuring shoreline change at a regional scale | |
| CN108648200B (en) | Indirect urban high-resolution impervious surface extraction method | |
| CN113870224A (en) | Flood monitoring method, system, equipment and medium | |
| Puissant et al. | Coastline extraction in VHR imagery using mathematical morphology with spatial and spectral knowledge | |
| Demir et al. | A nonparametric fuzzy shoreline extraction approach from Sentinel-1A by integration of RASAT pan-sharpened imagery | |
| Liu et al. | A Novel Impervious Surface Extraction Method Based on Automatically Generating Training Samples From Multisource Remote Sensing Products: A Case Study of Wuhan City, China | |
| Lewiński et al. | Testing texture of VHR panchromatic data as a feature of land cover classification | |
| Zhang et al. | Semantic segmentation method accelerated quantitative analysis of the spatial characteristics of traditional villages | |
| Rettelbach et al. | From images to hydrologic networks-understanding the arctic landscape with graphs |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150708 Termination date: 20180704 |