CN102663394B - Method of identifying large and medium-sized objects based on multi-source remote sensing image fusion - Google Patents

Method of identifying large and medium-sized objects based on multi-source remote sensing image fusion Download PDF

Info

Publication number
CN102663394B
CN102663394B CN 201210054296 CN201210054296A CN102663394B CN 102663394 B CN102663394 B CN 102663394B CN 201210054296 CN201210054296 CN 201210054296 CN 201210054296 A CN201210054296 A CN 201210054296A CN 102663394 B CN102663394 B CN 102663394B
Authority
CN
China
Prior art keywords
target
feature
sub
goal
level
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN 201210054296
Other languages
Chinese (zh)
Other versions
CN102663394A (en
Inventor
李波
胡蕾
丁浩
季艳
田越
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beihang University
Original Assignee
Beihang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beihang University filed Critical Beihang University
Priority to CN 201210054296 priority Critical patent/CN102663394B/en
Publication of CN102663394A publication Critical patent/CN102663394A/en
Application granted granted Critical
Publication of CN102663394B publication Critical patent/CN102663394B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)

Abstract

The invention provides a method of identifying large and medium-sized objects based on multi-source remote sensing image fusion. The method comprises that multispectral and SAR images are segmented into image areas according to spectrum distribution between different objects and differences of electromagnetic radiation, and a potential area of an object is extracted according to experience and knowledge of object distribution, so that characteristic extraction and object identification are more targeted while identification efficiency and identification accuracy of the system are improved; that based on the above, object areas are determined by extracting object contour, spatial layout and the like, and sub-object areas are divided according to spatial layout relation of the object contour and sub-object spatial layout; and that sub-object characteristics are extracted, and with the guidance of object characteristic set and object prior knowledge, sub-object identification and object verification are finally realizedby execution of identification rules and matching of characteristics.

Description

Big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration
Technical field
The present invention relates to a kind of remote sensing images target automatic identifying method, relate in particular to and a kind ofly realize the automatically method of identification of big-and-middle-sized target based on the multi-source remote sensing image feature extraction techniques, belong to the digital image processing techniques field.
Background technology
The application of remote sensing technology has contained many aspects such as environmental monitoring, resource exploration, the reallocation of land, diaster prevention and control, ground mapping, weather forecast, for great contribution has been made in the national economic development.Wherein the identification of big-and-middle-sized terrain object such as airport, harbour, bridge, oil depot, boats and ships, aircraft is one of research focus of remote sensing technology application.Along with the fast development of aviation/spacer remote sensing technology, the image of various airborne, visible lights that satellite borne sensor obtains, radar, type such as multispectral provides abundant information source for target identification, also makes the utilization of information become complicated.There is complementarity between the information to same target different sensors acquisition, can more intactly reflects the characteristic of target, be conducive to improve discrimination, get rid of false-alarm.
Multi-source image merges from information sign level, can be divided into Pixel-level, feature level and three levels of decision level, and identification belongs to feature level fusing method based on the multi-source image fusion goal, and these class methods are carried out analysis-by-synthesis and processing with the characteristic information that extracts.In the current target automatic identifying method, the most representative have based on statistics, based on knowledge and based on the target automatic identifying method of model.Target automatic identifying method based on statistics is on the basis that great amount of samples is trained, and obtains the statistical study of target property, adopts the characteristic matching of space length tolerance to identify target.This method relatively effectively, but is difficult to solve problem such as object construction variation under the low situation of target background complexity.Based on the target automatic identifying method of knowledge mainly be with expert system application in automatic target identification, overcome the limitation of classical statistics pattern-recongnition method to a certain extent, but be difficult in scene, find effectively and organization knowledge.Based on the target automatic identifying method of model, on the basis to object modeling, extract certain target signature, come the model parameter of target-marking in conjunction with some supplementary knowledges, identifying is the matching process of model and target signature.The changeable structure that is difficult to be adapted to complex target based on the method for model.
Remote sensing images have the characteristics of wide coverage, and full figure extracts that the target signature calculated amount is big, and interference is many, the potential zone of target be positioned with the scope that helps dwindle target's feature-extraction and analysis.And pay little attention to for the potential zone of target in the existing method.In the existing target's feature-extraction process, based on the method for statistics, be that extraction target signature as much as possible is analyzed especially.And in the realistic objective identifying for certain specific objective, the validity difference of its different characteristic is selected effective feature, is conducive to reduce the complexity of target's feature-extraction and analysis.Although the version of target is changeable, generally speaking, the formation of target (sub-goal) is relatively stable, and the spatial relationship between the parton target also is certain.Existing recognition methods fails to bring into play formation and the effect of the spatial relationship between sub-goal in identification of target.The complementarity of combining target feature in the multi-source remote sensing image, the present invention proposes a kind of big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration from the aspects such as effective checking of effective extraction of effective extraction in the potential zone of target, target signature and analysis, target.
Summary of the invention
Technical matters to be solved by this invention is to provide a kind of big-and-middle-sized target identification method that merges based on multi-source remote sensing image (panchromatic, multispectral, SAR) feature.This method synthesis utilizes Feature Extraction Technology and the fusion method of multi-source remote sensing image, realization has been finished the automatic identification to big-and-middle-sized targets such as airport, harbour, bridge, oil depot, boats and ships, aircrafts from a series of images treatment steps such as Region Segmentation, feature extraction and analysis, target identification checkings.
For achieving the above object, the present invention adopts following technical proposals:
Based on the big-and-middle-sized target identification method of multi-source remote sensing image co-registration, its characteristics are to comprise the steps:
(1) feature hierarchyization and relationship analysis: in conjunction with human cognitive, target signature is carried out stratification divide, set up the selection criterion of high-level characteristic centering low-level feature;
(2) the potential extracted region of Region Segmentation and target: utilize multispectral and SAR remote sensing images to carry out Region Segmentation, the zone at combining target and targeted environment place, determine image cut zone (terrain classification type), and according to spectrum and the electromagnetic scattering characteristic of all kinds of atural objects multispectral and SAR image are carried out Region Segmentation respectively; Which zone is combining target geographic distribution knowledge analyze and may have target to be identified in the cut zone of multispectral and SAR image, obtain the potential zone of target.
(3) target area is extracted with the sub-goal zone and is divided: according to the potential regional location of target, utilize features such as space layout and profile further to draw a circle to approve the target area in full-colour image; Utilize the spatial relationship of sub-goal in the concept characteristic, target is decomposed into separate sub-goal zone;
(4) sub-goal identification and target checking: the selection criterion that utilizes attributive character, topological structure feature centering low-level feature, in panchromatic, multispectral and SAR image, extract the local feature of sub-goal respectively, and feature analyzed, thereby judge whether it is sub-goal; According to the sub-goal that identifies, and the spatial relationship between sub-goal, target configuration knowledge etc., target is verified.
The recognition methods that the present invention proposes has following different with classic method: entire identification process is divided into Region Segmentation in the present invention and extract and the division of sub-goal zone, sub-goal identification and target checking the potential extracted region of target, target area.Under the guidance of target concept feature, extract the attributive character of target etc., realize the Region Segmentation of image, obtain the attributive character of targeted environment and the attributive character of target self, thereby obtain the potential zone of target, dwindled target signature and further extracted and the scope of analyzing.By contour feature and the space layout feature in the potential extracted region target of target, mark off the sub-goal zone, and identify by the middle low-level feature antithetical phrase target area of sub-goal, finish the checking of target by the identification of sub-goal and the space layout relationship analysis between sub-goal.Coming evaluating objects from the space layout between target configuration (sub-goal) and target configuration is that other recognition methods institutes are NM, the validity of low-level feature extraction and analysis in so not only being conducive to, the false-alarm that does not meet target configuration and constitute space layout can also be effectively got rid of, efficient and the accuracy of target identification can be improved.
Description of drawings
The present invention is described in further detail below in conjunction with the drawings and specific embodiments.
Fig. 1 is that this is based on the big-and-middle-sized target identification method process flow diagram of multi-source remote sensing image co-registration;
Fig. 2 is in this big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration, sets up feature selecting criterion process example;
Fig. 3 is in this big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration, multispectral image Region Segmentation decision tree;
Fig. 4 is in this big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration, the potential extracted region process flow diagram of target;
Fig. 5 is in this big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration, the main type of space layout feature;
Fig. 6 is in this big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration, utilizes the process synoptic diagram of profile and space layout feature extraction sub-goal;
Fig. 7 is that this is based on the reasoning flow process of the big-and-middle-sized target identification method of multi-source remote sensing image co-registration.
Embodiment
As shown in Figure 1, it is four key steps that realization of the present invention is divided into, and is respectively: the potential extracted region of feature hierarchyization and relationship analysis, Region Segmentation and target, target area are extracted with the sub-goal zone and are divided, sub-goal identification and target checking.Be identified as example with big-and-middle-sized targets such as the airport in the multi-source remote sensing image (multispectral, SAR, panchromatic), harbours below, concrete implementation step of the present invention is described in detail.
(1) feature hierarchyization and relationship analysis: target signature is divided into five levels, and according to the correlation experience knowledge of target, analyzes the selection effect that high-level characteristic centering low-level feature has, set up the selection criterion of high-level characteristic centering low-level feature;
1) multiple stratification of target signature is represented
The target correlated characteristic is many and complicated, for can better the descriptive analysis target signature, the present invention proposes a kind of multi-level method for expressing of target signature of five layer architectures, be specially: feature is divided into low-level feature, middle level feature, topological structure feature, attributive character and five levels of concept characteristic, every layer of feature mainly comprise in have:
1. low-level feature, the feature relevant with single pixel concluded low-level feature, typically comprise atural object radiation value in tone, brightness, gray-scale value, the remote sensing images (in multispectral image, show as the pixel spectral value, in full-colour image, show as the pixel gray scale);
2. middle level feature, the feature relevant with block of pixels (a plurality of pixels are assembled the zone that forms) concluded the middle level feature, typically comprise the edge in texture (as gray scale symbiosis square, gradient, entropy etc.), region shape (as area, length and width, length breadth ratio, geometric moment, skeleton etc.) and zone (interior) in the zone etc.;
3. the topological structure feature is summarized as topological structure level high-level characteristic with the spatial relationship between single object or a plurality of object, typically comprise space layout relation between contours of objects, target configuration part (as adjacent, from, parallel etc.) and direction etc.;
4. attributive character, the feature that reflection object physical attribute, state and imaging mechanism is relevant is summarized as attributive character, the polarization mode the when wave band when typically comprising time, the remote sensing images imaging of atural object material (as water, plant, culture etc.), remote sensing images imaging, SAR image imaging etc.;
5. concept characteristic is summarized as concept characteristic with object entity and the entity knowledge relevant with the object entity;
Topological structure feature, attributive character and concept characteristic belong to high-level characteristic, and Pixel-level low-level feature and region class middle level feature belong to middle low-level feature.The main foundation that level is divided is: by the visual information of low layer, progressively rise to semantic information.
2) set up the selection criterion of target high-level characteristic centering low-level feature
Mainly set up the feature selecting criterion according to the level division of feature used in the target identification experimental knowledge, feature correspondence and the relation between feature.Mainly set up concept characteristic to the selection criterion of attributive character, topological structure feature and middle low-level feature, and the selection criterion of attributive character, topological structure feature centering low-level feature.
Preferentially extract the significant attributive character of target, topological structure feature and middle low-level feature according to target concept, wherein the relation of target attributive character significant with it, topological structure feature and middle low-level feature is as shown in table 1.In like manner, according to the relation between middle low-level feature and high-level characteristic, set up the selection criterion of attributive character, topological structure feature centering low-level feature, typical criterion is as shown in table 2.In Fig. 2, aircraft is selected attributive character, topological structure features such as slickness, metal, adjacent, symmetry under the guidance of experimental knowledge, smooth attributive character selects textural characteristics such as gradient to extract, and adjacent and symmetric space layout relationship feature selecting edge and shape facility extract.
The relationship analysis of table 1 target concept feature and attributive character, topological structure feature, middle low-level feature
Target Attributive character The topological structure feature Middle low-level feature
The airport Smooth, material Parallel, adjacent Area
The harbour Material Parallel, vertical Do not have
Bridge Material Parallel, adjacent Do not have
Oil depot Material Be separated by, array Do not have
Ship Material Do not have Area, shape
Aircraft Material, smooth Symmetrical, adjacent Area
The relationship analysis of table 2 attributive character, topological structure feature and middle low-level feature
Middle low-level feature
Smooth attributive character Texture, backscatter intensity
The material properties feature Spectrum, backscatter intensity
The topological structure feature Edge, shape
(2) the potential extracted region of Region Segmentation and target
At first, utilize multispectral and SAR remote sensing images to carry out Region Segmentation, wherein multi-spectral remote sensing image adopts the partitioning scheme of cutting apart decision tree based on the multispectral image of spectral signature, and the SAR remote sensing images adopt the Markov dividing method.
The Multi-spectral Remote Sensing Data that invention is mainly cut apart is the multispectral image of the 8bit Quickbird of 2.44 meters resolution, selected part water, vegetation and culture wait for that cut zone is as sample areas, to these sample areas from each band spectrum value, spectral ratio and NDVI features such as (normalized differential vegetation indexs) is carried out statistical study, obtain cutting apart decision tree based on the multispectral image of spectral signature as shown in Figure 3, B wherein, G, R and N represent blue wave band respectively, green wave band, the spectral value of red wave band and near-infrared band
Figure BDA0000140184510000061
Min and max represent 4 minimum and maximum spectral values in the wave band, R respectively MaxRepresent maximum ratio between 4 wave bands.Utilize multispectral image to cut apart decision tree multispectral image is cut apart, obtain water, vegetation, common culture, large-sized artificial building and the zone of cutting apart such as highlighted.
At the characteristics of SAR image, the present invention adopts the image partition method based on MRF, and the zone in the SAR image is divided into weak back scattering zone, common back scattering zone and strong back scattering zone.
Then, utilize prioris such as the environmental characteristic of target and attributive character, in the potential zone that the big regional center that has extracted sets the goal.In the potential extracted region process of target, involved multispectral and SAR image cut zone and part principal character are as shown in table 3, and idiographic flow as shown in Figure 4.For not omission target, the used qualifications of its analytic process is often not strong.
Cut zone and feature that the potential regional analysis of table 3 target mainly utilizes
Target Multispectral image The SAR image
The airport The large-sized artificial construction area, area is greater than 100,000 square meters Weak back scattering zone
The harbour The large-sized artificial construction area is positioned at the land and water boundary place Strong back scattering zone
Bridge Common artificial construction area, both sides are the waters Strong back scattering zone
Oil depot Highlight regions, circle Strong back scattering zone
Ship Non-aqua region is in the waters Strong back scattering zone
Aircraft Highlight regions is positioned at the airport Strong back scattering zone
(3) target area is extracted with the sub-goal zone and is divided: according to the potential regional location of target, utilize features such as profile and space layout further to draw a circle to approve the target area in full-colour image; Utilize the spatial relationship of sub-goal in the concept characteristic, target is decomposed into separate sub-goal zone;
1) extract the target area
The level set method that at first adopts based target and background contrasts to strengthen is extracted contour feature.Its step is as follows:
1. utilize wavelet transformation with image mapped for wavelet low frequency component (LL) and level, vertical, to angular direction high fdrequency component (HL, LH, HH).
2. set up the disaggregated model of Wavelet Component, specific as follows:
F ( c ) = μ · Length ( c ) + Σ k = 1 4 λ k ∫ c i | u k ( x , y ) - u k , 1 | 2 dxdy + Σ k = 1 4 λ k ∫ c o | u k ( x , y ) - u k , 2 | 2 dxdy
Wherein, c is profile, c iBe the target area, c 0Be the background area, Length (c) is the length of profile, u K, 1Be the Wavelet Component value of target area, u K, 2Be background area Wavelet Component value, u k(x y) is respectively LL, HL, LH and the HH component of wavelet transformation, μ, λ 1, λ 2, λ 3And λ 4Be weighting coefficient, the weights λ of Wavelet Component 1, λ 2, λ 3And λ 4Set according to its conforming power.
Then profile is carried out vector quantization, utilize the method for perception marshalling that the vector line segment is analyzed, comprise mainly that the line segment on the same direction connects, the vertical relation between line segment between parallel relation, line segment etc.
2) sub-goal Region Decomposition
Result in conjunction with the contour vector marshalling, determine the contour edge of remarkable sub-goal earlier according to the length of remarkable sub-goal, curvature, rim space relation etc., thereby obtain remarkable sub-goal, recycling sub-goal space layout feature and objective contour edge obtain other non-remarkable sub-goals.Wherein significantly sub-goal refers to have that area is big, the edge long, the target configuration part of regular shape characteristic, and other target configurations partly are called non-remarkable sub-goal.The remarkable sub-goal of the part of target and non-remarkable sub-goal definition see Table 4.The object space spatial layout feature comprise in abutting connection with, intersect, from, comprise, parallel, vertical, pass through, array etc., see Fig. 5.Fig. 6 has provided airport target, by contour feature, in conjunction with the space layout feature, obtains the process in sub-goal zone.
The main sub-goal of table 4 target
Target Remarkable sub-goal Non-remarkable sub-goal
The airport Runway, secondary runway Hardstand, taxiway
The harbour The berth Access bridge
Bridge Bridge floor Do not have
Oil depot Oil drum Do not have
Ship Hull Pilothouse
Aircraft Fuselage, wing Tail
(4) sub-goal identification and target checking: the selection criterion that utilizes attributive character, topological structure feature centering low-level feature, in panchromatic, multispectral and SAR image, extract the local feature of sub-goal respectively, and feature analyzed, thereby judge whether it is sub-goal; According to the sub-goal that identifies, and the spatial relationship between sub-goal, target configuration knowledge etc., target is verified.The feature of sub-goal comprises spectral value, the backscatter intensity in the SAR image and the shape facility in the full-colour image of each wave band in the multispectral image.Adopt bottom-up target verification mode, low-level feature is identified sub-goal in namely utilizing, and utilizes topological structure feature and attributive character etc. to the target checking, and gets rid of false-alarm.
1) sub-goal identification
Utilize the attributive character in the embodiment (1), the selection criterion of topological structure feature centering low-level feature, the attributive character of combining target, topological structure feature, in panchromatic, multispectral and SAR image, extract the local feature of sub-goal respectively, and according to local feature, sub-goal is identified.
The middle low-level feature that extracts in full-colour image mainly comprises features such as the area of target, external square length, form parameter, and wherein, the account form of form parameter is P 2/ (4 π S), P is the girth of target, S is the area of target.
The middle low-level feature that extracts in multispectral image mainly comprises spectral value, the spectral ratio between wave band, NDVI value, each band spectrum average of zone of each wave band etc.
The middle low-level feature that extracts in the SAR image mainly comprises gray-scale value, regional average, gradient etc.
2) target checking
According to the sub-goal that identifies, and the spatial relationship between sub-goal, target configuration knowledge etc., target is verified.Identification as the airport; if the sub-goal that identifies comprises runway, secondary runway, hardstand and connecting taxiway; then think the formation knowledge that meets target; further; be to meet between sub-goal the space layout relation between runway and secondary runway if the runway that identifies and secondary runway are parallel space layout relationship, connecting taxiway; then think to meet the space layout relation between sub-goal, thereby realize the checking of target.
3) reasoning process
For realizing above feature and knowledge to the support effect of target identifying, constructed the knowledge base that is used for supporting target identification reasoning process, be used for the intermediate result that the storage reasoning process produces factbase, be used for the feature database of each category feature empirical value of storage target.
Knowledge base is represented knowledge with the form of production, and the expression structure of rule is described below with the BNF normal form:
＜rule〉:=RULE＜rule number 〉
PREMISE ($AND{＜condition〉})
ACTION{＜action〉}
＜condition〉:=＜simple condition〉| ($OR{＜simple condition〉})
Wherein, simple condition represents that with two class simple function ASK_FACT and ASK_CHARACTER its BNF normal form is described below:
(ASK_FACT＜object〉＜attribute〉＜value 〉)
(ASK_CHARACTER＜object〉＜attribute〉＜value 〉)
When inference machine makes an explanation execution to rule, ASK_FACT function indication inference machine inquiry factbase confirms whether this object has this property value, and ASK_CHARACTER function indication inference machine query characteristics storehouse confirms whether the feature (feature set) of this object meets definition in advance.
Action in the rule represents that with simple function CONCLUDE its property value with object is added factbase as the reasoning conclusion, and the BNF normal form of CONCLUDE function is described below:
(CONCLUDE＜object〉＜attribute〉＜value 〉)
The present invention shows with above-mentioned form shfft and has collected 45 rules in knowledge base that wherein, 19 of rules are detected in potential target zone and target area, and are as shown in table 5; 26 of identification proof rules, as shown in table 6.
Rule is detected in potential target zone and target area in table 5 knowledge base
Regular number Former piece Consequent
R27 Inquire target type to be identified The potential zone of search related objective
R28 The potential zone of search airport target The large-sized artificial building
R29 The searching code head marks potential zone The land and water boundary place
… … …
Target identification proof rule in table 6 knowledge base
Regular number Former piece Consequent
R01 There is target to be declared Check target type to be declared
R02 Target type to be declared is the airport Whether this airport is waited to declare target and can be divided
R03 Target type to be declared is harbour Whether this harbour is waited to declare target and can be divided
… … …
Potential target zone and target area are detected rule and are mainly used in the extraction instructing the potential zone of target, wait to declare target and the extraction of relevant middle low-level feature; The major function of identification proof rule is that low-level feature is identified each ingredient of big-and-middle-sized target in utilizing, and utilizes the composition of target and spatial relationship feature to finish and treat the checking of declaring target.
Be example with sub-goal recognition rule R11 wherein, be expressed as follows:
Rule 11:
If: the airport waits that the sub-goal of declaring target meets the primary runway feature;
So: this specific item is designated as primary runway.
Its formalized description is as follows:
RULE11
PREMISE(ASK_CHARACTER SUB_TARGET EIGENVECTOR
RUNWAY_CHARACTER)
ACTION(CONCLUDE SUB_TARGET TYPE RUNWAY)
Wherein, whether the feature set that regular former piece is illustrated in inquiry primary runway in the feature database conforms to the feature set of this sub-goal, and rreturn value is 0 or 1; The rule consequent is represented the type mark of this sub-goal for " primary runway " and deposit in the factbase.
The matched rule of feature is true in the min/max interval of priori eigenwert for the eigenwert of input, otherwise is false.By the eigenwert statistics to related objective in all kinds of images, the present invention has summed up the empirical features value scope of all kinds of targets and has been stored in the feature database, and is as shown in table 4.
The empirical features value of all kinds of targets in table 4 feature database
Figure BDA0000140184510000101
Reasoning process adopts the inference method that knowledge drives and data-driven combines, and finishes in multi-source image search and identification to intended target in conjunction with the depth-first search strategy.Reasoning process is constantly searched for the correlated characteristic of target to be identified by the feature selecting criterion, activates corresponding feature extraction operator.The result that the action that these regular consequents comprise produces can be used as the new fact again, activates other rules and further finds new fact.So analogize, the formation rule chain is finally finished identifying.Identification reasoning with airport target is example, and the reasoning flow process that is caused by said process as shown in Figure 7.
Big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration provided by the present invention mainly proposes for big-and-middle-sized object recognition rate is special in the raising high-resolution remote sensing image.But obviously, method provided by the present invention can also be expanded for the identification to other type target, and obtained beneficial effect also is similar.
More than the big-and-middle-sized target identification method based on the multi-source remote sensing image co-registration provided by the present invention is had been described in detail, but obvious specific implementation form of the present invention is not limited thereto.For the those skilled in the art of the art, the various apparent change of under the situation that does not deviate from claim scope of the present invention it being carried out is all within protection scope of the present invention.

Claims (1)

1. based on the big-and-middle-sized target identification method of multi-source remote sensing image co-registration, it is characterized in that following steps:
(1) on the basis that the target signature level is divided, set up the selection criterion of target high-level characteristic centering low-level feature, comprise concept characteristic to the selection criterion of attributive character, topological structure feature and middle low-level feature, and the selection criterion of attributive character, topological structure feature centering low-level feature;
(2) utilize the concept characteristic of target to the selection criterion of attributive character, spectral distribution and electromagnetic scattering characteristic according to atural object are carried out Region Segmentation respectively to multispectral and SAR image, obtain airport, harbour, oil depot, bridge, aircraft, the potential zone of boats and ships classification target in multispectral and SAR image cut zone;
(3) in the potential zone of the target of full-colour image correspondence, utilize space layout and contour feature to extract the target area, the spatial relationship of sub-goal in the recycling concept characteristic is decomposed into separate sub-goal with target;
(4) utilize the selection criterion of attributive character, topological structure feature centering low-level feature, in the target area of panchromatic, multispectral and SAR image, extract the feature of sub-goal respectively, and carry out the identification of sub-goal and the checking of target;
In the described step (1), the target signature level is divided clarification of objective is divided into five levels: low-level feature, middle level feature, topological structure feature, attributive character and concept characteristic;
In the described step (2), the Region Segmentation type is waters, vegetation territory, common culture territory, large-sized artificial building territory and highlight regions;
In the described step (3), space layout is divided into: in abutting connection with, intersect, from, comprise, parallel, vertical, pass through, array type; The level set method that the method for utilizing contour feature to extract the target area adopts based target and background contrasts to strengthen;
In the described step (3), utilize the method for perception marshalling that objective contour is described, determine the contour edge of remarkable sub-goal earlier in conjunction with length, curvature, the spatial relationship of contour edge, thereby obtain remarkable sub-goal, recycling sub-goal spatial relationship and objective contour edge obtain other non-remarkable sub-goals;
In the described step (4), the feature of sub-goal comprises: the middle low-level feature that extracts in full-colour image mainly comprises the area of target, external square length, form parameter; The middle low-level feature that extracts in multispectral image mainly comprises spectral value, the spectral ratio between wave band, NDVI value, each band spectrum average of zone of each wave band; The middle low-level feature that extracts in the SAR image mainly comprises gray-scale value, regional average, gradient;
In the described step (4), adopt bottom-up target verification mode, low-level feature is identified sub-goal in namely utilizing, and utilizes topological structure feature and attributive character that target is verified, and gets rid of false-alarm.
CN 201210054296 2012-03-02 2012-03-02 Method of identifying large and medium-sized objects based on multi-source remote sensing image fusion Active CN102663394B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 201210054296 CN102663394B (en) 2012-03-02 2012-03-02 Method of identifying large and medium-sized objects based on multi-source remote sensing image fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201210054296 CN102663394B (en) 2012-03-02 2012-03-02 Method of identifying large and medium-sized objects based on multi-source remote sensing image fusion

Publications (2)

Publication Number Publication Date
CN102663394A CN102663394A (en) 2012-09-12
CN102663394B true CN102663394B (en) 2013-09-25

Family

ID=46772878

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 201210054296 Active CN102663394B (en) 2012-03-02 2012-03-02 Method of identifying large and medium-sized objects based on multi-source remote sensing image fusion

Country Status (1)

Country Link
CN (1) CN102663394B (en)

Families Citing this family (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103020605B (en) * 2012-12-28 2016-04-06 北方工业大学 Bridge identification method based on decision-making layer fusion
CN103020975A (en) * 2012-12-29 2013-04-03 北方工业大学 Wharf and ship segmentation method combining multi-source remote sensing image characteristics
CN103456006B (en) * 2013-08-06 2016-04-20 北京航空航天大学 A kind of oil depot object localization method of SAR image
CN103544502A (en) * 2013-10-29 2014-01-29 上海市城市建设设计研究总院 High-resolution remote-sensing image ship extraction method based on SVM
CN103761741B (en) * 2014-01-23 2017-02-08 电子科技大学 Variation polarized SAR image partition method based on polarization target decomposition feature
CN104794478B (en) * 2015-05-04 2017-12-19 福建师范大学 A kind of building extracting method for being used in remote sensing image have uniform spectral characteristic
CN105512616B (en) * 2015-11-27 2019-01-22 吉林大学 Remove the method that atmospheric noise influences to rebuild remote sensing monitoring data set
CN105528787A (en) * 2015-12-07 2016-04-27 清华大学 Polarimetric SAR image bridge detection method and device based on level set segmentation
CN105868794B (en) * 2016-04-19 2019-03-12 哈尔滨工业大学 A kind of Ship Target fuzzy recognition method based on ISAR image
CN106408027A (en) * 2016-09-23 2017-02-15 交通运输部规划研究院 Method and system for extracting image of logistics warehouse behind storage yard at port
CN106407941A (en) * 2016-09-23 2017-02-15 交通运输部规划研究院 Method and system for monitoring port sea reclamation based on object-oriented image analysis
CN106408572A (en) * 2016-09-23 2017-02-15 交通运输部规划研究院 Method and system for monitoring container storage yard at port by utilizing remote sensing image
CN107292245B (en) * 2017-06-02 2018-04-10 同济大学 A kind of harbour detection method on high score remote sensing image
CN108154471A (en) * 2017-11-15 2018-06-12 北京海致网聚信息技术有限公司 A kind of method and apparatus for remote sensing image splicing
CN108256419B (en) 2017-12-05 2018-11-23 交通运输部规划研究院 A method of port and pier image is extracted using multispectral interpretation
CN109115770B (en) * 2018-06-14 2019-05-24 中科禾信遥感科技（苏州）有限公司 A kind of a wide range of crops remote-sensing monitoring method and device
CN109508641A (en) * 2018-10-15 2019-03-22 广州地理研究所 A kind of recognition methods of urban water-body, device, storage medium and equipment
CN109431497B (en) * 2018-10-23 2020-08-11 南京医科大学 Electroencephalogram signal processing method and epilepsy detection system
CN109711446B (en) * 2018-12-18 2021-01-19 中国科学院深圳先进技术研究院 Ground feature classification method and device based on multispectral image and SAR image
CN111611921B (en) * 2020-05-21 2021-05-14 苏州科知律信息科技有限公司 Solar panel identification system based on remote sensing big data
CN114076923B (en) * 2020-08-20 2024-05-03 西安电子科技大学 Target recognition method based on time domain echo model of multilayer background and target
CN112580509B (en) * 2020-12-18 2022-04-15 中国民用航空总局第二研究所 Logical reasoning type road surface detection method and system
CN115019163A (en) * 2022-04-01 2022-09-06 湖南大学 City factor identification method based on multi-source big data

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102063710A (en) * 2009-11-13 2011-05-18 烟台海岸带可持续发展研究所 Method for realizing fusion and enhancement of remote sensing image
CN102254311A (en) * 2011-06-10 2011-11-23 中国科学院深圳先进技术研究院 Method and system for fusing remote sensing images

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102063710A (en) * 2009-11-13 2011-05-18 烟台海岸带可持续发展研究所 Method for realizing fusion and enhancement of remote sensing image
CN102254311A (en) * 2011-06-10 2011-11-23 中国科学院深圳先进技术研究院 Method and system for fusing remote sensing images

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
《多源遥感影像数据融合技术探讨》;姚静等;《测绘与空间地理信息》;20080229;第31卷(第1期);第29-31页 *
《多源遥感影像数据融合技术研究》;姜芸等;《测绘与空间地理信息》;20090430;第32卷(第2期);第46-50页 *
姚静等.《多源遥感影像数据融合技术探讨》.《测绘与空间地理信息》.2008,第31卷(第1期),第29-31页.
姜芸等.《多源遥感影像数据融合技术研究》.《测绘与空间地理信息》.2009,第32卷(第2期),第46-50页.

Also Published As

Publication number Publication date
CN102663394A (en) 2012-09-12

Similar Documents

Publication Publication Date Title
CN102663394B (en) Method of identifying large and medium-sized objects based on multi-source remote sensing image fusion
Turker et al. Field-based sub-boundary extraction from remote sensing imagery using perceptual grouping
CN101840581B (en) Method for extracting profile of building from satellite remote sensing image
CN106503739A (en) The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics
Liu et al. Multimorphological superpixel model for hyperspectral image classification
CN106529508A (en) Local and non-local multi-feature semantics-based hyperspectral image classification method
CN107832797B (en) Multispectral image classification method based on depth fusion residual error network
CN103632363A (en) Object-level high-resolution remote sensing image change detection method based on multi-scale fusion
CN103020605A (en) Bridge identification method based on decision-making layer fusion
Bunting et al. Characterisation and mapping of forest communities by clustering individual tree crowns
CN104103076A (en) Nuclear power plant planned restricted zone remote sensing inspecting method based on high-resolution remote sensing images
CN106228130A (en) Remote sensing image cloud detection method of optic based on fuzzy autoencoder network
CN103136545A (en) High resolution remote sensing image analysis tree automatic extraction method based on space consistency
Kahraman et al. Road extraction techniques from remote sensing images: A review
Sahu et al. Deep learning models for beans crop diseases: Classification and visualization techniques
Fan et al. Pavement cracks coupled with shadows: A new shadow-crack dataset and a shadow-removal-oriented crack detection approach
Singh et al. A hybrid approach for information extraction from high resolution satellite imagery
CN107121681B (en) Residential area extraction system based on high score satellite remote sensing date
Liu et al. A multi-angle comprehensive solution based on deep learning to extract cultivated land information from high-resolution remote sensing images
Khryashchev et al. Optimization of convolutional neural network for object recognition on satellite images
Dahiya et al. A comparative analysis of different land-use and land-cover classifiers using hyperspectral data
Wang et al. [Retracted] Remote Sensing Satellite Image‐Based Monitoring of Agricultural Ecosystem
CN101393641B (en) Airport target decomposition method based on context
Li et al. Measuring detailed urban vegetation with multisource high-resolution remote sensing imagery for environmental design and planning
CN115115948A (en) Forest land information fine extraction method based on random forest and auxiliary factors

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