CN109635828A - A kind of typical geographical national conditions elements recognition system and method in ecological protection red line area - Google Patents
A kind of typical geographical national conditions elements recognition system and method in ecological protection red line area Download PDFInfo
- Publication number
- CN109635828A CN109635828A CN201811592299.9A CN201811592299A CN109635828A CN 109635828 A CN109635828 A CN 109635828A CN 201811592299 A CN201811592299 A CN 201811592299A CN 109635828 A CN109635828 A CN 109635828A
- Authority
- CN
- China
- Prior art keywords
- image
- scale
- scale collection
- region
- segmentation
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 238000000605 extraction Methods 0.000 claims abstract description 26
- 238000003709 image segmentation Methods 0.000 claims abstract description 23
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 22
- 239000000284 extract Substances 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 9
- 230000011218 segmentation Effects 0.000 claims description 51
- 238000012549 training Methods 0.000 claims description 13
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 9
- 238000012706 support-vector machine Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 230000003595 spectral effect Effects 0.000 claims description 7
- 239000000203 mixture Substances 0.000 claims 1
- 239000000523 sample Substances 0.000 description 24
- 238000001228 spectrum Methods 0.000 description 16
- 230000000875 corresponding effect Effects 0.000 description 14
- 238000013507 mapping Methods 0.000 description 8
- 230000008859 change Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000012937 correction Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 235000013399 edible fruits Nutrition 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000009432 framing Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- VAYOSLLFUXYJDT-RDTXWAMCSA-N Lysergic acid diethylamide Chemical compound C1=CC(C=2[C@H](N(C)C[C@@H](C=2)C(=O)N(CC)CC)C2)=C3C2=CNC3=C1 VAYOSLLFUXYJDT-RDTXWAMCSA-N 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 239000012141 concentrate Substances 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000003703 image analysis method Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 230000004083 survival effect Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000012468 concentrated sample Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 230000009182 swimming Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of typical geographical national conditions elements recognition system and methods in ecological protection red line area, which includes: data processing module, for handling satellite image data；Scale collection divides module, and the scale collection model for establishing scale collection model, and based on foundation carries out Image Segmentation to the satellite image data, obtains image scale collection；Information extraction modules extract arable land, forest land, water body, these four types of typical geographical national conditions elements of settlement place for carrying out the information extraction of object-oriented based on the image scale collection.Present system and method can carry out typical geographical national conditions element automatic interpretation and extract, and improve efficiency and accuracy.
Description
Technical field
The present invention relates to ECOLOGICAL ENVIRONMENTAL MONITORING technical fields, the in particular to a kind of typical geographical national conditions in ecological protection red line area
Elements recognition system and method.
Background technique
Geographical national conditions elements recognition is geographical national conditions generaI investigation and the basic work in monitoring, wherein arable land, woods
Ground, water body, 4 class of settlement place are typical geographical national conditions element in geographical national conditions element.The essence of ecological protection red line is ecology
The baseline of Environmental security extracts the weight that research and application is National Ecological Security to the ground mulching element of ecological protection red line
It ensures.At this stage, the ecological protection red line area geography national conditions elements recognition method based on high-resolution remote sensing image is with artificial
Based on visual interpretation, heavy workload, the degree of automation is low, limits the raising of geographical national conditions elements recognition efficiency.For this purpose, research
A kind of geographical national conditions elements recognition technology of automation is problem urgently to be resolved in the monitoring of ecological protection red line area geography national conditions.
Summary of the invention
The purpose of the present invention is to provide a kind of typical geographical national conditions elements recognition system and method in ecological protection red line area,
Realize that geographical national conditions element automation is extracted.
To achieve the goals above, the present invention the following technical schemes are provided:
A kind of typical geographical national conditions elements recognition system and method in ecological protection red line area, comprising:
Data processing module, for handling satellite image data；
Scale collection divides module, and the scale collection model for establishing scale collection model, and based on foundation is to the satellite shadow
As data progress Image Segmentation, image scale collection is obtained；
Information extraction modules extract ecology for carrying out the information extraction of object-oriented based on the image scale collection
Protect red line area arable land, forest land, water body, these four types of typical geographical national conditions elements of settlement place.
On the other hand, present invention implementation additionally provides a kind of typical geographical national conditions elements recognition system in ecological protection red line area
And method, comprising the following steps:
Handle satellite image data；
Scale collection model is established, and the scale collection model based on foundation carries out Image Segmentation to the satellite image data,
Obtain image scale collection；
The information extraction that object-oriented is carried out based on the image scale collection extracts ecological protection red line area arable land, woods
These four types of typical geographical national conditions elements of ground, water body, settlement place.
Compared with prior art, present invention has the advantage that
Using the image division method of region merging technique, hierarchical relationship up and down is established；By region merging technique cost criterion, establish
Divide scale index；Scale collection is expressed using Two Binomial Tree Model.Change of scale can be quickly realized by scale collection, to obtain
Obtain the Image Segmentation result under any scale.The it is proposed of scale collection can promote the efficiency of Image Segmentation, and by once dividing, just
The segmentation result under any scale can be obtained, convenient for the selection of optimal scale.
Based on image scale collection, for different samples, by change of scale, training is selected on optimal scale
Sample realizes the samples selection across scale；Image classification is carried out under all scales, the image classification knot being superimposed under different scale
Fruit realizes the image classification across scale.Object-oriented classification method across scale can be to avoid area caused by single scale
Domain expression inaccuracy and interpretation precision be not high, improves the precision of image classification.
Be particularly suitable for ecological protection red line area arable land, forest land, water body, 4 quasi-representative geography national conditions element of settlement place from
It is dynamic to extract, it can also be popularized and applied in the extraction of other regions or planar geography national conditions element.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly introduced, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as to model
The restriction enclosed for those of ordinary skill in the art without creative efforts, can also be according to these
Attached drawing obtains other relevant drawings.
Fig. 1 show the comprising modules figure of typical geographical national conditions elements recognition system of the present invention.
Fig. 2 is merging process schematic diagram.
Fig. 3 is scale collection institutional framework schematic diagram.
Fig. 4 is image segmentation flow diagram.
Fig. 5 a, 5b, 5c are respectively over-segmentation, appropriately divide, the schematic diagram of less divided.
Fig. 6 a, 6b are respectively the practical national conditions data in forest land, forest land interpretation data.
Specific embodiment
Below in conjunction with attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear and complete
Description.It should be appreciated that described herein, the specific embodiments are only for explaining the present invention, is not intended to limit the present invention.It is based on
The embodiment of the present invention, those skilled in the art's every other embodiment obtained under the premise of no creative work,
Belong to protection scope of the present invention.
The typical geographical national conditions elements recognition system and method in ecological protection red line provided in this embodiment area, with high-resolution
Remote sensing image data is data source, carries out initial segmentation to image using the fast partition method based on graph theory, is closed using region
And method obtain different scale Image Segmentation as a result, using binary tree structure express scale collection model；Pass through change of scale
Method is based on image scale collection, the Image Segmentation result being quickly obtained under any scale；Using the feature extraction side of object-oriented
Method extracts the spectrum and Texture eigenvalue of image patch；Training sample is selected under different scale, is obtained based on algorithm of support vector machine
Final image classification extracts ecological protection red line area arable land, forest land, water body, settlement place this 4 class as a result, on this basis
Geographical national conditions element.
Specifically, referring to Fig. 1, typical geographical national conditions elements recognition system, ecological protection red line provided in this embodiment area
System, comprising: data processing module, scale collection divide module, information extraction modules.
Wherein, data collection module is for handling high score (high-resolution) satellite image data, DIGITAL PLANNING map datum
And Law of DEM Data；Wherein, strong high-resolution 1:50000DOM (digital orthoimage) achievement data of Up-to-date state is made
For orthography production and the basic data of geographical national conditions elements recognition；1:10000DEM (digital elevation model) data, as
The elevation control data that high resolution image is just penetrating correction uses.The picture control achievement in elements recognition region can also be collected simultaneously
(including aviation image photo control point and satellite image photo control point) and sky three encrypt achievement data, and the main of correction is just being penetrated as image
The accuracy checking data of control data and orthography achievement.Treatment process is for example with satellite image, Control data, DLG data
And dem data is basic document, and correction is just being penetrated to panchromatic image, it is right using the panchromatic image after correcting as reference
Multispectral image carries out registration correction.Panchromatic image is merged with multispectral image, fusion evaluation is inlayed by map sheet
It is embedding and cut, and the processing such as Imaging enhanced, color adjustment are carried out to framing image, produce DOM achievement (including whole scape image,
Framing image).
Wherein, scale collection segmentation module is for establishing scale collection model, and the scale collection based on foundation carries out Image Segmentation,
Obtain image scale collection.Specifically, the method merged using level regions, the spectrum, texture and shape for comprehensively utilizing image are special
Sign carries out the scale collection modeling of image.By establishing the scale collection model of image, obtaining one includes all scale Image Segmentations
As a result set, and hierarchical structure tissue is carried out, and then facilitate subsequent Fast Segmentation result queries and application.
Scale is enabled to integrate as S=(H, λ+), wherein H records the set of multi-scale division algorithm all areas generated, λ+Note
Record the dimensional information in each region in H.There are space correlation relationship between region in H, the level group between different zones can establish
Knit relationship；Again by the dimensional properties in region, the expression of results of image under corresponding scale can be quickly obtained.Scale collection is not ruler
The set for spending parameter, is a series of set in regions with scale properties, and gathers internal each region with relatively special
Hierarchical relationship, to facilitate the expression-form for obtaining image under any scale.Scale collection described in the present embodiment has following 3
Kind feature:
(1) scale collection structure is a regular binary tree, in addition to leaf node all nodes there are two child node,
The number N of its child nodes corresponds to the number of prime area；The total number of node is 2N-1.
(2) there is scale as index using it in all nodes, map that one-dimensional scale axis.Each node has
Its specific scale parameter: there is the Scaling interval of scale and survival.Scale portrayal wherein occur is seat of the node in scale axis
Mark, the Scaling interval of survival show as its difference for scale and disappearance scale (scale occurs in father node) occur.
(3) node indicates domain entities, and the line between node indicates regional level relationship.Wherein, leaf node represents institute
There is prime area, non-leaf nodes represents the region generated in the merging process of region.
Scale integrates object-based image interpretation procedure decomposition as image multi-scale expression-Knowledge based engineering image solution
It translates.In the first stage, by the way that image is carried out the multi-scale expression based on region, the multiple dimensioned knot of image based on region is established
Structure selects corresponding segmentation result by given parameter in the application.Compared to traditional multi-scale division, scale collection is most
Big advantage is to be indexed according to scale, and the segmentation result of any scale of quick obtaining realizes any ruler of segmentation result
Degree conversion.
Wherein, information extraction modules are used to carry out the information extraction of object-oriented based on the image scale collection, extract
Arable land, forest land, water body, the typical geographical national conditions element in settlement place these fourth types ecological protection red line area.Specifically, information extraction modules
On the basis of the Image Segmentation result based on scale collection, using the Feature Extraction Technology of object-oriented, spectrum, the line of image patch are extracted
The features such as reason；It using scale collection model, is switched fast in the segmentation result of different scale, across scale selection training sample；It adopts
Image classification result is obtained using the training sample of selection and the feature vector of each image patch with algorithm of support vector machine；In this base
On plinth, arable land, forest land, water body, 4 class geography national conditions element of settlement place are extracted.
More specifically, wherein it includes initial segmentation submodule, region merging technique submodule, checkrow wire that scale collection, which divides module,
Introduction module.
Wherein, initial segmentation submodule carries out initial segmentation to image for the Fast Segmentation Algorithm based on graph theory, and will
Scale collection regional ensemble is inserted into the region of initial segmentation.Image weighted graph is abstracted into table based on the image segmentation algorithm of graph theory
Show, wherein image is made of vertex set V and side collection E, G=(V, E), vertex v ∈ V, is single pixel in this programme, even
Meet the side (v of an opposite vertexesi,vj) ∈ E have weight w (vi,vj), indicate the dissimilar degree between vertex.When initialization each
Pixel is all a vertex, then gradually merges and obtains a region.The foundation of merging is the distance between two o'clock.
Wherein, region merging technique submodule, for carrying out level regions merging, and the intermediate region group that merging process is generated
At the regional ensemble of scale collection.
In tradition image analysing computer pixel-based, pixel is most basic analytical unit, often with each wave band of pixel
The essential characteristic of gray value expression pixel.Spectral signature expressed by these gray values is also the classification for carrying out image, target knowledge
Not, the basis of the applications such as feature extraction, quantitative inversion.In object-based image analysing computer, region is most basic analysis list
Member, each region includes multiple independent pixels, and each pixel spectra feature is not fully identical in region, therefore using single
Gray average when carrying out expression region, be often lost the statistical information of pixel in region.Since different regions is shown
Different spectrum, texture and statistical information etc., when individually carrying out region description using a certain category feature, will cause can between region
The reduction of distinction causes Zonal expression inaccurate, even wrong expression.Therefore, it is necessary to from spectrum, texture, space etc.
Various aspects model region, form the region merging technique cost criterion that multiple features combine.
During merging creation scale collection using level regions, the merging cost in region determines the suitable of region merging technique
Sequence, and finally decide the segmentation quality of image.The cost for measuring merging in this programme using heterogeneous change, from light
The incrementss of region merging technique bring heterogeneity have been measured in terms of spectrum and shape feature two:
Ci,j=wcolor×Δhcolor+wshape×Δhshape (4-1)
wcolorAnd wshapeIt is to adjust the coefficient of spectrum and shape feature weight, and w respectivelycolor+wshape=1.Δhcolor
Indicate the incrementss of the heterogeneity of the spectrum as caused by region merging technique, is defined as:
Wherein i and j indicates two regions, and i ∪ j indicates the region after i and j merging, and n indicates the area in region, wcIndicate c
The weight of wave band, σcIndicate the standard deviation of wave band c.
ΔhshapeThe incrementss for indicating the heterogeneity of the shape due to caused by region merging technique, by the shape smoothness and shape in region
It is formed in terms of shape compactedness two, calculation formula is as follows:
Δhshape=wcompt×Δhcompt+wsmooth×Δhsmooth (4-3)
wcomptAnd wsmoothIt is to adjust the coefficient of compact shape and shape smoothness weight, and w respectivelycompt+wsmooth=
1。ΔhcomptIndicate the value added of compactness, Δ hsmoothThe value added for indicating smoothness, is defined as follows:
Wherein l indicates that perimeter, n indicate that area, b indicate the area of minimum circumscribed rectangle.
Checkrow wire introduction module, for the merging sequence during region merging technique to be carried out tissue, building regional level is closed
System establishes the scale index in region, realizes the scale collection expression of image.
The characteristic that scale collection structure is different from general rule hierarchical tree is: for each node, having corresponding
Scale parameter maps that scale axis.What the scale parameter indicated is the scale that region representated by node occurs.It is generating
During scale collection, each region merging technique can all generate a new region, correspond to the node that scale is concentrated, at this point,
It is indexed using the correlated condition of region merging technique as scale parameter.Therefore, theoretically scale should be it is gradually incremental,
Specifically be exactly father region scale it is centainly bigger than the scale of subregion.Scale is incremented by but also the section of large scale is to figure
The expression of picture is more coarse, and the section of lower scale is more fine and smooth to the expression of image.In order to guarantee the monotonic increase of scale,
Scale of this programme during region merging technique using summary table up to error as scale collection node indexes.Institute is merged for kth time
Its scale of scale collection node index of acquisition calculates shown in following formula:
Wherein CiIndicate i-th merge region merging technique cost ,+1 be in order to make express error amount less than 1 when scale parameter
Greater than 0.It for the Area Node obtained when initial segmentation, is had existed before carrying out region merging technique, therefore at the beginning of it is set
The scale index for all areas divided that begin is 0.
Level regions merge using initial segmentation as starting point, sometimes even carry out region conjunction by starting point of individual pixel
And.When image is very big, the number of prime area is very more, and calculation amount is very big.In addition, it is one that level regions, which merge,
A global optimization's process, itself is accelerated parallel it is extremely difficult, can be with therefore when handling large format remote sensing image
Using the Fast Segmentation of the double-deck scale collection model realization image.
Specifically, image is divided into a series of fritter first, and establish respectively when handling large format remote sensing image
Scale collection, then select a moderate scale parameter, retrieval, splicing obtain a corresponding full figure Image Segmentation as a result, and with
Based on this, the scale collection of a covering full figure is constructed.When the segmentation scale of needs is smaller, concentrated from each piece of corresponding scale
Corresponding segmentation result is obtained, and carries out the splicing of segmentation result；When the segmentation scale of needs is larger, from second layer scale collection
It is middle to obtain corresponding segmentation result.
The initial cross-section of second layer scale collection is spliced by the result of bottom scale collection, and connection upper layer and lower layer knot is played
The effect of structure, referred to as connection section.In order to adapt to different heterogeneous criterion, intensity value ranges, wave band number of image etc. are avoided
Caused by influence, this programme direct corresponding scale parameter in setting connection section, and being come with the region average area of setting
Calculate specific scale parameter.The region area of setting is converted to the number in region first, formula is as follows.
WiAnd HiIndicate the width and height of image, NsIndicate the number in region, SavrIndicate region area.By bottom scale
All nodes are concentrated to be ranked up according to scale size.For containing NsThe initial cross-section in a region, the ruler constructed using it
Degree collection will contain 2Ns- 1 node.It estimates and retains the biggish 2N of sequence node mesoscales- 1 node is found out corresponding at this time
Scale parameter Slink, and as scale threshold value, corresponding segmentation result is retrieved from each piecemeal, is spliced and is connected
Connect section.
The scale collection information of currently processed image is compressed in the specific format and is stored as binary stream data, is convenient for subsequent shadow
As the quick reading and parsing of segmentation.In the data memory format of scale collection, the result (distance of swimming code of initial segmentation is recorded respectively
Compressed format) and merging process node data.The nodes records of merging process are when two regions time merged in this document,
Newly-generated zone number and this time merge the scale at place.
The merge node of scale collection is successively parsed, each node can be disassembled as a binary tree, wherein be merged
Two regions are as branch, and the region merged is as root, and combined scale is as index.When all merge nodes are handled
After finishing, a complete scale collection structure is obtained.
Given scale parameter λ, successively judges whether the scale-value of all merge nodes is less than λ, and be marked；Establish one
The list (original marking in Fig. 4-5) of a initial segmentation region labeling modifies the mapping of label then according to combined node
Table obtains new mapping table (mapping table in Fig. 4-5)；Finally, new mapping table is ranked up, it will sequentially make in order, obtain new
Mapping table (new mappings table in Fig. 4-5)；Finally, the region labeling of initial segmentation is subjected to mapping processing according to new mapping table,
Obtain corresponding segmentation result under the scale.
In actually segmentation, it usually needs determine an optimal segmentation scale.For the segmentation result of each scale, divide
Homogeney, interregional heterogeneous two angles carry out quantitative measurement to the quality of segmentation not out of region；On this basis, it utilizes
The objective function of Image Segmentation superiority and inferiority is measured in homogeney and heterogeneous building.The maximum value of objective function corresponds to optimal segmentation
Scale.
The homogeney of segmentation result is the weighted sum of all areas homogeney under the segmentation result under a certain scale.Utilize mark
Quasi- difference carrys out the homogeney of gauge region.Region homogeney is better, then interior pel intensity profile in region is more concentrated, and standard deviation is got over
It is small；Homogeney is poorer, then interior pel intensity profile in region is more dispersed, and standard deviation is bigger.
Scale SλThe calculation formula of corresponding homogeney is as follows:
Wherein V (Sλ) indicate scale SλUnder region homogeney；M indicates the number of pixels of image, and N indicates the cross section
Number；niIndicate scale SλLower region RiArea；δiIndicate scale SλLower region RiThe weighted sum of each wave band standard deviation.It is counting
Slide rule degree SλWhen corresponding standard deviation, area can be made biggish in this way as weight using its area in each region
Specific gravity shared by region increases, and avoids the merging of smaller area to unstable caused by total.V(Sλ) smaller, illustrate whole
The homogeney in a section is better, heterogeneous smaller.
With the continuous progress of region merging technique, region similar in property is constantly merged, and the homogeney in region constantly drops
It is low, V (Sλ) constantly increase, finally reach maximum value.
Heterogeneity is used to measure the otherness between segmentation gained region.Segmentation result is better, then heterogeneous bigger, otherwise
It is heterogeneous smaller.Statistical method using not blue index (Moran ' s I, MI) as interregional heterogeneity.
Wherein n indicates the number of object；wi,jIndicate region RiAnd RjNeighbouring relations, the w if adjacenti,j=1, otherwise
wi,j=0；xiIndicate RiMean value,The spectrum average of full figure is indicated when MI value is higher, illustrate between region similitude compared with
Height, can discrimination it is poor, it is heterogeneous poor；When otherwise MI value is lower, can discrimination it is good, heterogeneity is good.
During region merging technique, a pair of of adjacent area of every merging can obtain the interregional heterogeneity of current cross-section.
With the continuous progress of region merging technique, region similar in property is merged, and interregional heterogeneity is caused to be continuously increased, and space is from phase
Closing property constantly reduces.And in general, decline trend, as region merging technique is continuously increased, autocorrelation is presented in not blue index
Reduced rate obviously increases, and the characteristic of convex curve is presented in general trend.Occur in the later period of region merging technique not blue index value
Certain fluctuation, this is because in the later period of region merging technique, since number of regions is less, so that statistical property and unstable.
In the present solution, being tied using global optimum's index (Overall Goodness F-measure) as segmentation is measured
The objective function of fruit quality, calculation method are as follows:
Wherein MInormAnd LVnormHeterogeneity and homogeney index respectively after segmentation result normalization.Parameter alpha is input
Adjustment parameter, the tendentiousness of model output is adjusted, as α > 1, it is intended to over-segmentation, otherwise tend to less divided.In a system
In the Image Segmentation result of column, so that the maximum result of OGF value is considered as optimal scale.The model can overcome conventional metric
Parameter is to the dynamic range of image, the susceptibility of atural object classification.The optimal ruler for being directed to different atural objects is obtained by adjustment parameter α
Degree, and there is preferable robustness.In the present solution, α value is 1.
Specifically, information extraction modules include: samples selection submodule, feature extraction submodule, elements recognition submodule.
Wherein, samples selection submodule is used to choose the training sample image patch of every class atural object.
In traditional image analysis methods pixel-based, since image resolution is lower, in this case, for
The selection of training sample only needs to select representative region, all regard pixel therein as training sample.So
And in the image analysing computer based on scale collection, scale collection contains the region of different scale Image Segmentation, and the area of different scale
There are hierarchical relationships between domain, thus the sample for the classification of scale collection is different from the object-based image analysis methods of tradition
Sample.
In the present solution, all node-classifications of image scale collection binary tree are related to the node of different scale, therefore sample
Selection needs to select sample appropriate from node by adjusting scale.Classification multiple dimensioned for the image of object-oriented, such as works as
When corresponding target is completely divided into a complete area, the feature extracted to it can preferably reflect the spy of atural object
Property；When atural object is divided relatively more broken, in the spectral signature for being able to reflect partial region to a certain degree, broken point
Class result can be eliminated in subsequent processing；When atural object is by less divided, the different atural object of multiclass is divided into one
In region, necessarily cause can not modified classification error, therefore the basic principle of samples selection is to can choose the area of over-segmentation
Domain, but less divided region cannot be selected.
In the image analysing computer based on machine learning, sample is accurate, sufficient quantity is the key that guarantee one of its accuracy rate.
By selecting principle above, the accurate sample in part can be obtained, however the quantity of sample is but difficult to ensure.Work as sample
When this is very few, the classifier and unstable of its training is utilized, it is difficult to obtain preferable performance.In order in less input sample
In the case of, obtain more stable classification, the upper and lower level that typical geography national conditions elements recognition system utilizes region to concentrate in scale
Relationship and selected characteristic of division, devise the recursive rule of classification samples.
Scale concentrates changing from small to big with scale, and region is always developed from over-segmentation to less divided.Thus, for any
Selected sample, the downward recursion of hierarchical relationship concentrated by scale or upward.When the characteristic of division of selection only includes spectrum, line
When the features such as reason, selected sample is characterized between the feature of its child node that linear weighted function is got, and has between feature at this time
Having can linearly can recurrence relation.At this point it is possible to which selected sample is extended to training in all child nodes that scale is concentrated
Sample.When the characteristic of division of selection includes shape feature, relationship is complicated between the feature of selected sample and the feature of child node,
Have no the relationship of linear weighted function.At this point, selected sample can not upwardly or downwardly recursion.
Wherein, feature extraction submodule is used to extract spectral signature, textural characteristics, geometrical characteristic and the index spy of image patch
Sign.
Classified using the method based on machine learning to all nodes that scale is concentrated, the provincial characteristics description of selection
Operator mainly includes following several:
Spectral signature: spectrum histogram and mean value, standard deviation has been respectively adopted.Histogram can recorde heterogeneity at it
In distribution situation, be widely used in image recognition application in.The descriptive power of histogram depends on the quantization number of histogram,
The spectrum in each channel is quantified as 32 values in this programme, is quantified in the hope of spectrum flat between ability and computation complexity
Weighing apparatus.The mean value and standard deviation of spectrum has recorded the population mean situation and details of spectrum in region with the feature of smaller dimension
Difference.
Textural characteristics: it for the texture in region, can be described in terms of the mode of texture and the intensity of texture two.
In this project, two above aspect is described using local binary pattern and local inverse differential are other.Partial binary
Mode describes the structure and intensity of texture by the spatial relationship and intensity difference of regional area pixel, and with the side of statistical analysis
Method establishes the model of texture description.It mainly include homogenieity (Homogeneity), contrast (Contrast), entropy (Entropy)
Three classes.
Homogenieity feature shows the homogenieity of image part, and value range is [0,1], when texture is more regular or surface ratio
It is worth when smoother bigger.
Contrast metric effectively detection image contrast, extraction object marginal information can enhance the information such as lineament, value
Range is [0, ∞].Neighboring gradation varies less inside open grain, original in practical application due to the compression of grey level
The lesser pixel of gray scale difference becomes same grey level upon compression in image, and contrast value at this time is close to 0.And thick
There are certain differences for the pixel gray scale of texture edge, even if this pixel is smaller to the probability of appearance, but amplified contrast
Value can also be allowed to effectively distinguish with the region inside open grain.
Entropy is characterized in that the measurement of image inconsistency, value range are [0, ∞].When the texture of image is extremely inconsistent, ash
The value for spending each element in co-occurrence matrix will be less than normal, then image has biggish entropy.If image does not have any texture, gray scale is total
The raw almost nil battle array of matrix, then entropy is close to zero.If image is filled with close grain, the numerical approximation of pixel pair is equal, then
The entropy of the image is maximum.If being dispersed with less texture in image, the numerical value difference of pixel pair is larger, then the entropy of the image
It is smaller.
Shape feature: including area, length, shape index, density, asymmetry.
Area features, that is, atural object element area of plane threshold value, value range are [0, the size of map sheet].For noly
The data of reference are managed, the area of single pixel is 1.As a result, the area of an imaged object is exactly the number for constituting its pixel
Amount.If image data has Geographic Reference, the area of an imaged object is exactly that the true area of pixel covering multiplies
To constitute the pixel quantity of this imaged object.
Length characteristic, that is, imaged object length/width ratio, is obtained by bounding box approximation.
Shape index refers to the boundary length of imaged object except it upper subduplicate 4 times of area, to describe image pair
As the smoothness on boundary.Imaged object is more broken, then its shape index is bigger.
Density feature, that is, imaged object area is except its upper radius, to describe the compactness of imaged object.In pixel
Ideal compact shape is a square in the figure of grid.The shape of one imaged object is closer to square, it close
It spends higher.
One imaged object is longer, its asymmetry is higher.For an imaged object, can be similar to one it is ellipse
Circle.Asymmetry feature is the length ratio for being expressed as elliptical short axle and long axis.
Layer feature: including normalized differential vegetation index (NDVI) and normalization water body index (NDWI).
The difference of normalized differential vegetation index, that is, near infrared band reflected value and the reflected value of red spectral band is than upper sum of the two.
The index value range is [- 1,1], and negative value indicates that covered ground is cloud, water, snow etc., to visible light high reflection；0 indicates rock
Stone or exposed soil etc.；Positive value indicates vegetative coverage, and increases with coverage and increase.
Difference processing is normalized with the specific band of remote sensing image in normalization water body index, to highlight in image
Water-Body Information is the normalized ratio index based on green wave band and near infrared band.
Wherein, elements recognition submodule is used for that (such as support vector machines, k to be most based on image sample set and selected characterization method
Nearest neighbour classification, Taxonomy and distribution and random forest) realize the automatic classification of image, and export arable land, forest land, water body, resident
Ground.
Select support vector machines as classification method.Support vector machines is a kind of sorting algorithm based on machine learning, is led to
It crosses and seeks structuring least risk to improve learning machine generalization ability, realize the minimum of empiric risk and fiducial range, thus
Reach in the case where statistical sample amount is less, can also obtain the purpose of good statistical law.It is a kind of two classification model,
Its basic model is defined as the maximum linear classifier in the interval on feature space, i.e. between the learning strategy of support vector machines is
Every maximization, it can finally be converted into the solution of a convex quadratic programming problem.Using the training sample across scale selection, it is based on picture
The feature vector of spot obtains image classification result using support vector machines method.
Applicating example
The part ecological protection red line area of certain province is chosen as Applied D emonstration area, uses typical geographical national conditions elements recognition system
System and method complete the automation in arable land, forest land, water body, settlement place within the scope of about more than 2600 square kilometres ecological protection red line areas
Interpretation.Based on Objects extraction derived from system as a result, establishing demonstration area experimental data base.Database includes arable land (0100), woods
(1000) four ground (0300), settlement place (0500), water body ground mulching level-one classes.
The high score satellite image that image data for the trial production of ecological protection red line area demonstration area is 2017, amounts to 12
Width.Wherein 3 scape of high score No.1 satellite image, No. two 3 scapes of satellite image of high score, No. three 6 scapes of image of resource.The image capturing time exists
In March, 2017 to May.The size of one scape raw video is more than 30000 × 30000 pixels, therefore in order to meet representative elements
The software and hardware requirement of extraction system operation, first by 12 scape image joints, resampling is the image of 2 meters of spatial resolutions.According still further to
1:5 width rule very much, is cut into 14 width framing images for remote sensing image.
Each width image is handled respectively using the typical geographical national conditions elements recognition system in this ecological protection red line area, is interpreted
After achievement, the fusion edge fit of data is completed, forms trial production raw data.After demonstration area universe remote sensing image is cut, often
Terrain and its features feature is different in width image coverage area, is broadly divided into city and two, mountain area type.City is with artificially
It based on object, picks up comprising house, road, structures, artificial heap, meadow, forest land, a variety of atural object classifications such as water body, and house etc.
The complicated multiplicity of building textural characteristics.
Construct scale collection when, choose Battz criterion be used as region merging technique criterion, be arranged initial Image Segmentation mode for based on
The Image Segmentation of graph theory, set distance threshold value are 50, and minimum area is 30 pixels, and form parameter is set as 0.3, edge parameters
It is set as 0.5, compactness is set as 0.5,0 section average area and is set as 100.
It interprets in achievement, forest land, water body interpretation effect are best, and successful recognition rate has reached 89%, and arable land is taken second place, and reaches
82%.System, which interprets achievement and updates achievement with national conditions in 2017, to be compared as shown in Figure 6 (forest land), Fig. 6 a be national conditions in 2017 at
Fruit data, Fig. 6 b are this system automatic interpretation data.
It is concentrated in scale, scale parameter is monotonic increase, therefore each scale parameter corresponds to a unique scale collection and cuts
Face, i.e., the segmentation result of one image.Each node and its father node difference on scale axis are existing for the subregion
Scaling interval.The advantages of this structure is, when given scale parameter, can by judge region Scaling interval whether include
The scale parameter knows that the region whether there is in given scale parameter.To obtain the segmentation knot under any scale parameter
Fruit.It is described from the data structure of scale collection, the image expression-form for any scale can be expressed as, can be led to
It crosses and the cut zone intersection in scale collection space is truncated and is quickly obtained.
By intercepting the section of scale collection in different scales, may be implemented to the image multi-scale expression based on region.
If given scale interval is sufficiently small, the multi-scale expression of nearly continuity may be implemented.
In the region merging technique face of entire scale collection, there are a generation scale and disappearance scales for each area surface, i.e.,
Each area surface is present in a Scaling interval, wherein the production scale in all initial segmentation faces is 0.When needing to obtain
It when the Image Segmentation result of some scale, is indexed first by scale, obtains existing area surface serial number under current scale；Then
By region merging technique tree construction, the initial segmentation face collection of each area surface is found；Finally, merging each initial segmentation face respectively
Subset is to get the Image Segmentation result for arriving specified scale.
It is ploughed using traditional visual interpretation method to demonstration area, forest land, water body, settlement place this 4 class elements recognition
Take around 100 work day people；Image is carried out to demonstration area using typical geographical national conditions elements recognition system of the invention to divide automatically
About 17 hours people of class time-consuming, editting and processing are ground mulching performance data on the basis of interpreting achievement herein, about time-consuming 80 works
Day people, extraction efficiency promote about 20%.That is, by the method for the invention and system can be obviously improved working efficiency.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art can easily think of the change or the replacement in the technical scope disclosed by the present invention, all should
It is included within the scope of protection of the present invention.
Claims (7)
1. a kind of typical geographical national conditions elements recognition system and method in ecological protection red line area characterized by comprising
Data processing module, for handling satellite image data；
Scale collection divides module, and the scale collection model for establishing scale collection model, and based on foundation is to the satellite image number
According to Image Segmentation is carried out, image scale collection is obtained；
Information extraction modules extract ecological protection for carrying out the information extraction of object-oriented based on the image scale collection
Red line area arable land, forest land, water body, these four types of typical geographical national conditions elements of settlement place.
2. system according to claim 1, which is characterized in that scale collection divides module and includes:
Initial segmentation submodule carries out initial segmentation to image for the Fast Segmentation Algorithm based on graph theory, and by initial segmentation
Region be inserted into scale collection regional ensemble；
Region merging technique submodule, for carrying out level regions merging, and the intermediate region that merging process is generated forms scale collection
Regional ensemble；
Checkrow wire introduction module constructs regional level relationship, builds for the merging sequence during region merging technique to be carried out tissue
The scale index in vertical region realizes the scale collection expression of image.
3. system according to claim 1, which is characterized in that information extraction modules include:
Samples selection submodule, for choosing the training sample image patch of every class atural object；
Feature extraction submodule, for extracting spectral signature, textural characteristics, geometrical characteristic and the index characteristic of image patch；
Elements recognition submodule, for realizing the automatic classification of image based on image sample set and support vector machine classification method,
And export arable land, forest land, water body, settlement place.
4. system according to claim 3, which is characterized in that the samples selection submodule is in selection training sample image patch
When, the sample image patch of corresponding atural object classification is selected under different scale.
5. a kind of typical geographical national conditions elements recognition system and method method in ecological protection red line area, which is characterized in that including with
Lower step:
Handle satellite image data；
Scale collection model is established, and the scale collection model based on foundation carries out Image Segmentation to the satellite image data, obtains
Image scale collection；
The information extraction that object-oriented is carried out based on the image scale collection, extracts ecological protection red line area arable land, forest land, water
These four types of typical geographical national conditions elements of body, settlement place.
6. according to the method described in claim 5, it is characterized in that, establish scale collection model based on the satellite image data,
And scale collection based on foundation carries out Image Segmentation, the step of obtaining image scale collection, comprising:
Initial segmentation is carried out to image based on the Fast Segmentation Algorithm of graph theory, and scale collection region is inserted into the region of initial segmentation
Set；
The regional ensemble for the intermediate region composition scale collection for carrying out level regions merging, and merging process being generated；
Merging sequence during region merging technique is subjected to tissue, constructs regional level relationship, establishes the scale index in region, it is real
The scale collection expression of existing image.
7. according to the method described in claim 5, it is characterized in that, described carry out object-oriented based on the image scale collection
Information extraction, extract ecological protection red line area arable land, forest land, water body, these four types of typical geographical national conditions elements of settlement place step
Suddenly, comprising:
Choose the training sample image patch of every class atural object；
Extract spectral signature, textural characteristics, geometrical characteristic and the index characteristic of image patch；
The automatic classification of image is realized based on image sample set and support vector machine classification method, and exports arable land, forest land, water
Body, settlement place.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811592299.9A CN109635828A (en) | 2018-12-25 | 2018-12-25 | A kind of typical geographical national conditions elements recognition system and method in ecological protection red line area |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811592299.9A CN109635828A (en) | 2018-12-25 | 2018-12-25 | A kind of typical geographical national conditions elements recognition system and method in ecological protection red line area |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109635828A true CN109635828A (en) | 2019-04-16 |
Family
ID=66077432
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811592299.9A Pending CN109635828A (en) | 2018-12-25 | 2018-12-25 | A kind of typical geographical national conditions elements recognition system and method in ecological protection red line area |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109635828A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881976A (en) * | 2020-07-27 | 2020-11-03 | 盐城郅联空间科技有限公司 | Multi-source image automatic interpretation method integrating artificial intelligence technology and big data |
| CN112669255A (en) * | 2020-11-03 | 2021-04-16 | 浙江工业大学 | Part machining surface prediction method based on space-time full convolution cyclic neural network |
| CN116188786A (en) * | 2023-05-04 | 2023-05-30 | 潍坊医学院附属医院 | Image segmentation system for hepatic duct and biliary tract calculus |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273813A (en) * | 2017-05-23 | 2017-10-20 | 国家地理空间信息中心 | Geographical space elements recognition system based on high score satellite remote sensing date |
-
2018
- 2018-12-25 CN CN201811592299.9A patent/CN109635828A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273813A (en) * | 2017-05-23 | 2017-10-20 | 国家地理空间信息中心 | Geographical space elements recognition system based on high score satellite remote sensing date |
Non-Patent Citations (2)
| Title |
|---|
| 孙家抦 等: "《文本分类中的特征选择方法研究》", vol. 1, 吉林大学出版社, pages: 224 - 227 * |
| 胡忠文: "基于区域的遥感影像多尺度表达方法与应用研究", 《中国博士学位论文全文数据库信息科技辑》, no. 2016, 15 January 2016 (2016-01-15), pages 6 - 101 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881976A (en) * | 2020-07-27 | 2020-11-03 | 盐城郅联空间科技有限公司 | Multi-source image automatic interpretation method integrating artificial intelligence technology and big data |
| CN111881976B (en) * | 2020-07-27 | 2021-02-09 | 盐城郅联空间科技有限公司 | Multi-source image automatic interpretation method integrating artificial intelligence technology and big data |
| CN112669255A (en) * | 2020-11-03 | 2021-04-16 | 浙江工业大学 | Part machining surface prediction method based on space-time full convolution cyclic neural network |
| CN112669255B (en) * | 2020-11-03 | 2024-02-13 | 浙江工业大学 | Part machining surface prediction method based on space-time full convolution cyclic neural network |
| CN116188786A (en) * | 2023-05-04 | 2023-05-30 | 潍坊医学院附属医院 | Image segmentation system for hepatic duct and biliary tract calculus |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Viana et al. | Land use/land cover change detection and urban sprawl analysis | |
| Koukoulas et al. | Mapping individual tree location, height and species in broadleaved deciduous forest using airborne LIDAR and multi‐spectral remotely sensed data | |
| Tomppo et al. | Multi-source national forest inventory: Methods and applications | |
| Zhang et al. | Object-oriented method for urban vegetation mapping using IKONOS imagery | |
| Ardila et al. | Markov-random-field-based super-resolution mapping for identification of urban trees in VHR images | |
| Bauer et al. | Satellite inventory of Minnesota forest resources | |
| CN101840581B (en) | Method for extracting profile of building from satellite remote sensing image | |
| González-Yebra et al. | Methodological proposal to assess plastic greenhouses land cover change from the combination of archival aerial orthoimages and Landsat data | |
| Matasci et al. | Mapping tree canopies in urban environments using airborne laser scanning (ALS): a Vancouver case study | |
| CN110309780A (en) | High resolution image houseclearing based on BFD-IGA-SVM model quickly supervises identification | |
| Yan | Pixel based and object oriented image analysis for coal fire research | |
| CN109635828A (en) | A kind of typical geographical national conditions elements recognition system and method in ecological protection red line area | |
| Duan et al. | Mapping the soil types combining multi-temporal remote sensing data with texture features | |
| CN112800827A (en) | Hyperspectral image classification experimental method | |
| Lefebvre et al. | Monitoring the morphological transformation of Beijing old city using remote sensing texture analysis | |
| Franklin | Land cover stratification using Landsat Thematic Mapper data in Sahelian and Sudanian woodland and wooded grassland | |
| Civco et al. | A hierarchical approach to land use and land cover mapping using multiple image types | |
| CN106991404A (en) | Ground mulching update method and system based on many source geodatas | |
| Khorram et al. | Accuracy assessment of land cover change detection | |
| Puestow et al. | Mapping of Salmon H-'-itat Parameters 1'-'Airborne Imagery and Digital Ancillary Data | |
| Provoost et al. | Remote sensing of coastal vegetation in the Netherlands and Belgium | |
| Wu et al. | Remote sensing of high resolution urban impervious surfaces | |
| Ozdarici Ok et al. | Automatic training site selection for agricultural crop classification: A case study on Karacabey plain, Turkey | |
| Lazuardi et al. | Assessment of coral reef life-form classification scheme using multiresolution images on Parang Island, Indonesia | |
| Tarantino et al. | Plastic covered vineyard extraction from airborne sensor data with an object-oriented approach |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190416 |