CN103390271A - Remote-sensing image segmentation method and device - Google Patents

Abstract

The invention provides a remote-sensing image segmentation method and device. The method comprises representing an original remote-sensing image in a way of a region adjacency graph, and data-partitioning the region adjacency graph according to a preset data partitioning mode to obtain a plurality of sub region adjacency graphs; counting the weights and the number of the arc segments of every sub region adjacency graph in parallel, calculating current segmentation state scale parameters of the region adjacency graph and determining whether the current segmentation state scale parameters are larger than preset scale parameters; if not, combining two region objects to be combined, which are associated with arc segments to be combined, updating the adjacency topological relationship between region objects, recounting the weights and the number of the arc segments of every sub region adjacency graph; if so, completing the segmentation. According to the remote-sensing image segmentation method and device, by counting every sub region adjacency graph in parallel and combining the region objects, the rapid segmentation of the original remote-sensing image can be achieved.

Description

Remote Sensing Image Segmentation and device

Technical field

The present invention relates to image processing techniques, relate in particular to a kind of Remote Sensing Image Segmentation and device.

Background technology

Development along with Aero-Space remote sensor data acquisition technology, people can obtain ultra-large high-definition remote sensing image data in a short period of time, as U.S. QuickBird satellite, WorldView-II satellite, GEOEye-I satellite and CBERS-2B satellite, per minute can gather respectively the image of about 373,2708,1145 and 120 mega pixels (megapixels).And increasing on-line system requires to process in real time remote sensing image data, as military target identification and terrain match, meteorological weather forecast, emergent disaster etc.

In prior art, when being carried out multi-scale division, mainly adopts high-resolution remote sensing image serial algorithm.Yet in the image auto Segmentation was become the process of some significant section objects, that must consider object cut apart the problems such as yardstick, syntople.Therefore, realize need to consuming super amount computational resource to the high-quality Accurate Segmentation of a large-scale high-resolution remote sensing image, this computation-intensive and data-intensive computational problem, as adopting serial algorithm processing speed on the personal desktop machine of current main-stream slow, efficiency is low.

Summary of the invention

When in conventional art, remote sensing image being carried out multi-scale division, adopt slow, the inefficient defect of serial algorithm speed, the embodiment of the present invention provides a kind of Remote Sensing Image Segmentation and device.

The embodiment of the present invention provides a kind of Remote Sensing Image Segmentation, comprising:

The mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, described Region adjacency graph being carried out data divides, obtain a plurality of subregion adjacent maps, described subregion adjacent map comprises section object and connects the segmental arc of two described section objects;

Segmental arc weights and the segmental arc number of segmental arc in parallel each described subregion adjacent map of statistics, the current cutting state scale parameter of the described Region adjacency graph of calculating, judge that whether described current cutting state scale parameter is greater than default scale parameter;

If described current cutting state scale parameter is not more than default scale parameter, merge two associated section objects to be combined of segmental arc to be combined, and upgrade in each described subregion adjacent map between section object in abutting connection with topological relation, return to segmental arc weights and segmental arc number in each described subregion adjacent map of statistics;

If described current cutting state scale parameter, greater than described default scale parameter, has been cut apart.

Preferably, the step of associated two section objects to be combined of described merging segmental arc to be combined specifically comprises:

, according to default merge algorithm, merge two associated section objects to be combined of segmental arc to be combined.

Preferably, described preset data dividing mode is the chessboard division mode; Described segmental arc weights are the heterogeneity value of two section objects connecting of described segmental arc.

Preferably, the current cutting state scale parameter of the described Region adjacency graph of described calculating comprises:

Described segmental arc weights in subregion adjacent map according to each, the minimum heterogeneous value of the described Region adjacency graph of statistics, wherein said minimum heterogeneous value equals the minimum segmental arc weights of all segmental arcs in described Region adjacency graph;

Described segmental arc weights and segmental arc number in subregion adjacent map according to each, calculate described average heterogeneous value, and wherein said average heterogeneous value equals the segmental arc weights sum of all segmental arcs in described Region adjacency graph divided by the segmental arc sum;

Calculate described current cutting state scale parameter according to described average heterogeneous value, wherein said current cutting state scale parameter equals described average heterogeneous value square root.

Preferably, described default merge algorithm is heterogeneous Minimum Area object merging algorithm;

The default merge algorithm of described basis merges two associated section objects to be combined of segmental arc to be combined, and upgrade in each described subregion adjacent map between section object in abutting connection with topological relation, comprising:

According to heterogeneous Minimum Area object merging algorithm, parallel each described subregion adjacent map of traversal, obtain segmental arc to be combined, and the segmental arc weights of described segmental arc to be combined equal described minimum heterogeneous value;

Judge whether two associated section objects to be combined of described segmental arc to be combined are the interior zone object, described interior zone object refers to that all section objects adjacent with described interior zone object are all in same described subregion adjacent map;

If two described section objects to be combined are described interior zone object, parallel two associated section objects to be combined of described segmental arc to be combined that merge, upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with topological relation;

If have the borderline region object in two described section objects to be combined, described segmental arc to be combined is stored in interim array, travel through described interim array and obtain described segmental arc to be combined, described borderline region object refers to that the section object adjacent with described borderline region object be not entirely in same subregion adjacent map;

Two described section objects to be combined that the described segmental arc to be combined of erial merge is associated, upgrade in Region adjacency graph between section object in abutting connection with between the topological relation section object in abutting connection with topological relation.

Preferably, if described two described section objects to be combined are described interior zone object, walk abreast and merge two associated section objects to be combined of described segmental arc to be combined, upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with between the topological relation section object in abutting connection with topological relation, comprising:

Parallel two the associated section objects to be combined of described segmental arc to be combined that merge;

The subregion adjacent map at parallel traversal described section object to be combined place, search other associated segmental arc of described section object to be combined;

Upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with between the topological relation section object in abutting connection with topological relation.

Preferably, two described section objects to be combined that the described segmental arc to be combined of described erial merge is associated, upgrade in Region adjacency graph between section object in abutting connection with topological relation, comprising:

Two described section objects to be combined that the described segmental arc to be combined of erial merge is associated;

The described Region adjacency graph of serial traversal, search other associated segmental arc of described section object to be combined;

Upgrade in described Region adjacency graph between section object in abutting connection with topological relation.

Preferably, the described Region adjacency graph of described serial traversal, search other associated segmental arc of described section object to be combined, comprising:

Obtain the primary importance information of described section object to be combined;

, according to the primary importance information of described section object to be combined, obtain the second place information of the subregion adjacent map at the section object place adjacent with described section object to be combined;

, to described subregion adjacent map, travel through described subregion adjacent map according to described second place Information locating, search other associated segmental arc of described section object to be combined.

In the technical scheme shown in the present embodiment, at first the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, Region adjacency graph is divided into a plurality of subregion adjacent maps.Segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map of parallel statistics, the minimum heterogeneous value of zoning adjacent map, average heterogeneous value and current cutting state scale parameter, when current cutting state scale parameter is not more than default scale parameter, according to heterogeneous Minimum Area object merging algorithm, heterogeneous minimum section object is merged, upgrade in Region adjacency graph between section object in abutting connection with topological relation, return to the iteration statistics and merge., by statistics that every sub regions adjacent map is walked abreast, can calculate fast minimum heterogeneous value and the current cutting state scale parameter of Region adjacency graph, thereby realize the Fast Segmentation of original remote sensing image is processed.

The embodiment of the present invention provides a kind of Remote Sensing Image Segmentation device, comprises that the first processing module, the second processing module, the 3rd processing module and manage module everywhere, wherein:

Described the first processing module, be used for the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, described Region adjacency graph being carried out data divides, obtain a plurality of subregion adjacent maps, described subregion adjacent map comprises section object and connects the segmental arc of two described section objects;

Described the second processing module, the segmental arc weights and the segmental arc number that are used for parallel each described subregion adjacent map segmental arc of statistics, calculate the current cutting state scale parameter of described Region adjacency graph, judge that whether described current cutting state scale parameter is greater than default scale parameter;

Described the 3rd processing module, be not more than default scale parameter if be used for described current cutting state scale parameter, merge two associated section objects to be combined of described segmental arc to be combined, upgrade in each described subregion adjacent map between section object in abutting connection with topological relation, notify the second processing module again to add up segmental arc weights and segmental arc number in each described subregion adjacent map;

Described the manages module everywhere,, if be used for described current cutting state scale parameter greater than described default scale parameter, finishes to cut apart flow process.

Preferably, described the second processing module also is used for segmental arc weights and the segmental arc number of parallel each described subregion adjacent map segmental arc of statistics, according to default merge algorithm, calculate the current cutting state scale parameter of described Region adjacency graph, judge that whether described current cutting state scale parameter is greater than default scale parameter.

Preferably, described preset data dividing mode is the chessboard division mode; Described segmental arc weights are the heterogeneity value of two section objects connecting of described segmental arc.

Preferably, described the second processing module comprises:

Parallel statistics submodule, be used for parallel segmental arc weights and segmental arc number of adding up each described subregion adjacent map segmental arc;

Minimum heterogeneous value calculating sub module, the described segmental arc weights that are used for subregion adjacent map according to each, add up the minimum heterogeneous value of described Region adjacency graph, wherein said minimum heterogeneous value equals the minimum segmental arc weights of all segmental arcs in described Region adjacency graph;

Average heterogeneous value calculating sub module, the described segmental arc weights and the segmental arc number that are used for subregion adjacent map according to each, calculate described average heterogeneous value, wherein said average heterogeneous value equals the segmental arc weights sum of all segmental arcs in described Region adjacency graph divided by the segmental arc sum;

Current cutting state scale parameter calculating sub module, be used for calculating described current cutting state scale parameter according to described average heterogeneous value, and wherein said current cutting state scale parameter equals described average heterogeneous value square root;

Current cutting state scale parameter judgement submodule, be used for judging that whether described current cutting state scale parameter is greater than default scale parameter.

Preferably, described default merge algorithm is heterogeneous Minimum Area object merging algorithm;

Described the 3rd processing module comprises:

Traversal is searched submodule, is used for according to heterogeneous Minimum Area object merging algorithm, and parallel each described subregion adjacent map of traversal, obtain segmental arc to be combined, and the segmental arc weights of described segmental arc to be combined equal described minimum heterogeneous value;

Interior zone object judgement submodule, be used for judging whether two associated section objects to be combined of described segmental arc to be combined are the interior zone object, described interior zone object refers to that all section objects adjacent with described interior zone object are all in same described subregion adjacent map;

The parallel submodule that merges, be described interior zone object if be used for two described section objects to be combined, parallel two associated section objects to be combined of described segmental arc to be combined that merge, upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with topological relation;

The erial merge submodule, if have the borderline region object in two described section objects to be combined, described segmental arc to be combined is stored in interim array, travel through described interim array and obtain described segmental arc to be combined, described borderline region object refers to that the section object adjacent with described borderline region object be not entirely in same subregion adjacent map; Two described section objects to be combined that the described segmental arc to be combined of erial merge is associated, upgrade in Region adjacency graph between section object in abutting connection with topological relation.

Preferably, described parallel merging submodule comprises:

Parallel merge cells, be used for parallel two the associated section objects to be combined of described segmental arc to be combined that merge;

Parallel traversal unit, be used for the parallel subregion adjacent map that travels through described section object to be combined place, searches other associated segmental arc of described section object to be combined;

The first updating block, be used for upgrading between the subregion adjacent map section object at described section object to be combined place in abutting connection with topological relation.

Preferably, described erial merge submodule comprises:

The erial merge unit, be used for two associated described section objects to be combined of the described segmental arc to be combined of erial merge;

Serial traversal unit, be used for the described Region adjacency graph of serial traversal, searches other associated segmental arc of described section object to be combined;

The second updating block, be used for upgrading between described Region adjacency graph section object in abutting connection with topological relation.

Preferably, described serial traversal unit comprises:

Primary importance acquisition of information subelement, for the primary importance information of obtaining described section object to be combined;

Second place acquisition of information subelement, be used for the primary importance information according to described section object to be combined, obtains the second place information of the subregion adjacent map at the section object place adjacent with described section object to be combined;

Serial traversal subelement, be used for arriving described subregion adjacent map according to described second place Information locating, travels through described subregion adjacent map, searches other associated segmental arc of described section object to be combined.

In the technical scheme shown in the present embodiment, at first the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, Region adjacency graph is divided into a plurality of subregion adjacent maps.Segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map of parallel statistics, the minimum heterogeneous value of zoning adjacent map, average heterogeneous value and current cutting state scale parameter, when current cutting state scale parameter is not more than default scale parameter, according to heterogeneous Minimum Area object merging algorithm, heterogeneous minimum section object is merged, upgrade in Region adjacency graph between section object in abutting connection with topological relation, return to the iteration statistics and merge., by statistics that every sub regions adjacent map is walked abreast, can calculate fast minimum heterogeneous value and the current cutting state scale parameter of Region adjacency graph, thereby realize the Fast Segmentation of original remote sensing image is processed.

Description of drawings

Fig. 1 is Remote Sensing Image Segmentation the first embodiment process flow diagram of the present invention;

Fig. 2 is that the data of Region adjacency graph are divided schematic diagram;

Fig. 3 is Remote Sensing Image Segmentation the second embodiment process flow diagram of the present invention;

Fig. 4 is Remote Sensing Image Segmentation device the first example structure schematic diagram of the present invention;

Fig. 5 is Remote Sensing Image Segmentation device the second example structure schematic diagram of the present invention.

Embodiment

After the original remote sensing image that satellite is taken transfers to ground-based computer, computing machine each ground object target in can't the Direct Recognition original remote sensing image, such as mountains and rivers, river, house etc.But, due to the image different in kind of different ground object targets, therefore can analyze original remote sensing image by image processing techniques, thereby propose interested ground object target.At present because the data volume of original remote sensing image is very large, computing machine is in processing procedure, at first the present invention needs original remote sensing image is expressed by the mode of Region adjacency graph, then Region adjacency graph being carried out data divides, Region adjacency graph is divided into a plurality of subregion adjacent maps, then respectively the every sub regions adjacent map after dividing is processed.The cutting procedure of original remote sensing image is the merging process of section object adjacent in Region adjacency graph, in order to guarantee that original remote sensing image is split into different images high homogeneity, interconnected zone, the present invention is based on heterogeneous Minimum Area object merging algorithm, the adjacent area object is merged.

Fig. 1 is Remote Sensing Image Segmentation the first embodiment process flow diagram of the present invention, and as shown in Figure 1, the Remote Sensing Image Segmentation that the embodiment of the present invention provides is carried out by the remote sensing image device, and the remote sensing image device can adopt the form of software and/or hardware to realize.The method comprises:

Step S100, represent the mode of original remote sensing image with Region adjacency graph, according to the preset data dividing mode, described Region adjacency graph carried out data and divide, and obtains a plurality of subregion adjacent maps;

Particularly, interested ground object target in original remote sensing image is represented with the mode of the segmental arc that is connected two adjacent area objects with section object, thereby original remote sensing image is converted into Region adjacency graph.Then according to the preset data dividing mode, Region adjacency graph is carried out data divide, obtain a plurality of subregion adjacent maps, wherein the subregion adjacent map comprises section object and connects the segmental arc of two described section objects.

Preferably, the preset data dividing mode is the chessboard division mode, Fig. 2 is that the data of Region adjacency graph are divided schematic diagram, as shown in Figure 2, Region adjacency graph is divided into the identical square of a plurality of sizes, each square is the subregion adjacent map, and every sub regions adjacent map comprises section object and the segmental arc that is connected two adjacent area objects.Be 1. for example section object, 1. 14 be related section object and segmental arc 4..Wherein, be 9. the interior zone object, namely with 9. be associated or adjacent section object all in same subregion adjacent map; 7. be the perimeter object, namely with 7. be associated or adjacent section object not entirely in same subregion adjacent map.

The dividing mode of Region adjacency graph need to meet the following conditions:

Make RAG represent whole Region adjacency graph, RAG is carried out data division expression RAG is divided into the n sub regions adjacent map (RAG that meets following 4 conditions ₁, RAG ₂..., RAG _n) set:

（1）RAG=RAG ₁∪RAG ₂∪……∪RAG _n；

(2) to all i and j, i ≠ j, have RAG _i∩ RAG _j=Φ;

(3) to i=1,2 ..., n, have P (RAG _i)=TRUE;

(4) to i ≠ j, P (RAG is arranged _i∪ RAG _j)=FALSE;

P (RAG wherein _i) be at set RAG to all _iThe logical predicate of middle element, Φ represents empty set.

Wherein, in mathematical logic: distributive and predicate are resolved in atomic proposition.Distributive is can self-existent object, and it can be concrete things or abstract concept.Predicate is to delineate the character of distributive or the word of things Relations Among.

Condition (1) expression equals former Region adjacency graph to the summation of whole subregion adjacent maps, i.e. data division must be completely, and in former Region adjacency graph, each section object must belong to some subregion adjacent maps.Condition (2) expression all subregion adjacent map is non-overlapping copies, and a section object can not belong to two sub regions adjacent maps simultaneously in other words.Condition (3) is illustrated in after data are divided the section object that belongs in same subregion adjacent map that obtains and has some identical characteristics.Condition (4) is illustrated in after data are divided the section object that belongs in different subregion adjacent maps that obtains and has some different characteristics.

Step S102, segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map of parallel statistics, the current cutting state scale parameter of zoning adjacent map;

Step S104 judges that whether described current cutting state scale parameter is greater than default scale parameter; , if described current cutting state scale parameter is not more than default scale parameter, enter step S106; If described current cutting state scale parameter, greater than described default scale parameter, enters step S108;

Add up segmental arc weights and the segmental arc number of segmental arc in every sub regions adjacent map, wherein, the segmental arc weights are the heterogeneity value of two section objects connecting of segmental arc.The internal data that only relates to this subregion adjacent map due to the statistic processes of every sub regions adjacent map, and between other subregion adjacent maps, data communication does not occur, segmental arc number and segmental arc weights in a plurality of subregion adjacent maps of the statistics that therefore can walk abreast.

Preferably, according to the segmental arc weights in every sub regions adjacent map, the minimum heterogeneous value of statistical regions adjacent map, wherein minimum heterogeneous value equals the minimum segmental arc weights of all segmental arcs in Region adjacency graph; According to the segmental arc weights in every sub regions adjacent map and segmental arc number, calculate average heterogeneous value, wherein average heterogeneous value equals the segmental arc weights sum of all segmental arcs in Region adjacency graph divided by the segmental arc sum; Calculate current cutting state scale parameter according to average heterogeneous value, wherein current cutting state scale parameter equals described average heterogeneous value square root.

Particularly, adopt parallel mode, the independent minimum heterogeneous value of son and the average heterogeneous value of son of calculating all subregion adjacent map.Namely adopt parallel mode, the segmental arc weights of segmental arc in independent statistics all subregion adjacent map, be designated as the results set that obtains: MHC '={ MHC _i: 1≤i≤n}; The independent average heterogeneous value of son of calculating all subregion adjacent map, the results set that obtains is designated as: AHC '={ AHC _j: 1≤j≤n}.According to set MHC ' and AHC ', the minimum heterogeneous value of this Region adjacency graph of serial computing, average heterogeneous value and current cutting state scale parameter.Current cutting state scale parameter computing formula is as follows:

Wherein K represents the segmental arc sum in this Region adjacency graph, e _kThe segmental arc weights that represent k segmental arc, the S root of making even.

Judge that whether current cutting state scale parameter is greater than default scale parameter; If current cutting state scale parameter, greater than default scale parameter, enters step S106, flow process finishes to have cut apart; , if current cutting state scale parameter is not more than default scale parameter, enter step S104.

Step S106, merge two associated section objects to be combined of segmental arc to be combined, upgrade in every sub regions adjacent map between section object in abutting connection with topological relation, return to step S102 and again add up segmental arc weights and segmental arc number in every sub regions adjacent map;

Step S108, flow process finishes, and has cut apart.

Preferably, merge segmental arc to be combined according to default merge algorithm, wherein default merge algorithm is heterogeneous Minimum Area object merging algorithm, this algorithm, for selecting two minimum section objects of heterogeneous value in Region adjacency graph to merge as section object to be combined, namely selects two associated section objects of the corresponding segmental arc of segmental arc weights minimum value to merge as section object to be combined.

Particularly,, according to heterogeneous Minimum Area object merging algorithm, obtain segmental arc to be combined, wherein segmental arc to be combined refers to that the segmental arc weights equal the segmental arc of minimum heterogeneous value.Merge two associated section objects to be combined of segmental arc to be combined, namely delete segmental arc to be combined and one of them section object to be combined, only keep a section object to be combined.Due to a section object to be combined related several segmental arcs simultaneously, after one of them section object to be combined is deleted, deleted section object to be combined except association segmental arc to be combined, may be also related other segmental arcs.When therefore deleting this section object to be combined, also need in abutting connection with topological relation, the upgrading of other associated segmental arc of this section object to be combined, with other segmental arcs originally associated deleted section object to be combined be updated to section object after merging., because therefore other segmental arcs may need the whole Region adjacency graph of traversal with segmental arc to be combined not in same subregion adjacent map, upgrade the segmental arc in every sub regions adjacent map.

After renewal was completed, variation had occurred in section object and segmental arc in former Region adjacency graph, therefore need to return to step S102, segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map after statistics is upgraded again.Through the statistics of iteration repeatedly with merge, during greater than default scale parameter, cutting procedure finishes when the current cutting state scale parameter of Region adjacency graph.

Further, when the section object in whole Region adjacency graph only remains the next one, cut apart flow process and also must finish.Therefore also comprise after step S102:

Add up the number of section object described in described Region adjacency graph;

Judge that whether the number of section object described in described Region adjacency graph is greater than one;

If enter step S104; Enter if not step S108, flow process finishes, and has cut apart.

In the technical scheme shown in the present embodiment, at first the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, Region adjacency graph is divided into a plurality of subregion adjacent maps.Segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map of parallel statistics, the minimum heterogeneous value of zoning adjacent map, average heterogeneous value and current cutting state scale parameter, when current cutting state scale parameter is not more than default scale parameter, according to heterogeneous Minimum Area object merging algorithm, heterogeneous minimum section object is merged, upgrade in Region adjacency graph between section object in abutting connection with topological relation, return to the iteration statistics and merge., by statistics that every sub regions adjacent map is walked abreast, can calculate fast minimum heterogeneous value and the current cutting state scale parameter of Region adjacency graph, thereby realize the Fast Segmentation of original remote sensing image is processed.

Further, before merging two associated section objects to be combined of segmental arc to be combined, judge whether these two section objects to be combined are arranged in same subregion adjacent map; If not, add up these two section objects to be combined and distinguish the number of the section object in the subregion adjacent map at places; To be arranged in the section object deletion to be combined of the more subregion adjacent map of section object number.

When treating the assembly section field object and merge, when section object to be combined lays respectively in two different subregion adjacent maps, prejudge the number of section object in the subregion adjacent map at two section object to be combined places, to be arranged in the section object deletion to be combined of a fairly large number of subregion adjacent map of section object,, in order to avoid the too much situation of section object quantity in some subregion adjacent maps occurs, carry out load balancing.

Fig. 3 is Remote Sensing Image Segmentation the second embodiment process flow diagram of the present invention, and as shown in Figure 3, the method comprises:

Step S200, represent the mode of original remote sensing image with Region adjacency graph, according to the preset data dividing mode, described Region adjacency graph carried out data and divide, and obtains a plurality of subregion adjacent maps;

Step S202, segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map of parallel statistics;

Step S204, according to described segmental arc weights and the segmental arc number in every sub regions adjacent map, the minimum heterogeneous value of zoning adjacent map, average heterogeneous value and current cutting state scale parameter;

Step S206, judge that whether described current cutting state scale parameter is greater than default scale parameter, if described current cutting state scale parameter is not more than default scale parameter, enter step S208, if described current cutting state scale parameter, greater than described default scale parameter, enters step S226;

Step S208, according to heterogeneous Minimum Area object merging algorithm, the every sub regions adjacent map of parallel traversal, obtain segmental arc to be combined;

Wherein, the segmental arc weights of segmental arc to be combined equal described minimum heterogeneous value;

Step S210, judge whether two associated section objects to be combined of segmental arc to be combined are the interior zone object; , if two section objects to be combined are described interior zone object, enter step S212; Otherwise enter step S218;

Wherein, the interior zone object refers to that all section objects adjacent with described interior zone object are all in same described subregion adjacent map.

Step S212, parallel two the associated section objects to be combined of segmental arc to be combined that merge;

Step S214, the subregion adjacent map at parallel traversal section object to be combined place, search other associated segmental arc of section object to be combined;

Step S216, upgrade in the subregion adjacent map at section object to be combined place between section object in abutting connection with topological relation, return to step S202;

Step S218, be stored to segmental arc to be combined in interim array, travels through interim array and obtain segmental arc to be combined;

Step S220, two section objects to be combined that erial merge segmental arc to be combined is associated;

Step S222, serial traversal Region adjacency graph, search other associated segmental arc of section object to be combined;

Step S224, upgrade in Region adjacency graph between section object in abutting connection with topological relation, return to step S202;

Step S226, flow process finishes, and has cut apart.

Particularly, according to the preset data dividing mode, Region adjacency graph is divided into a plurality of subregion adjacent maps, Region adjacency graph is as host node, and every sub regions adjacent map is partial node.

In the every sub regions adjacent map of the parallel statistics of each partial node, data communication,, because statistic processes is carried out in every sub regions adjacent map inside, do not occur in segmental arc weights and the segmental arc number of segmental arc between partial node, and statistics therefore can walk abreast.Host node is according to the statistical computation result of each partial node, and the heterogeneous value of the minimum of the whole Region adjacency graph of serial computing, average heterogeneity are worth and current cutting state scale parameter.

The segmentation object of remote sensing image is under the yardstick corresponding with interested ground object target or spatial structure characteristic of appointment, Image Segmentation is become different images high homogeneity, that interlink zone, corresponding with interested ground object target or spatial structure characteristic.Therefore whether judge current cutting state scale parameter greater than default scale parameter, if be not more than this remote sensing image of explanation, need to further be cut apart.

Heterogeneous Minimum Area object merging algorithm belongs to the region growing strategy towards the overall situation, and the segmental arc in the every sub regions adjacent map of parallel traversal is obtained the segmental arc weights and equaled the segmental arc to be combined of minimum heterogeneous value.Segmental arc to be combined may exist a plurality of, and may not belong to same subregion adjacent map, therefore further judges whether two associated section objects to be combined of segmental arc to be combined all belong to the interior zone object.

When if two associated section objects to be combined of segmental arc to be combined are the interior zone object, associated two section objects to be combined of this segmental arc to be combined and the section object in other subregion adjacent maps do not have annexation.After these two section objects to be combined were merged, only the segmental arc in the subregion adjacent map at this segmental arc to be combined place need to be upgraded.Therefore, when the segmental arc to be combined of related interior zone object while being a plurality of, data communication, do not occur in the parallel section object to be combined that merges in the subregion adjacent map of each partial node, and the parallel annexation of upgrading other segmental arcs in all subregion adjacent map between partial node.

If have the borderline region object in associated two section objects to be combined of segmental arc to be combined, there is annexation in the section object in this section object to be combined and other subregion adjacent maps.Therefore, after two associated section objects to be combined of segmental arc to be combined merged to this, other subregion adjacent maps adjacent with the borderline region object also needed to upgrade.Therefore need to carry out erial merge to the associated section object to be combined of each segmental arc to be combined from host node, namely travel through every sub regions adjacent map, after obtaining a plurality of segmental arcs to be combined, at first this type of segmental arc to be combined is stored in interim array.Then travel through interim array, obtain segmental arc to be combined, two section objects to be combined that each segmental arc to be combined of host node erial merge is associated.The whole Region adjacency graph of last serial traversal, obtain other associated segmental arc of section object to be combined, upgrades the annexation of other segmental arcs in whole Region adjacency graph.

Upgrade complete after, return to segmental arc weights and the segmental arc number of again adding up segmental arc in every sub regions adjacent map, carry out second iteration.

In the technical scheme shown in the present embodiment, at first the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, Region adjacency graph is divided into a plurality of subregion adjacent maps.Segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map of parallel statistics, the minimum heterogeneous value of zoning adjacent map, average heterogeneous value and current cutting state scale parameter, when current cutting state scale parameter is not more than default scale parameter, according to heterogeneous Minimum Area object merging algorithm, heterogeneous minimum section object is merged, upgrade in Region adjacency graph between section object in abutting connection with topological relation, return to the iteration statistics and merge.By statistics that every sub regions adjacent map is walked abreast, can calculate fast minimum heterogeneous value and the current cutting state scale parameter of Region adjacency graph, when the associated section object to be combined of segmental arc to be combined is the interior zone object, walk abreast each segmental arc to be combined is merged, further accelerated the dividing processing to original remote sensing image.

Further, when there was the borderline region object in described section object to be combined, step S222 also comprised:

Obtain the primary importance information of described section object to be combined;

Wherein, primary importance information is unique label of section object to be combined.

, according to the primary importance information of described section object to be combined, obtain the second place information of the subregion adjacent map at the section object place adjacent with described section object to be combined;

Wherein, second place information is unique coding of subregion adjacent map.

, to described subregion adjacent map, travel through described subregion adjacent map according to described second place Information locating, search other associated segmental arc of described section object to be combined.

Particularly, the width of supposing Region adjacency graph is w, is highly h, and the preset data dividing mode is the chessboard division mode, and Region adjacency graph is divided into a plurality of little square that the length of side is s, i.e. subregion adjacent map.The width of Region adjacency graph and highly respectively divided by s, can obtain line number (div_row) and the columns (div_col) of subregion adjacent map.Use again the width of Region adjacency graph and highly to respectively to the s complementation, with the width remainder of the Region adjacency graph that obtains and height remainder, be averagely allocated to each row and each row of chessboard, can obtain final data and divide width (div_w) and the height (div_h) of rear every sub regions adjacent map, further calculate thus initial row coordinate (block_y) and the initial row coordinate (block_x) of every sub regions adjacent map.Circular is as follows:

With the width of subregion adjacent map with highly respectively divided by s, can obtain line number (div_row) and the columns (div_col) of subregion adjacent map, tried to achieve by following formula:

$\left\{\begin{array}{c}\mathrm{div}\_\mathrm{row}\left\{\begin{array}{c}1,\mathrm{if}h<=s\\ h/s,\mathrm{if}h>s\end{array}\right.\\ \mathrm{diw}\_\mathrm{col}\left\{\begin{array}{c}1,\mathrm{if}w<=s\\ w/s,\mathrm{if}w>s\end{array}\right.\end{array}\right.$

Use again the width of Region adjacency graph and highly to respectively to the s complementation, the width remainder of the Region adjacency graph that obtains and height remainder are averagely allocated to the subregion adjacent map of each row and each row, final data divide the width (div_w) of all subregion adjacent map and highly (div_h) be respectively:

$\left\{\begin{array}{c}\mathrm{div}\_h=\left\{\begin{array}{c}h,\mathrm{if}\mathrm{div}\_\mathrm{row}=1\\ s+(h\%s)/\mathrm{div}\_\mathrm{row},\mathrm{if}\mathrm{div}\_\mathrm{row}>1\end{array}\right.\\ \mathrm{div}\_w=\left\{\begin{array}{c}w,\mathrm{if}\mathrm{div}\_\mathrm{col}=1\\ s+(w\%s)/\mathrm{div}\_\mathrm{col},\mathrm{if}\mathrm{div}\_\mathrm{col}>1\end{array}\right.\end{array}\right.$

Thus, m is capable, and row-coordinate (block_y) and the row coordinate (block_x) of the subregion adjacent map of n row are respectively:

$\left\{\begin{array}{c}\mathrm{block}\_y=i\&times;\mathrm{div}\_h(i=1,...,\mathrm{div}\_\mathrm{row})\\ \mathrm{block}\_x=j\&times;\mathrm{div}\_w(j=1,...,\mathrm{div}\_\mathrm{col})\end{array}\right.$

Finally, due to resulting remainder partly exist can not be just by mean allocation to all subregion adjacent map may,, at this, the remaining part of remainder is assigned to respectively in the subregion adjacent map of last column (row) that data divide.The height (block_h) of the subregion adjacent map of last column (row) and width (block_w) are respectively so:

$\left\{\begin{array}{c}\mathrm{block}\_h=\begin{array}{}\end{array}\left\{\begin{array}{c}\mathrm{div}\_h,\mathrm{if}i<\mathrm{div}\_\mathrm{row}\\ h-i\&times;\mathrm{div}\_h,\mathrm{if}i=\mathrm{div}\_\mathrm{row}\end{array}(i=1,...,\mathrm{div}\_\mathrm{row})\right.\\ \mathrm{block}\_w=\left\{\begin{array}{cc}\mathrm{div}\_w,\mathrm{if}j<\mathrm{div}\_\mathrm{col}& \\ w-j\&times;\mathrm{div}\_w,\mathrm{if}j=\mathrm{div}\_\mathrm{col}& \end{array}(j=1,...,\mathrm{div}\_\mathrm{col})\right.\end{array}\right.$

After Region adjacency graph is carried out chessboard division, can obtain div_row * div_col sub regions adjacent map, for ease of follow-up fast finding all subregion adjacent map, all subregion adjacent map be encoded, make unique coding corresponding to every sub regions adjacent map.I is capable, and coding (block_id) formula of j row subregion adjacent map is:

block_id=i×div_col+j(i=1,...,div_row;j=1,...,div_col)

Each section object is also encoded, make label corresponding to each section object, establish the width of Region adjacency graph and highly be respectively w and h, the coordinate of section object in Region adjacency graph is (x, y), unique label computing formula of this section object is as follows:

region_id=y×w+x(y=1,...,h;x=1,...,w)。

If unique label of a known section object, the decoding computing formula of its coordinate is as follows:

$\left\{\begin{array}{c}y=\mathrm{region}\_\mathrm{id}/w+1\\ x=\mathrm{region}\_\mathrm{id}\%w+1\end{array}\right.$

The label of a known section object, coding and decoding process according to the section object label, can the pixel coordinate of this section object in whole Region adjacency graph, row-coordinate is y, the row coordinate is x, and row (i) and the row (j) of the subregion adjacent map at this section object place in data are divided can be tried to achieve by following formula:

Bring the i that obtains and j into subregion adjacent map coding formula and get final product to obtain unique coding of subregion adjacent map, and by unique coding of subregion adjacent map, namely can navigate to fast this subregion adjacent map.

After the large format remote sensing image being carried out the data division, the subregion adjacent map quantity of its division is more, and general subregion adjacent map size is larger, and the section object quantity in the subregion adjacent map is also more, and the general area number of objects is huge especially.Navigate to rapidly the position of the subregion adjacent map that is adjacent by unique coding of section object, thereby need not to travel through and search in whole Region adjacency graph, greatly reduced the scope of searching, accelerate the seek rate of other associated segmental arcs of section object to be combined, improved to a great extent the efficiency of remote sensing image Parallel segmentation.

Fig. 4 is Remote Sensing Image Segmentation device the first example structure schematic diagram of the present invention, as shown in Figure 4, the remote sensing image device that the embodiment of the present invention provides is used for carrying out the Remote Sensing Image Segmentation of above-described embodiment, can adopt the form of software and/or hardware to realize, this device comprises that the first processing module 11, the second processing module 12, the 3rd processing module 13 and manage module 14 everywhere, wherein:

Described the first processing module 11, be used for the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, described Region adjacency graph being carried out data divides, obtain a plurality of subregion adjacent maps, described subregion adjacent map comprises section object and connects the segmental arc of two described section objects;

Described the second processing module 12, the segmental arc weights and the segmental arc number that are used for parallel each described subregion adjacent map segmental arc of statistics, calculate the current cutting state scale parameter of described Region adjacency graph, judge that whether described current cutting state scale parameter is greater than default scale parameter;

Described the 3rd processing module 13, be not more than default scale parameter if be used for described current cutting state scale parameter, merge two associated section objects to be combined of described segmental arc to be combined, upgrade in each described subregion adjacent map between section object in abutting connection with topological relation, notify the second processing module again to add up segmental arc weights and segmental arc number in each described subregion adjacent map;

Described the manages module 14 everywhere,, if be used for described current cutting state scale parameter greater than described default scale parameter, finishes to cut apart flow process.

Particularly, the first 11 pairs of processing modules Region adjacency graph carries out data and divides, the second processing module 12 comprises host node and partial node, partial node is used for adding up segmental arc weights and the segmental arc number of every sub regions adjacent map segmental arc, the internal data that only relates to the subregion adjacent map due to the data statistics for every sub regions adjacent map, need not to carry out data communication between each partial node, so each partial node segmental arc weights and segmental arc number of segmental arc in the every sub regions adjacent map of statistics that can walk abreast.

Further, host node is according to the statistical computation result of each partial node, and the current cutting state scale parameter of the whole Region adjacency graph of serial computing, judge that whether described current cutting state scale parameter is greater than default scale parameter.If current cutting state scale parameter, greater than default scale parameter, notifies the to manage module 14 everywhere and finish to cut apart flow process; , if current cutting state scale parameter is not more than default scale parameter, notifies the 3rd processing module 13 to carry out section object and merge.The 3rd processing module 13 merges two associated section objects to be combined of described segmental arc to be combined according to default merge algorithm.Upgrade the annexation of other segmental arcs in each described subregion adjacent map.Then notify the second processing module 12 again to add up segmental arc weights and segmental arc number in every sub regions adjacent map, carry out next iteration statistics and merge, until the current cutting state scale parameter of Region adjacency graph, greater than default scale parameter, is cut apart the flow process end.

Preferably, described preset data dividing mode is the chessboard division mode; Described segmental arc weights are the heterogeneity value of two section objects connecting of described segmental arc, and wherein default merge algorithm is heterogeneous Minimum Area object merging algorithm.

Further, when the section object in whole Region adjacency graph only remains the next one, cut apart flow process and also must finish.Therefore the second processing module also is used for the number of adding up section object described in described Region adjacency graph; Judge that whether the number of section object described in described Region adjacency graph is greater than one; If notify the 3rd processing module 13; Notify if not the to manage module 14 everywhere and finish to cut apart.

The Remote Sensing Image Segmentation device that various embodiments of the present invention provide is used for carrying out the Remote Sensing Image Segmentation that the embodiment of the present invention provides, and possesses corresponding functional module, repeats no more herein.

In the technical scheme shown in the present embodiment, at first the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, Region adjacency graph is divided into a plurality of subregion adjacent maps.Segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map of parallel statistics, the minimum heterogeneous value of zoning adjacent map, average heterogeneous value and current cutting state scale parameter, when current cutting state scale parameter is not more than default scale parameter, according to heterogeneous Minimum Area object merging algorithm, heterogeneous minimum section object is merged, upgrade in Region adjacency graph between section object in abutting connection with topological relation, return to the iteration statistics and merge., by statistics that every sub regions adjacent map is walked abreast, can calculate fast minimum heterogeneous value and the current cutting state scale parameter of Region adjacency graph, thereby realize the Fast Segmentation of original remote sensing image is processed.

Further, the 3rd processing module 13 also is used for: before merging two associated section objects to be combined of segmental arc to be combined, judge whether these two section objects to be combined are arranged in same subregion adjacent map; Add up if not these two section objects to be combined and distinguish the number of the section object in the subregion adjacent map at places; To be arranged in the section object deletion to be combined of the more subregion adjacent map of section object number.

When treating the assembly section field object and merge, when section object to be combined lays respectively in two different subregion adjacent maps, prejudge the number of section object in the subregion adjacent map at two section object to be combined places, to be arranged in the section object deletion to be combined of a fairly large number of subregion adjacent map of section object,, in order to avoid the too much situation of section object quantity in some subregion adjacent maps occurs, carry out load balancing.

Fig. 5 is Remote Sensing Image Segmentation device the second example structure schematic diagram of the present invention, as shown in Figure 5, this device comprises the first processing module 21, the second processing module 22, the 3rd processing module 23 and is managed module 24 everywhere, the first processing module 21, it is identical with corresponding functional module structure in above-described embodiment that the second processing module 22 and the is managed module 24 everywhere, wherein the second processing module 22 comprises parallel statistics submodule 221, minimum heterogeneous value calculating sub module 222, average heterogeneous value calculating sub module 223, current cutting state scale parameter calculating sub module 224, and current cutting state scale parameter judgement submodule 225:

Parallel statistics submodule 221, be used for parallel segmental arc weights and segmental arc number of adding up each described subregion adjacent map segmental arc;

Minimum heterogeneous value calculating sub module 222, the described segmental arc weights that are used for subregion adjacent map according to each, add up the minimum heterogeneous value of described Region adjacency graph, wherein said minimum heterogeneous value equals the minimum segmental arc weights of all segmental arcs in described Region adjacency graph;

Average heterogeneous value calculating sub module 223, the described segmental arc weights and the segmental arc number that are used for subregion adjacent map according to each, calculate described average heterogeneous value, wherein said average heterogeneous value equals the segmental arc weights sum of all segmental arcs in described Region adjacency graph divided by the segmental arc sum;

Current cutting state scale parameter calculating sub module 224, be used for calculating described current cutting state scale parameter according to described average heterogeneous value, and wherein said current cutting state scale parameter equals described average heterogeneous value square root;

Current cutting state scale parameter judgement submodule 225, be used for judging that whether described current cutting state scale parameter is greater than default scale parameter.

Wherein the 3rd processing module 23 comprises:

Traversal is searched submodule 231, is used for according to heterogeneous Minimum Area object merging algorithm, and parallel each described subregion adjacent map of traversal, obtain segmental arc to be combined, and the segmental arc weights of described segmental arc to be combined equal described minimum heterogeneous value;

Interior zone object judgement submodule 232, be used for judging whether two associated section objects to be combined of described segmental arc to be combined are the interior zone object, described interior zone object refers to that all section objects adjacent with described interior zone object are all in same described subregion adjacent map;

The parallel submodule 233 that merges, be described interior zone object if be used for two described section objects to be combined, parallel two associated section objects to be combined of described segmental arc to be combined that merge, upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with topological relation;

Erial merge submodule 234, if have the borderline region object in two described section objects to be combined, described segmental arc to be combined is stored in interim array, travel through described interim array and obtain described segmental arc to be combined, described borderline region object refers to that the section object adjacent with described borderline region object be not entirely in same subregion adjacent map; Two described section objects to be combined that the described segmental arc to be combined of erial merge is associated, upgrade in Region adjacency graph between section object in abutting connection with topological relation.

Further, the parallel submodule 233 that merges comprises:

Parallel merge cells 2331, be used for parallel two the associated section objects to be combined of described segmental arc to be combined that merge;

Parallel traversal unit 2332, be used for the parallel subregion adjacent map that travels through described section object to be combined place, searches other associated segmental arc of described section object to be combined;

The first updating block 2333, be used for upgrading between the subregion adjacent map section object at described section object to be combined place in abutting connection with topological relation.

Further, erial merge submodule 234 comprises:

Erial merge unit 2341, be used for two associated described section objects to be combined of the described segmental arc to be combined of erial merge;

Serial traversal unit 2342, be used for the described Region adjacency graph of serial traversal, searches other associated segmental arc of described section object to be combined;

The second updating block 2343, be used for upgrading between described Region adjacency graph section object in abutting connection with topological relation.

Further, serial traversal unit 2342 comprises:

Primary importance acquisition of information subelement 23421, for the primary importance information of obtaining described section object to be combined;

Second place acquisition of information subelement 23422, be used for the primary importance information according to described section object to be combined, obtains the second place information of the subregion adjacent map at the section object place adjacent with described section object to be combined;

Serial traversal subelement 23423, be used for arriving described subregion adjacent map according to described second place Information locating, travels through described subregion adjacent map, searches other associated segmental arc of described section object to be combined.

The Remote Sensing Image Segmentation device that various embodiments of the present invention provide is used for carrying out the Remote Sensing Image Segmentation that the embodiment of the present invention provides, and possesses corresponding functional module, repeats no more herein.

In the technical scheme shown in the present embodiment, at first the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, Region adjacency graph is divided into a plurality of subregion adjacent maps.Segmental arc weights and the segmental arc number of segmental arc in the every sub regions adjacent map of parallel statistics, the minimum heterogeneous value of zoning adjacent map, average heterogeneous value and current cutting state scale parameter, when current cutting state scale parameter is not more than default scale parameter, according to heterogeneous Minimum Area object merging algorithm, heterogeneous minimum section object is merged, upgrade in Region adjacency graph between section object in abutting connection with topological relation, return to the iteration statistics and merge.By statistics that every sub regions adjacent map is walked abreast, can calculate fast minimum heterogeneous value and the current cutting state scale parameter of Region adjacency graph, when the associated section object to be combined of segmental arc to be combined is the interior zone object, walk abreast each segmental arc to be combined is merged, further accelerated the dividing processing to original remote sensing image.

It should be noted that: above embodiment is only unrestricted in order to the present invention to be described, the present invention also is not limited in above-mentioned giving an example, and all do not break away from technical scheme and the improvement thereof of the spirit and scope of the present invention, and it all should be encompassed in claim scope of the present invention.

Claims (16)

1. a Remote Sensing Image Segmentation, is characterized in that, comprising:

The mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, described Region adjacency graph being carried out data divides, obtain a plurality of subregion adjacent maps, described subregion adjacent map comprises section object and connects the segmental arc of two described section objects;

Segmental arc weights and the segmental arc number of segmental arc in parallel each described subregion adjacent map of statistics, the current cutting state scale parameter of the described Region adjacency graph of calculating, judge that whether described current cutting state scale parameter is greater than default scale parameter;

If described current cutting state scale parameter is not more than default scale parameter, merge two associated section objects to be combined of segmental arc to be combined, and upgrade in each described subregion adjacent map between section object in abutting connection with topological relation, return to segmental arc weights and segmental arc number in each described subregion adjacent map of statistics;

If described current cutting state scale parameter, greater than described default scale parameter, has been cut apart.

2. Remote Sensing Image Segmentation according to claim 1, is characterized in that, the step of two section objects to be combined that described merging segmental arc to be combined is associated specifically comprises:

, according to default merge algorithm, merge two associated section objects to be combined of segmental arc to be combined.

3. Remote Sensing Image Segmentation according to claim 1, is characterized in that, described preset data dividing mode is the chessboard division mode; Described segmental arc weights are the heterogeneity value of two section objects connecting of described segmental arc.

4. Remote Sensing Image Segmentation according to claim 1, is characterized in that, the current cutting state scale parameter of the described Region adjacency graph of described calculating comprises:

Described segmental arc weights in subregion adjacent map according to each, the minimum heterogeneous value of the described Region adjacency graph of statistics, wherein said minimum heterogeneous value equals the minimum segmental arc weights of all segmental arcs in described Region adjacency graph;

Described segmental arc weights and segmental arc number in subregion adjacent map according to each, calculate described average heterogeneous value, and wherein said average heterogeneous value equals the segmental arc weights sum of all segmental arcs in described Region adjacency graph divided by the segmental arc sum;

Calculate described current cutting state scale parameter according to described average heterogeneous value, wherein said current cutting state scale parameter equals described average heterogeneous value square root.

5. Remote Sensing Image Segmentation according to claim 2, is characterized in that, described default merge algorithm is heterogeneous Minimum Area object merging algorithm;

The default merge algorithm of described basis merges two associated section objects to be combined of segmental arc to be combined, and upgrade in each described subregion adjacent map between section object in abutting connection with topological relation, comprising:

According to heterogeneous Minimum Area object merging algorithm, parallel each described subregion adjacent map of traversal, obtain segmental arc to be combined, and the segmental arc weights of described segmental arc to be combined equal described minimum heterogeneous value;

Judge whether two associated section objects to be combined of described segmental arc to be combined are the interior zone object, described interior zone object refers to that all section objects adjacent with described interior zone object are all in same described subregion adjacent map;

If two described section objects to be combined are described interior zone object, parallel two associated section objects to be combined of described segmental arc to be combined that merge, upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with topological relation;

If have the borderline region object in two described section objects to be combined, described segmental arc to be combined is stored in interim array, travel through described interim array and obtain described segmental arc to be combined, described borderline region object refers to that the section object adjacent with described borderline region object be not entirely in same subregion adjacent map;

Two described section objects to be combined that the described segmental arc to be combined of erial merge is associated, upgrade in Region adjacency graph between section object in abutting connection with between the topological relation section object in abutting connection with topological relation.

6. Remote Sensing Image Segmentation according to claim 5, it is characterized in that, if described two described section objects to be combined are described interior zone object, walk abreast and merge two associated section objects to be combined of described segmental arc to be combined, upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with between the topological relation section object in abutting connection with topological relation, comprising:

Parallel two the associated section objects to be combined of described segmental arc to be combined that merge;

The subregion adjacent map at parallel traversal described section object to be combined place, search other associated segmental arc of described section object to be combined;

Upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with between the topological relation section object in abutting connection with topological relation.

7. Remote Sensing Image Segmentation according to claim 5, is characterized in that, two described section objects to be combined that the described segmental arc to be combined of described erial merge is associated, upgrade in Region adjacency graph between section object in abutting connection with topological relation, comprising:

Two described section objects to be combined that the described segmental arc to be combined of erial merge is associated;

The described Region adjacency graph of serial traversal, search other associated segmental arc of described section object to be combined;

Upgrade in described Region adjacency graph between section object in abutting connection with topological relation.

8. Remote Sensing Image Segmentation according to claim 7, is characterized in that, the described Region adjacency graph of described serial traversal, search other associated segmental arc of described section object to be combined, comprising:

Obtain the primary importance information of described section object to be combined;

, according to the primary importance information of described section object to be combined, obtain the second place information of the subregion adjacent map at the section object place adjacent with described section object to be combined;

, to described subregion adjacent map, travel through described subregion adjacent map according to described second place Information locating, search other associated segmental arc of described section object to be combined.

9. a Remote Sensing Image Segmentation device, is characterized in that, comprises that the first processing module, the second processing module, the 3rd processing module and manage module everywhere, wherein:

Described the first processing module, be used for the mode of original remote sensing image with Region adjacency graph represented, according to the preset data dividing mode, described Region adjacency graph being carried out data divides, obtain a plurality of subregion adjacent maps, described subregion adjacent map comprises section object and connects the segmental arc of two described section objects;

Described the second processing module, the segmental arc weights and the segmental arc number that are used for parallel each described subregion adjacent map segmental arc of statistics, calculate the current cutting state scale parameter of described Region adjacency graph, judge that whether described current cutting state scale parameter is greater than default scale parameter;

Described the 3rd processing module, be not more than default scale parameter if be used for described current cutting state scale parameter, merge two associated section objects to be combined of described segmental arc to be combined, upgrade in each described subregion adjacent map between section object in abutting connection with topological relation, notify the second processing module again to add up segmental arc weights and segmental arc number in each described subregion adjacent map;

Described the manages module everywhere,, if be used for described current cutting state scale parameter greater than described default scale parameter, finishes to cut apart flow process.

10. Remote Sensing Image Segmentation device according to claim 9, it is characterized in that, described the second processing module also is used for segmental arc weights and the segmental arc number of parallel each described subregion adjacent map segmental arc of statistics, according to default merge algorithm, calculate the current cutting state scale parameter of described Region adjacency graph, judge that whether described current cutting state scale parameter is greater than default scale parameter.

11. Remote Sensing Image Segmentation device according to claim 9, is characterized in that, described preset data dividing mode is the chessboard division mode; Described segmental arc weights are the heterogeneity value of two section objects connecting of described segmental arc.

12. remote sensing image Parallel segmentation device according to claim 9, is characterized in that, described the second processing module comprises:

Parallel statistics submodule, be used for parallel segmental arc weights and segmental arc number of adding up each described subregion adjacent map segmental arc;

Minimum heterogeneous value calculating sub module, the described segmental arc weights that are used for subregion adjacent map according to each, add up the minimum heterogeneous value of described Region adjacency graph, wherein said minimum heterogeneous value equals the minimum segmental arc weights of all segmental arcs in described Region adjacency graph;

Average heterogeneous value calculating sub module, the described segmental arc weights and the segmental arc number that are used for subregion adjacent map according to each, calculate described average heterogeneous value, wherein said average heterogeneous value equals the segmental arc weights sum of all segmental arcs in described Region adjacency graph divided by the segmental arc sum;

Current cutting state scale parameter calculating sub module, be used for calculating described current cutting state scale parameter according to described average heterogeneous value, and wherein said current cutting state scale parameter equals described average heterogeneous value square root;

Current cutting state scale parameter judgement submodule, be used for judging that whether described current cutting state scale parameter is greater than default scale parameter.

13. Remote Sensing Image Segmentation device according to claim 10, is characterized in that, described default merge algorithm is heterogeneous Minimum Area object merging algorithm;

Described the 3rd processing module comprises:

Traversal is searched submodule, is used for according to heterogeneous Minimum Area object merging algorithm, and parallel each described subregion adjacent map of traversal, obtain segmental arc to be combined, and the segmental arc weights of described segmental arc to be combined equal described minimum heterogeneous value;

Interior zone object judgement submodule, be used for judging whether two associated section objects to be combined of described segmental arc to be combined are the interior zone object, described interior zone object refers to that all section objects adjacent with described interior zone object are all in same described subregion adjacent map;

The parallel submodule that merges, be described interior zone object if be used for two described section objects to be combined, parallel two associated section objects to be combined of described segmental arc to be combined that merge, upgrade in the subregion adjacent map at described section object to be combined place between section object in abutting connection with topological relation;

The erial merge submodule, if have the borderline region object in two described section objects to be combined, described segmental arc to be combined is stored in interim array, travel through described interim array and obtain described segmental arc to be combined, described borderline region object refers to that the section object adjacent with described borderline region object be not entirely in same subregion adjacent map; Two described section objects to be combined that the described segmental arc to be combined of erial merge is associated, upgrade in Region adjacency graph between section object in abutting connection with topological relation.

14. Remote Sensing Image Segmentation device according to claim 13, is characterized in that, described parallel merging submodule comprises:

Parallel merge cells, be used for parallel two the associated section objects to be combined of described segmental arc to be combined that merge;

Parallel traversal unit, be used for the parallel subregion adjacent map that travels through described section object to be combined place, searches other associated segmental arc of described section object to be combined;

The first updating block, be used for upgrading between the subregion adjacent map section object at described section object to be combined place in abutting connection with topological relation.

15. Remote Sensing Image Segmentation device according to claim 13, is characterized in that, described erial merge submodule comprises:

The erial merge unit, be used for two associated described section objects to be combined of the described segmental arc to be combined of erial merge;

Serial traversal unit, be used for the described Region adjacency graph of serial traversal, searches other associated segmental arc of described section object to be combined;

The second updating block, be used for upgrading between described Region adjacency graph section object in abutting connection with topological relation.

16. Remote Sensing Image Segmentation device according to claim 15, is characterized in that, described serial traversal unit comprises:

Primary importance acquisition of information subelement, for the primary importance information of obtaining described section object to be combined;

Second place acquisition of information subelement, be used for the primary importance information according to described section object to be combined, obtains the second place information of the subregion adjacent map at the section object place adjacent with described section object to be combined;

Serial traversal subelement, be used for arriving described subregion adjacent map according to described second place Information locating, travels through described subregion adjacent map, searches other associated segmental arc of described section object to be combined.

Images