CN106548141B - A kind of object-oriented farmland information extraction method based on the triangulation network - Google Patents
A kind of object-oriented farmland information extraction method based on the triangulation network Download PDFInfo
- Publication number
- CN106548141B CN106548141B CN201610934973.1A CN201610934973A CN106548141B CN 106548141 B CN106548141 B CN 106548141B CN 201610934973 A CN201610934973 A CN 201610934973A CN 106548141 B CN106548141 B CN 106548141B
- Authority
- CN
- China
- Prior art keywords
- triangulation network
- cluster
- central point
- oriented
- polygon
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- 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
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30188—Vegetation; Agriculture
Abstract
The present invention relates to a kind of object-oriented farmland information extraction method based on the triangulation network, the steps include: to be split high spatial resolution image using multi-scale division method；It rejects strip cutting object (road, irrigation canals and ditches etc.)；Extract the central point of surplus division object；The triangulation network is constructed using central point；Skinning operations are carried out to the triangulation network；The triangulation network is constructed using AUTOCLUST clustering algorithm, and is clustered；Using V constraint diagram, optimize cluster result, avoids excessively clustering and owing cluster；It is constrained using maximum variance, rejects remaining fragmentary forest land；It obtains plant extraction result and carries out accuracy evaluation.The present invention overcomes high spatial resolution remote sense image data volume is big, handle difficult problem, the semantic information that object after making full use of segmentation provides, the object interfered is extracted to farmland information by rejecting settlement place and road etc., to efficiently carry out automatically extracting for farmland information, and guarantee the globality of plant extraction.
Description
Technical field
The present invention relates to a kind of farmland information extraction methods, more particularly to a kind of the efficient non-supervisory of object-oriented
Farmland information extracting method.
Background technique
Automatically extracting for farmland information has vital work for the resource management of China rural holding and the reallocation of land
With.Currently, the related art method that farmland information charts is mainly based on remote sensing technology.On different space scales,
There is the research much extracted about farmland information, is broadly divided into part, area and Global Regional.The research of regional area mainly relates to
And single area or irrigated area basin or specific region；Geographic coverage mainly includes large river basins and continental region；And it is global
Scale then relates generally to the farmland information drawing of global range.Under different spaces scale, in order to reach optimal farmland information
Extraction effect, different sensing datas need to be summarized using different drawing modes according to the research achievement of many scholars,
The farmland information of regional area, which extracts, uses Landsat TM/ETM+, SPOT, LISS, ASTER, CBERS data can reach compared with
Good cost performance, information extracting method relate generally to photo reading, image digital operation, image classification, segmentation, visual fusion.
Geographic coverage mainly uses Landsat TM/ETM+, MODIS, MERIS, AVHRR, and SPOT VGT data pass through time series point
Analysis method, supervision or non-supervised classification, masking method identify farmland information.Global Scale mainly uses MODIS, MERIS,
AVHRR, SPOT VGT data, information extracting method relate generally to Non-surveillance clustering, the machine learning for time series data
Algorithm, and use other auxiliary datas (including statistical data, ground real time data etc.).
The spatial resolution for studying the remote sensing image used above is lower, when the arable land of identification small range and vital area is past
Past relatively difficult, this just needs to be aided with the image data of higher resolution.Particularly, China starts from the high-resolution of Eleventh Five-Year Plan
The implementation of earth observation systems special project, so that high spatial resolution image data (< 10m) acquiring technology is further developed.
However, foreign matter is serious with spectrum phenomenon since high spatial resolution image spectral information is more abundant, it is difficult with pixel-based point
Class method extracts arable land.In addition, the difference for different phenologys of ploughing, for example, in unmanned plane high score image, the color in nonirrigated farmland and paddy field
Coloured silk differs greatly, and nonirrigated farmland is yellow, and paddy field is green, is difficult the two while extracting.Therefore, the arable land special topic in high score image
The extraction difficulty of information is bigger.
For the arable land Extracting Thematic Information of high score image, has many scholars and done correlative study work, Liu et al. 2008
Year writes articles " High quality prime farmland extraction in " GIS and architectural environment joint conference "
Pattern based on object-oriented image analysis " passes through side using aviation image data (3m)
Edge detection segmentation image, subsequent computing object characteristic value are finally extracted decision tree using C4.5 building farmland information, are realized high-quality
The detection in amount standard farmland.It is write articles at " International Journal of Remote Sensing " within Lu etc. 2007
“Comparison between several feature extraction/classification methods for
mapping complicated agricultural land use patches using airborne
Hyperspectral data " extracts complicated agricultural land overlay area using Airborne Hyperspectral data (2m).Duro etc. 2012
Year writes articles " A comparison of pixel-based and at " Remote Sensing of Environment "
object-based image analysis with selected machine learning algorithms for the
Classification of agricultural landscapes using SPOT-5HRG imagery ", uses SPOT shadow
Picture is utilized respectively the supervised classification methods such as random gloomy, support vector machines, decision tree, is realized arable land using multi-scale division technology
The extraction of information.The suchlike object based on after segmentation, the method for extracting farmland information using measure of supervision is also very much, but
It is that high score image provides spectrum more abundant and texture feature information, above method are mostly secret operation, there is no sufficiently benefits
With the exclusive feature of arable land ground block message itself, cause extraction effect bad.
Currently, seldom specifically for the non-supervisory farmland information extracting method of high score image development, the degree of automation is not
Height utilizes the semantic information of the geographic object after segmentation and insufficient.For example, Sun and Xu 2009 exists
" Transactions of the CSAE " writes articles " Comer extraction algorithm for high-resolution
Imagery of agricultural land " first carries out image data using Quickbird panchromatic wave-band image is directed to
Segmentation, (the arable land shape rule, generally with 4 or 4 or more corner characteristics of the shape information in conjunction with provided by corner characteristics point
Point) extract the farming land information of the arable land after segmentation and the regular shapes such as swag.The research is high resolution spatial panchromatic image arable land
Information extraction opens a new thinking, utilizes the geometry and line in plot of ploughing in high resolution image data to a certain extent
Feature is managed, and the peculiar property in arable land plot carries out algorithm design, thus extraction farmland information rapidly and efficiently.But it is only
Single shape feature is utilized, the extraction effect to complex environment is simultaneously bad.In view of farmland information has apparent concentrate
Feature in flakes, and since the arable land object after high score Image Segmentation is larger, and for concentrating settlement place region, due to internal atural object
It is abundant, including house, forest land, road, cement floor etc., the cutting object to come in every shape is formed, opposite region of ploughing, the cut section
Domain seems that comparison is broken.Based on this feature, it is contemplated that the method for cluster rejects the atural objects such as broken settlement place, forest land,
To automatically extract farmland information.
Summary of the invention
The technical problem to be solved by the present invention is overcome the above-mentioned deficiency of the prior art, provide a kind of based on the triangulation network
It is poly- to combine object-oriented classification, triangulation network analysis and AUTOCLUST for object-oriented farmland information extraction method
Class algorithm is based on high score remote sensing image, automatically, efficiently and accurately carries out the extraction of farmland information.
In order to solve the above technical problems, a kind of object-oriented farmland information based on the triangulation network provided by the invention is automatic
Extracting method, comprising the following steps:
Step 1 is split high spatial resolution remote sense image using multi-scale division method；
Step 2 rejects strip cutting object；
Step 3, the central point for extracting surplus division object；
Step 4 constructs Delaunay triangulation network using central point；
Step 5 carries out skinning operations to the Delaunay triangulation network that step 4 obtains, and removes the outer of Delaunay triangulation network
Triangle；
Step 6 clusters the Delaunay triangulation network after peeling in step 5, and it is intensive to obtain triangulation network Node distribution
Several sparse two kinds of cluster results with distribution；
Step 7 is modified cluster result using Vornoio constraint diagram, and modification method is as follows: utilizing Delaunay tri-
Angle net node constructs Voronoi diagram, calculates Voronoi area of a polygon A, Voronoi area of a polygon in Voronoi diagram and is less than
When lower threshold, primarily determine that the corresponding central point of Voronoi polygon belongs to broken region, Voronoi area of a polygon
When greater than upper limit threshold, primarily determine the corresponding central point of the polygon for arable land region；It will be determined as the center in broken region
Point is left out or is included into the intensive cluster of Node distribution, and the central point in the region that is judged to ploughing is included into sparse poly- of Node distribution
In class；
Central point corresponding to the densely distributed cluster of step 8, deletion of node；
If the maximum spectral variance Max.diff for obtaining cutting object after the segmentation of step 9, the step 1 is greater than default threshold
Value, then determine all centers that the cutting object for forest land, rejects central point corresponding to the cutting object, finally remains
The corresponding cutting object of point is arable land, obtains plant extraction result.
The present invention also has feature further below:
1, in step 1, segmentation repeatedly is carried out to high spatial resolution image using image analysis software and is attempted, and is observed
The distribution characteristics of cutting object, selection carry out Image Segmentation to the preferable segmentation scale of arable land plot segmentation.
2, cutting object of the rectangle fitting degree character numerical value less than 0.5 is determined as strip cutting object, and rejects.
3, in step 3, the area of a polygon of surplus division object is calculated, and calculates the coordinate of central point, institute according to area
State the mass center that central point is cutting object polygon.
4, in step 5, distance parameter d is selected, deletes the side that Delaunay triangulation network intermediate cam shape side length is greater than d, d's takes
Being worth range is [100m, 150m].
5, in step 7, lower threshold isUpper limit threshold isWherein,
Voronio_mean is the average value of the area of the Thiessen polygon generated,WithFor coefficient, and
6、
7, it before step 7 or step 8 execute, makes the following judgment, if j-th of cluster meetsThe cluster is then included into sparse poly- of Node distribution
Class, whereinIndicate the mean value of the triangle area of j-th of cluster,Indicate all poly-
The area mean of mean of the triangle sets of class,Indicate the gore productive set of j-th of cluster
The variance of conjunction.
8, preset threshold is the maximum spectral variance mean value of 2 times of all cutting objects and the maximum spectrum of all cutting objects
The sum of variance criterion difference.
9, it is clustered in step 6 using AUTOCLUST method.
The present invention is based on Delaunay triangulation network combination AUTOCLUST clustering algorithms and the realization pair of a variety of optimization constrained procedures
Farmland information automatically extracts.The present invention can be on the basis of object-oriented classification by the triangulation network and clustering algorithm knot
It closes, automatically and accurately therefrom chooses reliable farmland information, meanwhile, it is strong to simplify human-computer interaction in farmland information extraction process
Degree, ensure that the globality of plant extraction.
Image division technology is used in step 1 of the invention, using high spatial resolution image, is attempted by segmentation repeatedly
And the distribution characteristics of cutting object is observed, discovery farmland information has apparent concentration feature in flakes, and the arable land block after segmentation is all
It is bigger, and due to the high resolution features of low latitude unmanned plane image, for concentrating settlement place region, internal atural object is abundant,
The cutting object to come in every shape is formed including house, forest land, road, cement floor etc., after segmentation, opposite region of ploughing, the segmentation
Region seems that comparison is broken.On the other hand, for Chinese vast rural area, settlement place region is distributed in arable land in checkerboard,
Generally, centralized residence settlement place and road area are the main interference factors that farmland information extracts, and road and irrigation canals and ditches etc. pair
As not being inconsistent with farmland information with apparent strip feature, therefore this class object is directly rejected in step 2.
Further, for the atural object of non-strip, the present invention is based on triangulation network analysis to the disturbing factor of bare place into
Row is gradually rejected, and will introduce AUTOCLUST cluster, Voronio constraint diagram and Variance Constraints in this process, a variety of clusters are about
Beam algorithm improves the precision of plant extraction step by step.
As it can be seen that the present invention based on the triangulation network, is main extraction algorithm using AUTOCLUST cluster, utilizes long short side
Cluster realizes the transmitting of clustering information, specifically, characterizes length side information using part and global variance, deletes in the triangulation network
Short side set and long line set are gone forward side by side a step expanded scope.
Compared with prior art, the present invention sufficiently combines the characteristic information and the triangulation network point after the Remote Sensing Image Segmentation of arable land
Analysis forms a kind of automatic, efficient, accurate farmland information extracting method.It specific innovative point and has the beneficial effect that:
First, the present invention can make full use of the variable density that can adapt to different zones point automatically of Delaunay triangulation network
Feature clustered using long short side and realize that the transmitting of clustering information is specifically adopted in conjunction with the AUTOCLUST algorithm based on graph theory
Length side information is characterized with part and global variance, so as to use this method to realize the cluster of the triangulation network.
Second, the present invention is further processed using Thiessen polygon constraint and is ignored on the basis of AUTOCLUST algorithm
The feature of triangulation network intermediate cam shape area obtains its dual graph Voronoi diagram according to Delaunay triangulation network, further passes through meter
The method for calculating Thiessen polygon area characterizes distribution density a little, and owing for occurring when solving using AUTOCLUST clustering algorithm is poly-
The phenomenon that class；Using cluster triangle web area average value constraint, the mistake cluster for solving the larger triangulation network in AUTOCLUST cluster is asked
Topic；It is constrained using maximum variance, advanced optimizes algorithm, reject the fragmentary forest land in plains region.
To sum up, the present invention sufficiently excavates the semantic information of high score remote sensing image, proposes one kind based on the triangulation network, knot
The plant extraction method of AUTOCLUST cluster is closed, the extraction of farmland information automatically, is efficiently and accurately carried out.Entire method, from
Dynamicization degree is higher, precision is higher, high stability.The farmland information extraction efficiency of high score remote sensing image is not only increased, together
When, it ensure that the globality of plant extraction.
Detailed description of the invention
The present invention will be further described below with reference to the drawings.
Fig. 1 is the method for the present invention flow chart.
Fig. 2 a is that unmanned plane tests raw video.
Fig. 2 b is the segmentation result that unmanned plane test raw video segmentation scale is 50.
Fig. 2 c is the segmentation result that unmanned plane test raw video segmentation scale is 100.
Fig. 2 d is the segmentation result that unmanned plane test raw video segmentation scale is 180.
Fig. 3 a is Delaunay triangulation network schematic diagram.
Fig. 3 b is the outer triangle exemplary diagram for removing Delaunay triangulation network.
Fig. 4 a is the triangulation network of Fig. 2 d segmentation result building.
Fig. 4 b is AUTOCLUST cluster result.
Fig. 4 c is the dendrogram of Voronio figure lower threshold constraint.
Fig. 4 d is the dendrogram of Voronio figure upper limit threshold constraint.
Fig. 5 a is the plant extraction result using the method proposed.
Fig. 5 b is to carry out the result that plant extraction obtains using image analysis software eCognition.
Specific embodiment
Below according to attached drawing, the present invention will be described in detail, and the objects and effects of the present invention will be more apparent.
It is as shown in Figure 1 the process of object-oriented farmland information extraction method of the embodiment of the present invention based on the triangulation network
Figure, present implementation the following steps are included:
Step 1 is split high spatial resolution image using multi-scale division method.In the present embodiment, use
ECognition carries out segmentation repeatedly to high spatial resolution image and attempts, and observes the distribution characteristics of cutting object, selects
Image Segmentation is carried out to the preferable segmentation scale of arable land plot segmentation.It is proposed that segmentation scale value range be [150,
200], to protrude the difference of the cutting object corresponding with arable land of the settlement place in high score image.
Fig. 2 is that three scape unmanned planes test image.In this step, carried out using image analysis software eCognition multiple dimensioned
Segmentation, partitioning parameters Shape are 0.4, Smoothness 0.5.Respectively implement three scales segmentation (including scale 50,
100,180), by observing the segmentation result of different scale, selection divides preferable scale to arable land plot and carries out subsequent analysis.
It is easy discovery, with the increase of segmentation scale, the area of cutting object is gradually increased.Opposite segment scale 50 and 100, works as segmentation
Scale at 180, can more preferably will in 0.2 meter of resolution ratio of unmanned plane image data with same pixel feature arable land plot
It splits.
Fig. 2 a is raw video, and Fig. 2 b is the segmentation result divided scale and be 50, and Fig. 2 c is the segmentation divided scale and be 100
As a result, Fig. 2 d is the segmentation result divided scale and be 180.
Step 2, the cutting object (road, irrigation canals and ditches etc.) for rejecting elongate in shape.In the present embodiment, by rectangle fitting degree feature
Cutting object of (Rectangular fit) numerical value less than 0.5 is rejected, and rectangle fitting degree feature (Rectangular fit) can
For rejecting the strip cutting object of road, irrigation canals and ditches etc. with arable land ground block feature notable difference.
After success is split unmanned plane image data, triangulation network cluster will be used to reject broken settlement place region, with
Extract farmland information.Central point (step 3), the central point number and the region segmentation pair of extraction are extracted first with cutting object
As number is identical.Although the distribution of the arable land Fig. 2 d is relatively concentrated, inevitably it is mingled with the atural objects such as road, irrigation canals and ditches, ridge, in high-resolution
In the image of rate, these atural objects can be divided out, and they belong to elongate in shape atural object, have with the rectangle in arable land larger
Difference.In the central point that step 3 is extracted, more central points of these atural objects, they will affect method precision, be improvement method
Reliability needs to reject these strip atural objects before constructing the triangulation network.So use space characteristic rectangle is fitted first
It spends (Rectangular fit) threshold value and rejects strip atural object (fit < 0.5 Rectangular) corresponding central point.
Step 3, the central point for extracting surplus division object.In this example, the polygon facet of surplus division object is calculated first
Product, and the coordinate of its central point is calculated according to area.
For the central point (center of gravity or mass center) for calculating cutting object polygon, the area of polygon is calculated first, if certain
The cut-off rule of a cutting object has N number of node (xi,yi) composition, wherein 0≤i≤N-1.Assuming that the last one node and first
Node is identical, then polygon Shape closed.So, the areal calculation formula of polygon indicates are as follows:
Wherein xN=x0, then it represents that polygon Shape closed.After area is calculated, polygon central point is (here actually
Indicate center of gravity or mass center) coordinate (Cx, Cy) can be calculate by the following formula:
Step 4 constructs Delaunay triangulation network, Delaunay triangulation network such as Fig. 3 a institute using the central point that step 3 is extracted
Show.
Delaunay triangulation network is the triangle sets for not overlapping each other and adjoining each other, and need to generally meet two criterion:
Other points are not present in the circumscribed circle of each triangle in Delaunay triangulation network.In all possible triangulation network,
The minimum angle of each triangle in Delaunay triangulation network is generally maximum.According to features above, many researchs and proposes different three
Angle net construction method, popular method include: divide and conquer, scan-line algorithm, delta algorithm, Fast incremental construction algorithm, convex closure
Algorithm.Emphasis of the present invention does not lie in triangulation network developing algorithm, and the present embodiment utilizes the mature algorithm (Divide-and- that divides and rules
Conquer Algorithm) building Delaunay triangulation network, firstly, dividing point set as far as possible to sufficiently small subset, to different sons
Collection constructs the sub- triangulation network, then integrates each sub- triangulation network, finally, the triangulation network for guaranteeing building using local optimum method is
Delaunay triangulation network, the advantages of building triangulation network is that time efficiency is higher in this way, and it is big that disadvantage is needed recursive operation amount, right
Memory headroom it is more demanding.
Step 5 carries out skinning operations to the Delaunay triangulation network that step 4 obtains, and removes the outer of Delaunay triangulation network
Triangle.In this example, using the Delaunay triangulation network obtained in step 4, select distance parameter d, distance parameter d that should be not less than
Longest edge inside the triangulation network deletes the long side for being greater than d of triangle edges it is proposed that the value range of d is [100m, 150m],
And it will be only with an associated side of triangle as the boundary of point group.It is as shown in Figure 3b outer the three of removing Delaunay triangulation network
The result at angle.
Point group distribution belongs to uncertain problem.In the research of GIS, it is commonly used for substitution point group distribution.
But when the distribution of point group is slim and sexy, it would be possible that being formed in this way comprising many there is no the recess area of point in convex hull
Convex hull can not really represent the distribution of point group.Generally, based on the sorting procedure of Delaunay triangulation network are as follows: 1.
The triangulation network is constructed using point group, such as Fig. 3 a；2. selecting a distance parameter d, " peeling " to the triangulation network is realized by parameter d,
Delete the long side for being greater than d of triangle edges；3. the boundary as point group will be extracted with the associated side of a triangle.Fig. 3 b shows,
It is operated by " peeling ", deletes the very big triangle of side length, so that the triangulation network after " peeling " divides closer to true point group
Cloth.
Step 6 carries out AUTOCLUST cluster to the Delaunay triangulation network after peeling in step 5, obtains triangulation network node
Several two kinds of cluster results densely distributed and that distribution is sparse.AUTOCLUST clustering algorithm is existing ripe algorithm,
It sees Estivill-Castro and Lee and was published in " Computers, Environment and urban in 2002
Systems " on document " Argument free clustering for large spatial point data sets
via boundary extraction from Delaunay Diagram".As the tetra- pieces of regions A, B, C, D, E, F are in Fig. 4 a
High density clusters position.
Other than using AUTOCLUST clustering method, K-means, density-based spatial clustering etc. also can be used
Method clusters center point, obtains several densely distributed and sparse distribution two kinds of clusters.
Step 7 is modified AUTOCLUST cluster result using Vornoio constraint diagram, avoids excessively clustering and owing poly-
Class represents the residential area region not being clustered in black quadrangle and triangle in figure, that is, owes cluster as shown in Figure 4 b.This reality
It applies in example, constructs Voronoi diagram using Delaunay triangulation network node (central point namely obtained in step 4), calculate
Voronoi area of a polygon in Voronoi diagram primarily determines the polygon when Voronoi area of a polygon is greater than upper limit threshold
Corresponding central point is that arable land primarily determines the Voronoi polygon pair when Voronoi area of a polygon is less than lower threshold
The central point answered belongs to broken region (bare place), leaves out or be included into the intensive cluster of Node distribution for the central point in broken region
In.The amendment to cluster result is completed, avoids excessively clustering and owing cluster phenomenon.
As illustrated in fig. 4 c, to use that lower threshold constrained as a result, the thicker Voronoi polygon of lines is in figure
Area is less than the polygon of lower limit value, primarily determines as bare place.As shown in figure 4d, the knot constrained for practical upper limit threshold value
Fruit, the thicker Voronoi polygon of lines is the polygon that area is greater than lower limit value in figure, is primarily determined as arable land.
Voronoi diagram is constructed based on Delaunay triangulation network, area of a polygon is calculated, is expressed as A.So Voronoi is more
The dot density of side shape can be characterized by 1/A.Generally, the value is smaller, and point group density is smaller, and the value is bigger, indicates that point group is close
It spends bigger.The area-constrained strategy of Thiessen polygon: generally, when AUTOCLUST is clustered, many regions occur owing cluster
The phenomenon that, then when being constrained using Thiessen polygon, SC service ceiling threshold value and lower threshold comprehensive constraint, generally,
When Thiessen polygon area is greater than upper limit threshold, the corresponding central point of the polygon must be arable land, when Thiessen polygon face
When product is less than lower threshold, the corresponding central point of the polygon must belong to broken region.Here bound threshold value is as far as possible
Setting is loose.High density point region and low-density point region are filtered with following formula:
Here Voronio_mean is the average value of the area of the Thiessen polygon generated,WithFor coefficient, andIt is apparent that coefficient here determines cluster degree a little, and by experiment, in order to avoid excessively clustering and owing cluster,
SettingAs general value, AUTOCLUST cluster result is modified.
During AUTOCLUST cluster, it also includes the biggish triangulation network that some clusters are main, and its area relatively around
Triangle web area is also big, this may be because the corresponding central point of two or three of arable land cutting objects of only a few is apart from close, and
Farther out apart from adjacent other points, this point is frequently not the central point of resident's region segmentation object.Triangle in this cluster
Net number is often few, and the average value of triangle area than the triangle web area in the broken region for needing to identify average value greatly very
It is more.
So being further introduced into the concept of cluster triangle web area mean value to remove this kind of mistake in AUTOCLUST cluster process
It accidentally clusters, equation is as follows:
WhereinIndicate the mean value of the triangle area of j-th of cluster, TiFor i-th of triangle
Area, nj be j-th cluster triangle number,Indicate being averaged for the triangle area mean value of all clusters
Value,Indicate the variance of the triangle area set of j-th of cluster, m indicates the number of cluster, and i is indicated
Triangle serial number in j-th of cluster.
If j-th of cluster meetsThen will
The cluster, which is included into, to be distributed in sparse cluster.
Central point corresponding to the densely distributed cluster of step 8, deletion of node, so far completes the center in bare place region
The rejecting of point.
Step 9 is constrained using maximum variance, rejects remaining fragmentary forest land, if obtaining segmentation pair after the step 1 segmentation
The maximum spectral variance Max.diff of elephant is greater than preset threshold, then determines the cutting object for forest land, reject the cutting object institute
Corresponding central point, cutting object corresponding to all central points finally remained are arable land, obtain plant extraction result
And carry out accuracy evaluation.In the present embodiment, the preset threshold is 2 times of maximum spectral variance mean values and maximum spectral variance standard
The sum of difference.
Although the method that the present embodiment proposes can cluster settlement place aggregation zone, and be rejected, for Plain
Also sporadicly there are many forest lands in area, since the textural characteristics in forest land are not much different, corresponding cutting object is often larger, very
Difficulty is rejected with triangulation network clustering method.And the coarse texture in forest land, variance are larger, plough relatively smooth, variance is smaller, institute
It is trimmed in the method for proposing Variance Constraints here to extract result to farmland information.The present embodiment is using 2 times of maximum spectrum
The sum of mean variance and maximum spectral variance standard deviation are identified as threshold value.Such as following formula:
Max.diff>2×mean_Max.diff+Standard Deviation_Max.diff (9)
Wherein mean_Max.diff indicates the mean value of the maximum spectral variance feature of all objects in segmentation figure layer；Feature
The standard deviation of Max.diff is expressed as Standard Deviation_Max.diff.
Fig. 5 a is the farmland information extracted using the object-oriented farmland information extraction method based on the triangulation network, Fig. 5 a
In, white indicates the arable land extracted, and black indicates the bare place of identification, and Fig. 5 b expression is exercised supervision using eCognition software
Classify the farmland information extracted, white is arable land in figure, and black is bare place.It was found that: extracting method of the present invention is guaranteeing
Under the premise of required precision, the concentration ploughed characteristic in flakes can not only be kept on the whole, moreover it is possible to avoid to a certain extent
Due to sowing type difference (phenology difference), caused fragmentary mistake divides phenomenon.Meanwhile avoiding supervised classification selection training sample
Process improves automatic processing degree.
In addition to the implementation, the present invention can also have other embodiments.It is all to use cutting object central point cluster etc.
The farmland information extractive technique scheme of form, falls within the scope of protection required by the present invention.
Claims (9)
1. a kind of object-oriented farmland information extraction method based on the triangulation network, comprising the following steps:
Step 1 is split high spatial resolution image using multi-scale division method；
Step 2 rejects strip cutting object；
Step 3, the central point for extracting surplus division object；
Step 4 constructs Delaunay triangulation network using central point；
Step 5 carries out skinning operations to the Delaunay triangulation network that step 4 obtains, and removes the outer triangle of Delaunay triangulation network；
Step 6 clusters the Delaunay triangulation network after peeling in step 5, obtains triangulation network Node distribution intensively and divides
Several sparse two kinds of cluster results of cloth；
Step 7 is modified cluster result using Vornoio constraint diagram, and modification method is as follows: utilizing Delaunay triangulation network
Node constructs Voronoi diagram, calculates Voronoi area of a polygon in Voronoi diagram, and Voronoi area of a polygon is less than lower limit
When threshold value, primarily determine that the corresponding central point of Voronoi polygon belongs to broken region, Voronoi area of a polygon is greater than
When upper limit threshold, primarily determine the corresponding central point of the polygon for arable land region；The central point for being determined as broken region is deleted
It goes or is included into the intensive cluster of Node distribution, the central point in the region that is judged to ploughing is included into the sparse cluster of Node distribution
In；
Central point corresponding to the densely distributed cluster of step 8, deletion of node；
If the maximum spectral variance Max.diff for obtaining cutting object after the segmentation of step 9, the step 1 is greater than preset threshold,
Determine that the cutting object for forest land, rejects central point corresponding to the cutting object, all central point institutes finally remained
Corresponding cutting object is arable land, obtains plant extraction result.
2. the object-oriented farmland information extraction method according to claim 1 based on the triangulation network, it is characterised in that:
Cutting object of the rectangle fitting degree character numerical value less than 0.5 is determined as strip cutting object, and rejects.
3. the object-oriented farmland information extraction method according to claim 1 based on the triangulation network, it is characterised in that:
In step 3, the area of a polygon of surplus division object is calculated, and calculates the coordinate of central point according to area, the central point is
The mass center of cutting object polygon.
4. the object-oriented farmland information extraction method according to claim 1 based on the triangulation network, it is characterised in that:
In step 5, distance parameter d is selected, deletes the side that Delaunay triangulation network intermediate cam shape side length is greater than d, the value range of d is
[100m,150m]。
5. the object-oriented farmland information extraction method according to claim 1 based on the triangulation network, it is characterised in that:
In step 7, lower threshold isUpper limit threshold isWherein, Voronio_
Mean is the average value of the area of the Thiessen polygon generated,WithFor coefficient, and
6. the object-oriented farmland information extraction method according to claim 5 based on the triangulation network, it is characterised in that:
7. the object-oriented farmland information extraction method according to claim 1 based on the triangulation network, it is characterised in that:
It before step 7 or step 8 execute, makes the following judgment, if j-th of cluster meetsThe cluster is then included into sparse poly- of Node distribution
Class, whereinIndicate the mean value of the triangle area of j-th of cluster,Indicate all poly-
The area mean of mean of the triangle sets of class,Indicate the gore productive set of j-th of cluster
The variance of conjunction.
8. the object-oriented farmland information extraction method according to claim 1 based on the triangulation network, it is characterised in that:
Preset threshold is the maximum spectral variance mean value of 2 times of all cutting objects and the maximum spectral variance standard deviation of all cutting objects
The sum of.
9. the object-oriented farmland information extraction method according to claim 1 based on the triangulation network, it is characterised in that:
Automatic cluster is carried out using AUTOCLUST method in step 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610934973.1A CN106548141B (en) | 2016-11-01 | 2016-11-01 | A kind of object-oriented farmland information extraction method based on the triangulation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610934973.1A CN106548141B (en) | 2016-11-01 | 2016-11-01 | A kind of object-oriented farmland information extraction method based on the triangulation network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106548141A CN106548141A (en) | 2017-03-29 |
| CN106548141B true CN106548141B (en) | 2019-07-26 |
Family
ID=58392223
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610934973.1A Active CN106548141B (en) | 2016-11-01 | 2016-11-01 | A kind of object-oriented farmland information extraction method based on the triangulation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106548141B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107221001A (en) * | 2017-05-18 | 2017-09-29 | 苏州恩巨网络有限公司 | A kind of area information acquisition methods and device |
| CN107564077B (en) * | 2017-09-04 | 2019-12-10 | 浪潮集团有限公司 | method and device for drawing road network |
| CN107705002B (en) * | 2017-09-21 | 2020-07-03 | 中国矿业大学(北京) | Method for determining influence range of abnormal high value of sampling point of heavy metal content in mining area soil |
| CN107766495A (en) * | 2017-10-19 | 2018-03-06 | 大连大学 | Distributed temporal index method based on Voronoi diagram |
| CN107958087A (en) * | 2017-10-19 | 2018-04-24 | 大连大学 | The representative point acquisition methods of the distributed temporal index of the row's of falling Thiessen polygon of portable medical |
| CN108021690A (en) * | 2017-10-19 | 2018-05-11 | 大连大学 | Arrange Voronoi indexes |
| CN107766496A (en) * | 2017-10-19 | 2018-03-06 | 大连大学 | Based on MapReduce and the extensive NN Query system for arranging Thiessen polygon |
| CN108961403A (en) * | 2018-06-22 | 2018-12-07 | 武汉大学 | A method of major trunk roads are extracted with open source street map |
| CN111768505B (en) * | 2020-07-01 | 2023-10-17 | 江西师范大学 | Planar point set shape reconstruction method and device and electronic equipment |
| CN112330700A (en) * | 2020-11-16 | 2021-02-05 | 四川航天神坤科技有限公司 | Cultivated land plot extraction method of satellite image |
| CN112926029B (en) * | 2021-01-29 | 2022-08-26 | 四川省环境政策研究与规划院 | Residential area identification and division method for rural domestic sewage treatment |
| CN112598724B (en) * | 2021-03-01 | 2021-06-01 | 武大吉奥信息技术有限公司 | Improved TIN-based vector data center line extraction method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102073981A (en) * | 2011-02-23 | 2011-05-25 | 中国测绘科学研究院 | Point group geographic entity selection method under the restriction of correlated elements |
| CN104036024A (en) * | 2014-06-27 | 2014-09-10 | 浙江大学 | Spatial clustering method based on GACUC (greedy agglomerate category utility clustering) and Delaunay triangulation network |
| JP2016062170A (en) * | 2014-09-16 | 2016-04-25 | 三菱製紙株式会社 | Light transmissive conductive material |
| CN105608474A (en) * | 2015-12-30 | 2016-05-25 | 二十一世纪空间技术应用股份有限公司 | High-resolution-image-based regional adaptive cultivated land extraction method |
| CN105719014A (en) * | 2016-01-18 | 2016-06-29 | 中国农业大学 | Provincial cultivated land back-up resource exploiting and zoning method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8982123B2 (en) * | 2008-06-19 | 2015-03-17 | Xerox Corporation | Reducing the number of vertices in a clustered delaunay mesh |
-
2016
- 2016-11-01 CN CN201610934973.1A patent/CN106548141B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102073981A (en) * | 2011-02-23 | 2011-05-25 | 中国测绘科学研究院 | Point group geographic entity selection method under the restriction of correlated elements |
| CN104036024A (en) * | 2014-06-27 | 2014-09-10 | 浙江大学 | Spatial clustering method based on GACUC (greedy agglomerate category utility clustering) and Delaunay triangulation network |
| JP2016062170A (en) * | 2014-09-16 | 2016-04-25 | 三菱製紙株式会社 | Light transmissive conductive material |
| CN105608474A (en) * | 2015-12-30 | 2016-05-25 | 二十一世纪空间技术应用股份有限公司 | High-resolution-image-based regional adaptive cultivated land extraction method |
| CN105719014A (en) * | 2016-01-18 | 2016-06-29 | 中国农业大学 | Provincial cultivated land back-up resource exploiting and zoning method and system |
Non-Patent Citations (3)
| Title |
|---|
| Delaunay三角网与Voronoi图在GIS中的应用研究;彭仪普等;《测绘工程》;20020930;第11卷(第3期);第39-41页 |
| Effective Privacy Preserved Clustering Based on Voronoi Diagram;Jinfei Liu等;《2011 Eighth International Symposium on Voronoi Diagrams in Science and Engineering》;20111231;第206-213页 |
| 农用地遥感影像信息的角提取方法;孙小丹等;《农业工程学报》;20091031;第25卷(第10期);第135-141页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106548141A (en) | 2017-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106548141B (en) | A kind of object-oriented farmland information extraction method based on the triangulation network | |
| AU2020103026A4 (en) | A Single Tree Crown Segmentation Algorithm Based on Super-pixels and Topological Features in Aerial Images | |
| CN101840581B (en) | Method for extracting profile of building from satellite remote sensing image | |
| Blaschke et al. | Image segmentation methods for object-based analysis and classification | |
| CN101408941B (en) | Method for multi-dimension segmentation of remote sensing image and representation of segmentation result hierarchical structure | |
| CN102096816B (en) | Multi-scale multi-level image segmentation method based on minimum spanning tree | |
| CN106815847A (en) | Trees dividing method and single tree extracting method based on laser radar point cloud | |
| CN106611421B (en) | The SAR image segmentation method constrained based on feature learning and sketch line segment | |
| CN109919944A (en) | A kind of joint super-pixel figure of complex scene building variation detection cuts optimization method | |
| CN103839267B (en) | Building extracting method based on morphological building indexes | |
| CN108537239A (en) | A kind of method of saliency target detection | |
| Su | Scale-variable region-merging for high resolution remote sensing image segmentation | |
| CN108985247A (en) | Multispectral image urban road identification method | |
| CN105608474A (en) | High-resolution-image-based regional adaptive cultivated land extraction method | |
| CN104268559A (en) | Paddy field and dry land distinguishing method based on oriented objects and medium-resolution-ration remote sensing image | |
| CN108830844A (en) | A kind of facilities vegetable extracting method based on multidate high-resolution remote sensing image | |
| CN107392133A (en) | Utilize the ermophyte remote sensing recognition method of object-oriented Multi-source Information Fusion | |
| Wang et al. | Color-and texture-based image segmentation for improved forest delineation | |
| CN107992856A (en) | High score remote sensing building effects detection method under City scenarios | |
| CN109766824A (en) | Main passive remote sensing data fusion classification method based on Fuzzy Evidence Theory | |
| Truax | Comparing spectral and object based approaches for classification and transportation feature extraction from high resolution multispectral imagery | |
| CN109829425A (en) | A kind of small scale terrain classification method and system of Farmland Landscape | |
| CN107194857B (en) | Available land resource analysis system based on high-resolution satellite remote sensing data | |
| Zuo et al. | A method of crop seedling plant segmentation on edge information fusion model | |
| CN102622345B (en) | High-precision land-utilization remote sensing updating technology with synergistic multisource spatio-temporal data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |