CN101706950A - High-performance implementation method for multi-scale segmentation of remote sensing images - Google Patents
High-performance implementation method for multi-scale segmentation of remote sensing images Download PDFInfo
- Publication number
- CN101706950A CN101706950A CN200910157530A CN200910157530A CN101706950A CN 101706950 A CN101706950 A CN 101706950A CN 200910157530 A CN200910157530 A CN 200910157530A CN 200910157530 A CN200910157530 A CN 200910157530A CN 101706950 A CN101706950 A CN 101706950A
- Authority
- CN
- China
- Prior art keywords
- algorithm
- segmentation
- image
- yardstick
- remote sensing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention provides a high-performance implementation method for multi-scale segmentation of remote sensing images, and in particular a method for implementing fast and multi-scale image segmentation of large amount of remote sensing images and establishing the structural relationship of segmentation results in the process of high-resolution remote sensing image information extraction. The method comprises the following steps: on the basis of analyzing the implementation procedure of the algorithm and finding out the computing-intensive segment of the algorithm, realizing the parallel segmentation of the algorithm-intensive segment based on MPI and OMP models, and carrying out date join on the parallel segmentation results; implementing the multi-scale image segmentation by storing the initial segmentation results of the algorithm and merging the subsequent multi-scale remote sensing images; and establishing a multi-scale object topological relation model. The generated multi-scale segmentation region and corresponding hierarchical relationship can be applied to various services; and the corresponding implementation method is applicable to various segmentation algorithms such as mean shift and the like, and can greatly improve the data amount processed by algorithm and the processing efficiency.
Description
Technical field
The present invention relates to remote sensing image treatment technology and remote sensing image information extracting method, specifically, the multi-scale division and the high-performance implementation method thereof that relate to remote sensing image, and the foundation of the yardstick topological relation of segmentation result, the present invention realizes applicable to the high-performance of multiple remote sensing image dividing method.
Background technology
OO remote sensing images analysis method has been widely used in application such as current remote sensing image processing, information extraction and analysis, Target Recognition, classification, and a kind of important method that has become remote sensing image, particularly high-resolution remote sensing image processing and analyzed.Wherein, the extraction of object, feature calculation and signature analysis are the important steps of processes such as remote sensing image information extraction and Target Recognition, and the extraction of object then is basis wherein, and its implementation procedure mainly adopts the dividing method of remote sensing image to realize.Particularly for the high-resolution remote sensing image information extraction, the demand of Target Recognition and OO classification of remote-sensing images etc., because high resolution image has abundant atural object geometry, texture information, make people may carry out the combined process of face of land information by different observation yardsticks, thereby press for people and remote sensing image is cut apart and extracted corresponding cutting object (segmented per-parcel), corresponding multi-scale division list of references comprises Dorin Comaniciu, Mean Shift:ARobust Approach Toward Feature Space Analysis[J] .IEEE Transactions On Pattern Analysis andMachine Intelligence.2002 24 (5) .603-619, Mukherjee D.P., Ore image segmentation by learningimage and shape features.Pattern Recognition Letters 2009 (30): 615-622 etc.
Information extraction for remote sensing image, processes such as classification and Target Recognition, many times need from same image, to extract different target atural object, and different atural object correspondences different best identified yardsticks, as discern highway in the image, the greenery patches, lake etc. and identification airport, the square, the harbour, needed yardstick such as park is different, so the different demands that need carry out multi-scale division and adaptation to image, corresponding list of references comprises Cl á udio Rosito Jung, Unsupervised multiscalesegmentation of color images, Pattern Recognition Letters, 2007, Chen, J., Perceptually-tunedmultiscale color-texture segmentation.In:Internat.Conf.Image Process., 2007.II:921-924 etc.
In the multiple dimensioned image cutting procedure, because the remote sensing image data amount is big, characteristics such as algorithm complexity, therefore corresponding partitioning algorithm need consume a large amount of internal memories, the cutting procedure of remote sensing image need consume a large amount of computational resources and cause efficient very low on the one hand, a lot of on the other hand images also can't be cut apart because need very big internal memory, and therefore a lot of actual application all need to make up the high-performance image that is suitable for and cut apart environment and realize cutting apart of remote sensing image; Demand for the multi-scale division of remote sensing image, owing to need the multiple dimensioned of segmentation result, therefore need to realize fast the multi-scale division (i.e. Qu Yu multiple dimensioned merging) of image, and set up level corresponding relation between different scale, be convenient to carry out between level reusing of yardstick conversion and corresponding knowledge, realize the parallelization high-performance treatments of partitioning algorithm, corresponding list of references comprises Tilton, J.C.Image segmentation byiterative parallel region growing with applications to data compression and image analysis, Institute of Electric and Electronic Engineer, 1998, Bing Zhang, robust parallel segmentation ofBASAG-GAUSSLAN MRFS with uncertain parameters, ICME, pp.92,2001 IEEE InternationalConference on Multimedia and Expo (ICME ' 01), 2001 etc.
Aspect the high-performance multi-scale division algorithm of remote sensing image, the research that is separated from each other at present visible patent/document is more, and high-performance realizes the multi-scale division of remote sensing image and it is all few to set up research and the practicable solution of topological relation fast, does not more have the while to set up a plurality of yardstick structures and its corresponding topological relation of object hierarchy structure and corresponding expression separately fast.
Summary of the invention
The purpose of this invention is to provide a kind of high-performance implementation method of cutting apart at remote sensing image, particularly the multi-scale division at high-resolution remote sensing image provides a kind of algorithm flow of multi-scale division fast and effectively, and makes up the level corresponding relation between the multi-scale division result simultaneously.
Thinking of the present invention is: select a kind of based on region growing and the remote sensing image dividing method that merges principle, merge section by computation-intensive section and the yardstick of finding out algorithm after the algorithm principle analysis, adopt high performance parallel image division method of the present invention to carry out corresponding image initial segmentation, the result who adopts piecemeal mechanism walk abreast to cut apart is carried out the data stitching, make up vector data object (and attribute), and the result of initial segmentation is stored; Formulate yardstick and merge rule, and on the basis of the initial segmentation result of storage, adopt iterative algorithm to carry out yardstick to merge,, and and then set up the topological hierarchical relationship of object between different scale until the image cutting procedure of finishing all yardsticks.
Basic image division method of the present invention can be selected watershed segmentation method, mean shift segmentation method, multiresolution dividing method, division merging method etc., these methods have characteristics such as boundary tracking is accurate, algorithm is stable comparatively speaking, can give play to bigger advantage in conjunction with the present invention again.
Technical scheme of the present invention provides the method that high-performance realizes and topological relation is set up of remote sensing image multi-scale division, it is characterized in that comprising following implementation step:
1) dividing method of selected remote sensing image is analyzed the algorithm condensed section and the yardstick merging section of respective algorithms;
2) the algorithm condensed section is carried out transformation based on the high-performance calculation method, can adopt based on MPI, OMP or model that the two is integrated and realize that implementation adopts the data parallel segmentation strategy;
3) the block parallel segmentation result being carried out the data cut-off rule sews up;
4) the high-performance image being cut apart intermediate result stores in order to merging in follow-up yardstick;
5) according to the requirement of multi-scale division, on the basis of step 4), merge rule and carry out the yardstick merging, generate not homotactic multi-scale division image according to yardstick;
6) the yardstick merging process of iteration step 4 is up to the generative process of finishing all yardstick split images; Carry out the object label of different scale segmentation result, make up corresponding vector quantization object, and the expression that adopts " object-attribute " is stored to the individual features of the variant cutting object mode with attribute in the corresponding vector quantization object;
7) set up topological hierarchical relationship between each yardstick, be convenient to realize that the image cutting object strides the object relationship visit and the information inquiry of yardstick, finish cutting procedure.
Above-mentioned implementation step is characterised in that:
Need in the step 1) selected remote sensing image partitioning algorithm is carried out analysis that algorithm condensed section and yardstick merge section and defines, and carry out step 2 on this basis) the high-performance of algorithm condensed section realize and the multiple dimensioned merging of step 6) realizes.
Step 2), be that the algorithm condensed section is carried out the implementation that the high-performance image is cut apart 3), need to adopt the image data high-performance of " deblocking-multiprocessors parallel processing-segmentation result merges " to cut apart flow process,, after disposing on each nuclear, carry out corresponding data line again and sew up according to carrying out deblocking and distribution according to the CPU check figure that is possessed on current available MPI node number and each node to eliminate the influence of separator bar.
The storage of initial segmentation result in the middle of step 4) realizes.
Step 5), 6), 7) carry out the merging of yardstick according to user's setting, make up the multi-scale division result and also set up object hierarchy structural topology relation.On the basis of step 4) intermediate result, different blocks are carried out the piece mark, and the syntople inquiry is also set up inquiry linked list, generates the syntople figure of different objects under the same yardstick, and and then by the object vector topological relation between mapping relations traversals different scale, set up topological relation figure.The image data segmentation result of different scale has the conforming characteristics at yardstick edge after yardstick merges.
The present invention compared with prior art has following characteristics: the abundant computational resource of appliance computer, realize the cutting procedure of remote sensing image in a kind of mode efficiently, on the basis of initial segmentation result, carry out the batch circulation merging of multiple yardstick and generate the multi-scale division result fast, generate the hierarchical structure topological relation between each yardstick simultaneously according to merging rule.High-performance data involved in the present invention calculates and yardstick hierarchical topology relation is not only applicable to the average drifting algorithm, also is fit to other algorithms based on region growing and merging, as watershed segmentation algorithm, multiresolution partitioning algorithm etc.
Description of drawings
Fig. 1 is a high-performance remote sensing image multi-scale division method synoptic diagram
Fig. 2 is the high-performance realization flow figure that multiple dimensioned remote sensing image is cut apart
Fig. 3 is the high-performance implementation method figure in the image initial segmentation
Fig. 4 is the network topology structure figure that the high-performance image is cut apart
Fig. 5 is based on the suture line two class intersection point synoptic diagram of the data segmentation result of method of partition
Fig. 6 is three kinds of situation synoptic diagram that generate time block in the sewing process of deblocking result
Fig. 7 is that deblocking (two) is cut apart and cut-off rule is sewed up synoptic diagram
Fig. 8 is different block subregion syntople synoptic diagram
Fig. 9 is based on variant subregion syntople lookup method figure among Fig. 8
Figure 10 is the expression chain hoist pennants of Fig. 9 data structure in calculator memory
Figure 11 is a somewhere SPOT5 image multi-scale division effect synoptic diagram
Embodiment
Fig. 1 is a high-performance remote sensing image multi-scale division method synoptic diagram, wherein the main part of cutting apart at image adopts the parallel high-performance calculation mode of cutting apart to realize, partly adopt circulation yardstick merging method to realize in multi-scale division, can realize the multi-scale division of image efficiently.The specific implementation techniqueflow of the high-performance implementation method of multiple dimensioned remote sensing image multi-scale division of the present invention has comprised 12 processing units as shown in Figure 2 among Fig. 2.Wherein, the course of work of the high-performance multi-scale division of image is (this embodiment is an example with the mean shift segmentation algorithm):
Image data finishes (as denoising, enhancing etc.) after the pre-service in early stage, and according to the mean shift segmentation algorithm principle, when the iterative process that data point moves to sample average, x ← m (x) forms the average drifting algorithm.In the iterative process x the position of process, promptly sequence x, m (x), m (m (x)) ... } be called the track of x.The direction of average drifting is always pointed to the place with maximum local density, and at density function maximum value place, drift value goes to zero,
So the average adjustment algorithm is the quick ascent algorithm of a kind of self-adaptation, it can find maximum local density somewhere by calculating, and carries out " drift " to its position along corresponding track.After selected gaussian kernel, have:
Show on the feature space of image, have:
X wherein
sBe the space segment of eigenvector, x
rBe the color part of eigenvector, k (x) uses identical nuclear, h in space and color gamut
s, h
rIt is wide to be respectively nucleus band, and C is corresponding normalization constant.By the wide parameter h=(h of control nucleus band
s, h
r) decide segmentation precision, so the C of the processing unit among Fig. 2 and D determined this algorithm cut apart the corresponding bandwidth parameter.Therefore processing unit E shown in Fig. 2 is the algorithm condensed section of this algorithm, adopt the high-performance treatments environment of similar integrated MPI parallel C luster structure shown in Figure 3 and the multi-core parallel concurrent model of OMP, and the method for employing deblocking is carried out.
For the average drifting image partitioning algorithm of gaussian kernel, the condensed section of algorithm should belong to based on spatial domain (h
s) and color gamut (h
r) the mean filter process, as the E processing unit among Fig. 2, therefore we have adopted the parallel image computing environment based on MPI (Message Passing Interface) model and OMP (OpenMP API) model as shown in Figure 3 to realize this computation-intensive section, and seldom need between each process nuclear can effectively improve raising the efficiency of computation-intensive section alternately.Its step is at first to need to calculate to can be used for the available CPU check figure certificate that image is cut apart, and its computing method are:
N
part＝∑N
nodeN
core
N wherein
PartFor data need be carried out the number of piecemeal, N
NodeBe MPI node number available in the system, N
CoreCPU check figure for the pairing computing machine of each MPI node.Therefore image data to be split at first need be divided into N
PartPart, and being dispensed to the mean filter process of carrying out corresponding data on each CPU process nuclear of system respectively, shown in the MPI/OMP among corresponding C PU process nuclear similar Fig. 4, network topology result is similar shown in Figure 4.
Illustrated the CPU process nuclear structure that can be used for cutting apart among Fig. 4, available N in the system
PartBe 8 (node A is 2, and Node B is 2, and node C is 4).The data block of dividing after carrying out mean filter on the different CPU process nuclear, is carried out data by host node again and is sewed up respectively.Fig. 5 has illustrated two class intersection points in the data sewing process.The point that the block profile of both sides crosses at " suture line " is called a class intersection point, and two edges of image block also are this type of intersection point; And one-sided profile and its intersection point are called b class intersection point.Line segment between two adjacent a class intersection points is called the category-A line segment, is shown by the thick line segment table among the figure, in addition then is called the category-B line segment.
Block for category-A line segment both sides can directly merge according to spectral value, area, form etc.; But for the category-B line segment, situation is more complicated then, because corresponding block is arranged also irregular following, if simply all pieces are merged, then caused less divided, so this paper utilization is to the purpose that reaches " inferior cutting apart " elimination " suture line " of respective regions.This paper so-called " inferior cutting apart " just at first finds out the cut zone combination relevant with the category-B line segment, forms some little sub-image pieces, and it is cut apart, again the result is filled out back in the whole segmentation result, do not have too big extra operand like this, also can make the credible result of sewing up algorithm.Roughly be divided into as shown in Figure 6 three class situations for seeking the inferior block relevant with the category-B line segment, preceding two kinds of situations are that several a classes and b class intersection point, no category-A line segment are arranged between two a class intersection points; The third situation then is to have comprised the category-A line segment in a complete category-B line segment.For these three kinds of situations, need find the combination of block according to different searching algorithm, and generate the sub-image piece according to its profile.
The stitching algorithm of data line is as follows:
1) traversal " suture line " is found out all a class intersection points { Pa} and b class intersection point { Pb};
2) connect all two adjacent Pa as the category-A line segment LA}, all the other are as category-B line segment { LB};
3) category-A is sewed up: merge all { blocks of LA} both sides, block is obtained by the FloodFill algorithm search, the block of both sides is noted by abridging respectively and is FFt (LA) and FFb (LA), the subscript t of FF and b represent that respectively the block of upside and downside (is the example explanation with level " suture line " here, therefore block is in both sides up and down), the index value after the merging determines that by formula (1) wherein prefix T represents label, AR represents area, and the label after therefore merging should be the weighted mean of both sides piecemeal;
4) category-B is sewed up: search from left to right run into also do not travel through the point on the LB} PLB} then carries out following iterative algorithm:
1. with { P
LBOn first P
LBAs seed points FloodFill algorithm search upside block FFt (P
LB), and itself and all abutments of " suture line " are done mark, formation point set { PBt};
2. so that { point among the PBt} is as seed points search downside block FFb (PBt), and when running into it with the binding site of " suture line " during search, { PBt} then therefrom deletes this point, and deletion { P if belong to
LBMiddle corresponding point, otherwise add point set to { among the PBb}; So until { PBt} is empty, after this if { PBb} is not empty, then carries out 3., otherwise carries out 4.;
3. so that { point among the PBb} is as seed points search downside block FFt (PBb), and when running into it with the binding site of " suture line " during search, { PBb} then therefrom deletes this point, and deletion { P if belong to
LBMiddle corresponding point, otherwise add point set to { among the PBt}; So until { PBb} is empty, after this if { PBt} is not empty, then carries out 2., otherwise carries out 4.;
4. will be 1.-the 3. continuous cut zone merging of formation, it is added to { in the Mask} image set, and form with its corresponding image in the taking-up original image and time cut apart image, remainder is filled by black, and the inferior image of cutting apart is added to and time cuts apart image set { among the PostImages}.
This step can all situations among Fig. 6 all with respect to, none is omitted ground and forms and time cut apart image set.
5) incite somebody to action the image among the PostImages} is cut apart again, and according to { Mask} fills back inferior segmentation result in the whole segmentation result, and algorithm is finished.
Fig. 7 has illustrated to use above-mentioned algorithm and has carried out the design sketch that data line is sewed up.Wherein A is a raw video, and B is the segmentation result behind the image block, and C is for removing the stitching algorithm effect of piecemeal line, and D is for adopting the different cutting objects of equal value representation, and E is for finally cutting apart the design sketch of object vector stack raw video.
Finish after the high-performance calculation part among Fig. 2, we realize to the multi-scale division shown in Fig. 2 further that again its method is that the initial segmentation result of the F processing unit among Fig. 2 is carried out based on the further merging process of giving dimensioning.G among Fig. 2, H, four processing units of I, J are exactly to finish the process that multiple dimensioned result merges.Because the algorithm complex of yardstick merging process in whole algorithm in this step is very low, therefore do not need the merging process in step to adopt parallel processing mode to this, only need be unified in that the unified yardstick that carries out merges and gets final product on the host node.
Set yardstick for each, yardstick will form similar image data block shown in Figure 8 after merging, its expression in calculator memory will be unified the mode of color data link table, promptly adopt the mode of the zones of different searching neighboring region among Fig. 9, again in internal memory with chained list expression shown in Figure 10, the topological relation of being convenient between different scale generates.The association attributes value sequence comprises that the spectrum average, form parameter, texture expression etc. of block are used for the property value of follow-up classification, identification and analysis etc. among Figure 10, the neighboring region sequence then is the neighboring region according to the digraph traversal of similar Fig. 9, the yardstick corresponding relation then is the block of a yardstick and next yardstick on this zone correspondence, can carry out the design of respective data structures according to actual needs.
Figure 11 has illustrated the multi-scale division effect synoptic diagram of the somewhere SPOT5 image data that carries out based on the method among the present invention. wherein ABCD is respectively yardstick 100,500,1000,2000, selected red area is partitioned into 7 respectively under four kinds of yardsticks, 4,3,1, and partitioning boundary is corresponding fully between variant yardstick, can set up topological hierarchical relationship between yardstick. the image data partitioning boundary of different scale is consistent, and can set up the topological layer aggregated(particle) structure on this basis, enrich data structure shown in Figure 10. in the objectification analysis of reality, in classification and the identifying, subsequent algorithm can be selected a kind of suitable yardstick and analyze on this basis and feature calculation in the segmentation result of different scale, also can carry out the conversion between yardstick, can go forward one by one according to the longitudinal layer structure that the topological hierarchical relationship of having set up carries out object during conversion, realize various more efficiently operations.
Example of the present invention is realized on the PC platform, the experiment proved that, the present invention can access the segmentation result that is preferably desirable, and than conventional method efficiency improvement largely (the present invention reaches 6.5 times for the raising of example image splitting speed) is arranged.Mentioned method can be widely used in the processes such as OO image processing, analysis, classification, identification of high-resolution remote sensing image, as large-scale application such as national land resources surveies for the second time among the present invention.
Claims (3)
1. the high-performance implementation method of a remote sensing image multi-scale division is characterized in that comprising following step:
Step 1, the dividing method of selected remote sensing image is analyzed the algorithm condensed section and the yardstick merging section of respective algorithms;
Step 2 is carried out transformation based on the high-performance calculation method to the algorithm condensed section, can adopt based on MPI, OMP or model that the two is integrated and realize that implementation adopts the data parallel segmentation strategy;
Step 3 is carried out the data cut-off rule to the block parallel segmentation result and is sewed up;
Step 4 is cut apart intermediate result to the high-performance image and is stored in order to merging in follow-up yardstick;
Step 5 according to the requirement of multi-scale division, merges rule according to yardstick and carries out the yardstick merging on the basis of step 4, generate not homotactic multi-scale division image;
Step 6, the yardstick merging process of iteration step 4 is up to the generative process of finishing all yardstick split images; Carry out the object label of different scale segmentation result, make up corresponding vector quantization object, and the expression that adopts " object-attribute " is stored to the individual features of the variant cutting object mode with attribute in the corresponding vector quantization object;
Step 7 is set up the topological hierarchical relationship between each yardstick, is convenient to realize that the image cutting object strides the object relationship visit and the information inquiry of yardstick, finishes cutting procedure.
2. selected image division method according to claim 1, it is characterized in that this method is OO multi-scale division method, the basic dividing method of selecting is for cutting apart based on the region growing and the remote sensing image of merging method, and passing through the multi-scale division process that control merges yardstick realization remote sensing image, selectable basic dividing method can comprise that watershed segmentation (watershed), mean shift segmentation (mean shift), multiresolution are cut apart (multi-resolution), (split and merge) method etc. is cut apart in the division merging.
3. the parallel high-performance dividing method of algorithm condensed section according to claim 1, it is characterized in that if algorithm has available MPI environment, then can select the MPI parallel environment to realize, if have available OMP environment, then can select the OMP parallel environment to realize, possess simultaneously then and can use simultaneously, then carry out image when not possessing and cut apart by normal algorithm pattern (unit is single-threaded).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101575306A CN101706950B (en) | 2009-07-13 | 2009-07-13 | High-performance implementation method for multi-scale segmentation of remote sensing images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101575306A CN101706950B (en) | 2009-07-13 | 2009-07-13 | High-performance implementation method for multi-scale segmentation of remote sensing images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101706950A true CN101706950A (en) | 2010-05-12 |
| CN101706950B CN101706950B (en) | 2012-04-25 |
Family
ID=42377172
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009101575306A Expired - Fee Related CN101706950B (en) | 2009-07-13 | 2009-07-13 | High-performance implementation method for multi-scale segmentation of remote sensing images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101706950B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101916432A (en) * | 2010-07-29 | 2010-12-15 | 武汉大学 | Parallelization method for image segmentation algorithm based on mean shift |
| CN102024258A (en) * | 2010-12-15 | 2011-04-20 | 中国科学院遥感应用研究所 | Multi-scale segmentation method for remote sensing image with boundary maintenance characteristics |
| CN102063707A (en) * | 2011-01-05 | 2011-05-18 | 西安电子科技大学 | Mean shift based grey relation infrared imaging target segmentation method |
| CN102385694A (en) * | 2010-09-06 | 2012-03-21 | 邬明权 | Hyperspectral identification method for land parcel-based crop variety |
| CN102496151A (en) * | 2011-12-08 | 2012-06-13 | 南京大学 | Method for multi-scale segmentation of high-resolution remote sensing images |
| WO2012118459A1 (en) | 2011-03-01 | 2012-09-07 | Ulusoy Ilkay | An object based segmentation method |
| CN102799898A (en) * | 2012-06-28 | 2012-11-28 | 浙江大学 | Efficient target identification processing method of high-resolution remote sensing image in context |
| CN102831613A (en) * | 2012-08-29 | 2012-12-19 | 武汉大学 | Parallel fractural network evolution image segmentation method |
| CN103617626A (en) * | 2013-12-16 | 2014-03-05 | 武汉狮图空间信息技术有限公司 | Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method |
| CN104392436A (en) * | 2014-11-11 | 2015-03-04 | 莱芜钢铁集团有限公司 | Processing method and device for remote sensing image |
| CN104537125A (en) * | 2015-01-28 | 2015-04-22 | 中国人民解放军国防科学技术大学 | Remote-sensing image pyramid parallel building method based on message passing interface |
| CN105574887A (en) * | 2016-02-29 | 2016-05-11 | 民政部国家减灾中心 | Quick high-resolution remote sensing image segmentation method |
| 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 |
| CN105787911A (en) * | 2016-03-21 | 2016-07-20 | 中国林业科学研究院资源信息研究所 | Image erosion and expansion processing method based on topology fractal algorithm |
| CN107341837A (en) * | 2017-06-26 | 2017-11-10 | 华中师范大学 | Grid and vector data conversion and continuous yardstick expression based on image pyramid |
| CN107424154A (en) * | 2017-04-21 | 2017-12-01 | 成都信息工程大学 | Watershed image parallel partition method based on dynamically distributes |
| CN108241148A (en) * | 2017-12-13 | 2018-07-03 | 交通运输部规划研究院 | A kind of method using remote sensing images attributes extraction port resources information |
| CN108986113A (en) * | 2018-07-06 | 2018-12-11 | 航天星图科技（北京）有限公司 | A kind of block parallel multi-scale division algorithm based on LLTS frame |
| CN109993753A (en) * | 2019-03-15 | 2019-07-09 | 北京大学 | The dividing method and device of urban function region in remote sensing image |
| CN110008854A (en) * | 2019-03-18 | 2019-07-12 | 中交第二公路勘察设计研究院有限公司 | Unmanned plane image Highway Geological Disaster recognition methods based on pre-training DCNN |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100538399C (en) * | 2007-09-27 | 2009-09-09 | 武汉大学 | Cutting apart and classification and variety detection integration method of high-resolution remote sensing image |
| CN101364259B (en) * | 2008-04-09 | 2010-06-02 | 武汉大学 | Method for extracting road various information of multi-level knowledge driven panchromatic remote sensing image |
| CN101408941B (en) * | 2008-10-20 | 2010-10-27 | 中国科学院遥感应用研究所 | Method for multi-dimension segmentation of remote sensing image and representation of segmentation result hierarchical structure |
-
2009
- 2009-07-13 CN CN2009101575306A patent/CN101706950B/en not_active Expired - Fee Related
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101916432A (en) * | 2010-07-29 | 2010-12-15 | 武汉大学 | Parallelization method for image segmentation algorithm based on mean shift |
| CN101916432B (en) * | 2010-07-29 | 2012-05-02 | 武汉大学 | Parallelization method for image segmentation algorithm based on mean shift |
| CN102385694A (en) * | 2010-09-06 | 2012-03-21 | 邬明权 | Hyperspectral identification method for land parcel-based crop variety |
| CN102024258A (en) * | 2010-12-15 | 2011-04-20 | 中国科学院遥感应用研究所 | Multi-scale segmentation method for remote sensing image with boundary maintenance characteristics |
| CN102024258B (en) * | 2010-12-15 | 2012-07-25 | 中国科学院遥感应用研究所 | Multi-scale segmentation method for remote sensing image with boundary maintenance characteristics |
| CN102063707A (en) * | 2011-01-05 | 2011-05-18 | 西安电子科技大学 | Mean shift based grey relation infrared imaging target segmentation method |
| WO2012118459A1 (en) | 2011-03-01 | 2012-09-07 | Ulusoy Ilkay | An object based segmentation method |
| CN102496151A (en) * | 2011-12-08 | 2012-06-13 | 南京大学 | Method for multi-scale segmentation of high-resolution remote sensing images |
| CN102799898A (en) * | 2012-06-28 | 2012-11-28 | 浙江大学 | Efficient target identification processing method of high-resolution remote sensing image in context |
| CN102831613A (en) * | 2012-08-29 | 2012-12-19 | 武汉大学 | Parallel fractural network evolution image segmentation method |
| CN102831613B (en) * | 2012-08-29 | 2014-11-19 | 武汉大学 | Parallel fractural network evolution image segmentation method |
| CN103617626A (en) * | 2013-12-16 | 2014-03-05 | 武汉狮图空间信息技术有限公司 | Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method |
| CN104392436A (en) * | 2014-11-11 | 2015-03-04 | 莱芜钢铁集团有限公司 | Processing method and device for remote sensing image |
| CN104392436B (en) * | 2014-11-11 | 2019-01-04 | 莱芜钢铁集团有限公司 | A kind of remote sensing image processing method and device |
| CN104537125A (en) * | 2015-01-28 | 2015-04-22 | 中国人民解放军国防科学技术大学 | Remote-sensing image pyramid parallel building method based on message passing interface |
| CN104537125B (en) * | 2015-01-28 | 2017-11-14 | 中国人民解放军国防科学技术大学 | A kind of remote sensing image pyramid parallel constructing method based on message passing interface |
| CN105608474B (en) * | 2015-12-30 | 2019-07-09 | 二十一世纪空间技术应用股份有限公司 | Region adaptivity plant extraction method based on high resolution image |
| 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 |
| CN105574887B (en) * | 2016-02-29 | 2019-08-02 | 民政部国家减灾中心 | High-resolution remote sensing image fast partition method |
| CN105574887A (en) * | 2016-02-29 | 2016-05-11 | 民政部国家减灾中心 | Quick high-resolution remote sensing image segmentation method |
| CN105787911A (en) * | 2016-03-21 | 2016-07-20 | 中国林业科学研究院资源信息研究所 | Image erosion and expansion processing method based on topology fractal algorithm |
| CN105787911B (en) * | 2016-03-21 | 2019-02-12 | 中国林业科学研究院资源信息研究所 | A kind of Image erosion and expanding processing based on topological fractal algorithm |
| CN107424154A (en) * | 2017-04-21 | 2017-12-01 | 成都信息工程大学 | Watershed image parallel partition method based on dynamically distributes |
| CN107424154B (en) * | 2017-04-21 | 2020-04-03 | 成都信息工程大学 | Watershed image segmentation parallel method based on dynamic distribution |
| CN107341837A (en) * | 2017-06-26 | 2017-11-10 | 华中师范大学 | Grid and vector data conversion and continuous yardstick expression based on image pyramid |
| CN107341837B (en) * | 2017-06-26 | 2020-07-10 | 华中师范大学 | Grid-vector data conversion and continuous scale expression method based on image pyramid |
| CN108241148A (en) * | 2017-12-13 | 2018-07-03 | 交通运输部规划研究院 | A kind of method using remote sensing images attributes extraction port resources information |
| CN108986113A (en) * | 2018-07-06 | 2018-12-11 | 航天星图科技（北京）有限公司 | A kind of block parallel multi-scale division algorithm based on LLTS frame |
| CN109993753A (en) * | 2019-03-15 | 2019-07-09 | 北京大学 | The dividing method and device of urban function region in remote sensing image |
| CN109993753B (en) * | 2019-03-15 | 2021-03-23 | 北京大学 | Method and device for segmenting urban functional area in remote sensing image |
| CN110008854A (en) * | 2019-03-18 | 2019-07-12 | 中交第二公路勘察设计研究院有限公司 | Unmanned plane image Highway Geological Disaster recognition methods based on pre-training DCNN |
| CN110008854B (en) * | 2019-03-18 | 2021-04-30 | 中交第二公路勘察设计研究院有限公司 | Unmanned aerial vehicle image highway geological disaster identification method based on pre-training DCNN |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101706950B (en) | 2012-04-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101706950B (en) | High-performance implementation method for multi-scale segmentation of remote sensing images | |
| Xiao et al. | A weakly supervised semantic segmentation network by aggregating seed cues: the multi-object proposal generation perspective | |
| CN106778821B (en) | Classification of Polarimetric SAR Image method based on SLIC and improved CNN | |
| Hsu et al. | Unsupervised cnn-based co-saliency detection with graphical optimization | |
| Ashraf et al. | A multichannel Markov random field framework for tumor segmentation with an application to classification of gene expression-based breast cancer recurrence risk | |
| CN102096816B (en) | Multi-scale multi-level image segmentation method based on minimum spanning tree | |
| CN105869173A (en) | Stereoscopic vision saliency detection method | |
| CN110188763B (en) | Image significance detection method based on improved graph model | |
| Wang et al. | Adaptive nonlocal random walks for image superpixel segmentation | |
| CN112712546A (en) | Target tracking method based on twin neural network | |
| Zohourian et al. | Superpixel-based Road Segmentation for Real-time Systems using CNN. | |
| CN105046689A (en) | Method for fast segmenting interactive stereo image based on multilayer graph structure | |
| Chong et al. | Erase then grow: Generating correct class activation maps for weakly-supervised semantic segmentation | |
| Yang et al. | Directional connectivity-based segmentation of medical images | |
| Wang et al. | A region-line primitive association framework for object-based remote sensing image analysis | |
| Qian et al. | Spatial contextual noise removal for post classification smoothing of remotely sensed images | |
| Liao et al. | Deep-learning-based object-level contour detection with CCG and CRF optimization | |
| Wang et al. | A novel coarse-to-fine hair segmentation method | |
| Kalinin et al. | A graph based approach to hierarchical image over-segmentation | |
| Zhao et al. | Multiple vision architectures-based hybrid network for hyperspectral image classification | |
| Li et al. | 2D amodal instance segmentation guided by 3D shape prior | |
| CN107392926B (en) | Remote sensing image feature selection method based on early-stage land thematic map | |
| Chen et al. | Collaborative classification of hyperspectral and lidar data with information fusion and deep nets | |
| Wu et al. | Instant water body variation detection via analysis on remote sensing imagery | |
| Zheng et al. | Category-wise fusion and enhancement learning for multimodal remote sensing image semantic segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120425 Termination date: 20150713 |
|
| EXPY | Termination of patent right or utility model |