CN102043958B - High-definition remote sensing image multi-class target detection and identification method - Google Patents
High-definition remote sensing image multi-class target detection and identification method Download PDFInfo
- Publication number
- CN102043958B CN102043958B CN2010105623195A CN201010562319A CN102043958B CN 102043958 B CN102043958 B CN 102043958B CN 2010105623195 A CN2010105623195 A CN 2010105623195A CN 201010562319 A CN201010562319 A CN 201010562319A CN 102043958 B CN102043958 B CN 102043958B
- Authority
- CN
- China
- Prior art keywords
- line segment
- image
- remote sensing
- area
- interest
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a high-definition remote sensing image multi-class target detection and identification method which comprises the following steps of: firstly, regulating the definition of an original remote sensing image to degrade the same into a plurality of images having different definitions; and then, respectively carrying out the following treatments on the images having different definitions from the lowest definition to the highest definition: (1) extracting regions of interest, (2) carrying out prior knowledge-aided identification, (3) extracting features, and (4) carrying out target identification with a previously trained classifier. By simply extracting and expressing shape features, analyzing shared features and differential features to select dynamic features and searching regions of interest according to the distribution of feature quantities, the invention provides a novel remote sensing image multi-stage multi-class target identification method based on multi-definition information and a large-scale texture analysis result. The method ensures high detection and identification reliability and simultaneously improves the information processing efficiency.
Description
Technical field
The invention belongs to digital image processing field, be specifically related to a kind of remote sensing images multi-class targets and detect recognition methods, to the target classification behave and make buildings, can discern automatically at different resolution multi-class targets.
Background technology
For remote sensing images, this area it is generally acknowledged that the remote sensing images smaller or equal to 10 meters imaging resolutions can be called high-resolution remote sensing image.The high-resolution remote sensing image that obtains now, except can carrying out the resource investigation, environmental monitoring etc. under the large scale, also for carrying out some other meticulousr analysis, like target detection identification, providing maybe; In addition, use, also need do the work of some target detection identifications in order to expand remote sensing images.
Use differently with Imaging Guidance etc., the target detection identification in the remote sensing images is being carried out in the scene on a large scale, and the target signal to noise ratio is low, and clutter is many.Target usually is mingled among the background that comprises bulk information.This just makes and from remote sensing images, extracts relatively difficulty of interested target.
The application of remote sensing images target detection identification at present mainly contains following dual mode: the one, and to simple target,, carry out the target detection recognition methods and accomplish through setting up the target signature model; Another kind is by the direct analysis image of interpretation personnel; Utilize the experience of its long-term training to carry out target detection and interpretation; But the workload of this Flame Image Process is very big, and data-handling efficiency is lower, and target detection and sentence read result receive the influence of interpretation personnel state.
The method of based target feature detection identification is often only to a class targets; Main through the evaluating objects background characteristics; The method of setting up target signature (invariant features or other validity features) model realizes, also can combine various prioris with the adaptive faculty of ensuring method to dynamic scene certainly.But there is following problem during to remote Sensing Image Analysis:
(1) is directed against a class targets, or even is arranged in certain specific objective of specific environment, when this method is used for remote Sensing Image Analysis, just need design a kind of method to each class targets.Detect the identification multi-class targets and with regard to adopting several different methods same image is repeatedly handled, its treatment effeciency can increase and sharply decline along with target classification number.
(2) it is limited that general target detects the image visual field that recognition methods is directed against, and can analyze its target/background characteristics, only need do among a small circle search and just can accomplish the target detection judgement of track rejection (or obtain).But in remote sensing image processing; Need be in very big scene domain inner analysis view data; Detect interested target and recognition objective classification, need the information processed amount huge, thus more than general target identification method will cause very high false alarm rate and misclassification rate.
First problem---the counting yield problem that identification is faced to the remote sensing image processing target detection, adopting the method for multi-class targets identification is a very natural solution.The main difficulty that the multi-class targets recognition methods faces is: each class targets is used identical feature description, and recognition efficiency can significantly descend along with the increase of target classification number.Therefore,, seek sharing feature and distinctive characteristic, make up multi-stage characteristics and select recognition system better to address this problem the target category classification.
To remote sensing image processing target detection identification faced second problem---remote sensing images comprise a large amount of background clutter information.Positive sample generally is target that from background, splits or the target that only comprises a small amount of background pixel when sorter is trained.Input has proposed higher requirement to sorter when identification like this, therefore needs to propose a kind of area-of-interest exacting method.
Make up the decision tree classification recognition system, extract area-of-interest, the target classification is carried out the extraction and the description of shape facility, this method logic flow is clear, is that the high-resolution remote sensing image multi-class targets is discerned a kind of new method.
Summary of the invention
The objective of the invention is to propose a kind of multi-class targets detection recognition methods of high-resolution remote sensing image; To target be artificial buildings; Based on shape facility target is described, added resolution and grain background classified information, carry out multistage multi-class targets detection identification in conjunction with sharing feature and feature selecting; The counting yield problem of remote sensing images multi-class targets identification can be solved, and higher recognition correct rate and recognition efficiency can be guaranteed.
Realize that the concrete technical scheme that the object of the invention adopted is following:
In order to improve counting yield; Earlier original remote sensing images are carried out resolution adjustment (amplitude of adjustment is confirmed according to other physics size of target class to be identified in the original remote sensing images in advance); Original remote sensing images are reduced to different resolution form the multi-resolution image group, again each image is handled respectively according to resolution order from low to high as follows:
(1) extracts area-of-interest
At first, extract the line segment in the image according to rim detection; Secondly, each pixel assignment in the image for the line segment length through this point, is obtained image line segment length distribution plan; Then, each pixel in the line segment length distribution plan is sued for peace in its a certain size neighborhood scope, obtain image line segment dense degree surface chart.At last, comprise peaked zone, be the area-of-interest of acquisition getting in the image line segment dense degree surface chart.
In this method, extract to the area-of-interest of remote sensing images target detection recognition application and to be different from traditional area-of-interest and to extract based on vision attention.Extracting area-of-interest needs to combine closely the target classification requirement of identification, finds the common feature between target, a small amount of false-alarm would rather occur, also can actual target locations be extracted.Through experimental observation, artificial buildings can show many line segment features, and the remote sensing images middle conductor is generally artificial buildings than the place of comparatively dense.
(2) priori aid identification
Land, waters classified information according to original image is carried out aid identification, and even detected area-of-interest is positioned at the land area, then gets into the land Target Recognition; If detected area-of-interest is positioned at the zone, waters, then get into the waters Target Recognition.
(3) feature representation
According to the selection of step (2), calculate the corresponding proper vector of selecting the area-of-interest in zone.At first, the line segment that rim detection in the step (1) is extracted is done pre-service: merge line segment, remove remaining weak point, the low contrast line segment.Then, pretreated line segment is constituted geometry according to spatial relationship, comprise parallel lines, U connects, and L connects, and remaining line segment is effective line segment.Geometry has constituted characteristic set with the attribute of effective line segment.At last, get suitable feature in the characteristic set and calculate its attribute as the corresponding proper vector of selecting the area-of-interest in zone.
Specific object is following: to effective line segment; Calculate the relative position relation attribute between each effective line segment; Comprise: the active line segment length; Effectively line segment is to the distance with reference to line segment, effectively line segment, with reference to line segment mid point line and with reference to the angle between the line segment, the angle of effective line segment, reference line elongated segment line; To geometry, calculate the absolute positional relation attribute at they and original image center, comprising: form the average length of each line segment of geometry, the distance that geometry and image center are nearest.
During to the line segment pre-service, length is called short line segment less than the straight line of certain value (like 15 pixels).Contrast (domain background pixel average and straight line pixel average ratio) is called the low contrast line segment less than the straight line of threshold value (as 10).
Shape facility has consequence in the remote sensing images feature extraction.Because artificial buildings can show tangible shape facility, as: linear edge, vertical crossing etc.The special target of each type all characterizes out the profile or the profile of certain pattern, and this character just makes shape facility can be used as a kind of validity feature of classification target in the different remote sensing images of difference.
(4) utilize the good sorter of training in advance to carry out Target Recognition
The good sorter of proper vector input training in advance with obtaining obtains recognition result and accomplishes the Target Recognition under the corresponding resolution.
According to the selection of step (2) and the feature representation of step (3), the sorter node that the proper vector input training in advance that obtains is good also judges whether to obtain recognition result.If do not obtain recognition result then continue to calculate individual features vector and import next sorter node; If obtain recognition result, promptly accomplish the Target Recognition under the corresponding resolution.Go out the land target that logical relation is set up to land Target Recognition procedure identification, go out the waters target that logical relation is set up to waters Target Recognition procedure identification.
After obtaining the Target Recognition result under the various image in different resolution, promptly accomplished the multi-class targets in the original high-resolution remote sensing image has been detected identification.
In flow process of the present invention, the area-of-interest that extracts some has improved the recognition efficiency of target detection identification greatly to carry out subsequent treatment; And use form feature model to describe different classes of artificial building target, improved the discrimination of remote sensing images target detection identification; Overall flow uses multistage multiclass decision tree to detect recognition structure, has improved remote sensing images target detection identification efficiency.
Description of drawings
Fig. 1: schematic flow sheet of the present invention
Fig. 2: 10 meters resolution identification process figure
Fig. 3: 5 meters resolution identification process figure
Fig. 4: parallel lines, L connect, the U syndeton is extracted synoptic diagram
Fig. 5: relative line segment attribute synoptic diagram
Fig. 6: absolute structure attribute synoptic diagram
Embodiment
Below in conjunction with accompanying drawing and specific embodiment the present invention is described further.
Consider six kinds of target classifications in the present embodiment altogether: airport, harbour, bridge, dam, highway hinge, railway terminal.The original image that uses is 1 meter high-resolution remote sensing image.According to the actual physical size of target, with the airport, the harbour, large bridge, large-scale dam are divided into one group, when image is 10 meters resolution, discern; With small bridge, small-sized dam, the highway hinge, railway terminal is divided into one group, when image is 5 meters resolution, discerns.10 meters with 5 meters identification process such as figure two, shown in the figure three.
In the present embodiment, at first original remote sensing images resolution is adjusted into 10 meters, after 10 meters detection identification process were accomplished, again remote sensing images being adjusted resolution was 5 meters, continued to detect the identification respective objects.
Concrete steps are following:
(1) extracts area-of-interest
A) at first, according to the line segment in the image after the rim detection extraction adjustment resolution; Secondly, each pixel assignment in the image for the line segment length through this point, is obtained image line segment length distribution plan; Then; To each pixel in the image line segment length distribution plan, getting with it is the center, suitably the square window (preferred size is the 150*150 pixel) of size; Line segment length numerical value in this window is added up, obtain characterizing the image surface chart of line dense degree thus.
B) be that the area-of-interest frame that initial size is the 250*250 pixel is got at the center with the peak value in the curved surface, to left and right sides four direction expansion up and down, largest extension is to the 500*500 pixel.Getting 10 meters step-lengths is 25 pixels, 5 meters step-lengths be 50 pixels as the candidate extended area, calculate the line length average of candidate extended area line segment through point; If greater than 50% of corresponding peak value; Then adding this extended area is the area-of-interest frame, by that analogy, obtains final area-of-interest.
Preferably, in order to obtain higher object recognition rate,,, can extract five area-of-interests with five peak regions of maximum according to step (b) to each width of cloth test pattern (during 5 meters resolution is the 4000*4000 size).This is because the remote sensing images background is very complicated, often has road, the square field, and background interference factors such as house influence maximal value.
(2) priori aid identification
According to land, the waters classified information of original image, detected area-of-interest is positioned at the land area, then gets into the airport, highway hinge and railway terminal identification; Detected area-of-interest is positioned at the zone, waters, then gets into the harbour, large bridge, large-scale dam, small bridge and the identification of small-sized dam.
(3) feature representation
1) target classification signature analysis
A) evaluating objects style characteristic
The airport: be in no aqua region, the parallel lines characteristic is very obvious.
The harbour: background has the waters, comprises many small-sized vertically, parallel, and more several font structures are arranged.
Bridge: background has the waters, removes assorted line in conjunction with the waters, comprises tangible vertical stratification.
Dam: background has the waters, removes assorted line in conjunction with the waters, comprises tangible vertical stratification, and the waters width on water center line both sides is different.
The highway hinge: be in no aqua region, straight line is mainly cross-shaped, disperses as radiation all around.
Railway terminal: be in no aqua region, straight line mainly is parallel shape, the very how unidirectional little broken line in target location.
B) construction feature group
According to other physical characteristics of target class and experimental data, use three stack features in the identification process:
Airport: parallel lines
The harbour, bridge, dam: active line, L connects, and U connects
The highway hinge, railway terminal: active line
2) extract line segment feature
According to image gradient information extraction image middle conductor, as: road edge, building edge, edge, harbour etc.Detailed process is following:
A) computed image I (x, gradient image G y) (x, y) amplitude size U (x, y) and direction θ (x, y), with θ (x, y) increase by 22.5 ° produce θ 1 (x, y).
B) (x, y) (x y) is divided into eight interval P (i) respectively according to direction, P1 (i) with θ 1 with θ.(x, the picture point that y) surpasses threshold value T according to whether the judgement of 8 connected domains draws connected set of points, is divided to zones of different R (i) respectively with connected set of points, R1 (i) as candidate point again with the U in each interval.
C) difference computation interval P (i), the region R (i) among the P1 (i), the minimum boundary rectangle of R1 (i); Obtain each boundary rectangle length l en, give the picture point of regional process separately with len, the picture point assignment that does not have process is 0; Generate two width of cloth zone boundary rectangle staple diagram L (x thus; Y), and L1 (x, y).
D) relatively the corresponding L of original image every bit (x, y), L1 (x, y) numerical values recited; If L (x, y)＞L1 (x, y), then corresponding region P (i) counting adds one; Otherwise corresponding region P1 (i) counting adds one, each region R (i), R1 (i) and ratio Ratio its area size (i), Ratio1 (i).
E) as if Ratio (i)＞50% or Ratio1 (i)＞50%, then corresponding zone greater than 50% is effective straight support zone, extracts original image I (x thus; Y) line segment in; And obtain the line segment attribute: length, gradient mean value, the contrast of background in straight line, the straight support territory.
The method that the image middle conductor is extracted in this area is a lot, and preferably above in this instance is the line segments extraction method on basis with the compute gradient.
3) extract the geometry characteristic
Original line segments is done pre-service: merge line segment, remove remaining length less than 15, contrast is lower than 10 line segment.With the geometric features formation geometry of pretreated line segment according to target.Geometry comprises: parallel lines, and L connects, and U connects.Remaining be effective line segment.Fig. 4 is that parallel lines, L connect, U connects synoptic diagram.Judgment rule is following: judge neighbouring relations according to distance between the line segment, for adjacent sets of line segments, judge angle and distance between them, then can get parallel lines, L connects, the U syndeton.
4) produce proper vector
To each how much parts, according to their geometry and spatial relationship computation attribute.Attribute comprises: relative attribute, absolute attribute.Concrete grammar is following:
A) effective relative attribute of line segment: { l, d, Φ, θ }
Computation rule is as shown in Figure 5: l representes the active line segment length, and d representes that line segment arrives the distance with reference to line segment m, and Φ representes the angle between l and m mid point line and the m, and θ representes the angle of two line segment extended lines.
B) parallel lines, L connects, and U connects absolute attribute: { l, d}
Computation rule is as shown in Figure 6: l representes prototype structure composition line segment average length, and d representes prototype structure and the nearest distance of image center.
C) set feature space.If how much parts attribute kinds are m, then they are mapped to a m dimension histogram.To each dimension regulation quantized interval of attribute histogram.The total dimension of proper vector that generates is:
In the formula, P representes geometry, and A representes the attribute of geometry, n (I
Ij) expression quantized interval quantity.The attribute of geometry is expressed the one dimension in the different coding form character pair vector by the coding form after quantizing.Its eigenwert is the quantity that satisfies this attribute coding of all geometries in the image.
(4) utilize the good sorter of training in advance to carry out Target Recognition
Before detecting identification, need the good sorter of training in advance.The method of training classifier is: to every type of target that needs identification, under its corresponding resolution (10 meters or 5 meters), gather about 100 width of cloth goal-orientations, comprise the training sample of a small amount of background, be cut to identical size.The proper vector of calculation training sample at corresponding recognition node, is set the proper parameter training classifier to the svm classifier device.
Calculate proper vector according to step (3), obtain recognition result in the input svm classifier device.
Claims (2)
1. a high-resolution remote sensing image multi-class targets detects recognition methods; At first original remote sensing images are carried out the resolution adjustment; Original remote sensing images are reduced to a plurality of different resolution respectively; Form the image that several have different resolution, again each image in different resolution handled respectively according to resolution order from low to high as follows:
(1) extracts area-of-interest
At first, extract the line segment in the image according to rim detection; Secondly, each pixel assignment in the image for the line segment length through this point, is obtained image line segment length distribution plan; Then, each pixel in the said line segment length distribution plan is sued for peace in its a certain size neighborhood scope, obtain image line segment dense degree surface chart; At last, comprise peaked zone, be the area-of-interest of acquisition getting in the said image line segment dense degree surface chart;
(2) priori aid identification
Land, waters classified information according to original image is carried out aid identification, determines this area-of-interest affiliated area type, confirms that promptly area-of-interest belongs to land area or zone, waters;
(3) feature extraction
Area type according to step (2) is confirmed calculates the proper vector that obtains area-of-interest;
(4) utilize the good sorter of training in advance to carry out Target Recognition
The good sorter of proper vector input training in advance with obtaining obtains recognition result and accomplishes the Target Recognition under the corresponding resolution;
After obtaining the Target Recognition result under the various image in different resolution, promptly accomplish detection identification to the multi-class targets in the original high resolution remote sensing images;
Wherein, the concrete computation process of proper vector is in the described step (3):
At first, the line segment that rim detection in the step (1) is extracted is done pre-service: merge line segment, remove the remaining weak point and the line segment of low contrast, wherein length is called short line segment less than the straight line of certain value, and contrast is called the low contrast line segment less than the straight line of threshold value; Then, pretreated line segment is constituted geometry according to spatial relationship, comprise parallel lines, U connects, and L connects, and the remaining line segment that does not constitute geometry is effective line segment, and said geometry has constituted characteristic set with the attribute of effective line segment; At last, from said characteristic set, choose the proper vector of its attribute of certain feature calculation as area-of-interest.
2. method according to claim 1 is characterized in that, said attribute is specially:
To effective line segment; Said attribute is the relative position relation attribute between each effective line segment, comprising: the active line segment length, effectively line segment is to reference to the distance of line segment, effective line segment and with reference to the line of line segment mid point and angle with reference to the angle between the line segment and effective line segment and reference line elongated segment line;
To geometry, said attribute is the absolute positional relation attribute at geometry and original image center, comprising: form the average length of each line segment of geometry, and geometry and the nearest distance of image center.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010105623195A CN102043958B (en) | 2010-11-26 | 2010-11-26 | High-definition remote sensing image multi-class target detection and identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010105623195A CN102043958B (en) | 2010-11-26 | 2010-11-26 | High-definition remote sensing image multi-class target detection and identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102043958A CN102043958A (en) | 2011-05-04 |
| CN102043958B true CN102043958B (en) | 2012-11-21 |
Family
ID=43910084
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010105623195A Expired - Fee Related CN102043958B (en) | 2010-11-26 | 2010-11-26 | High-definition remote sensing image multi-class target detection and identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102043958B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104346775B (en) * | 2013-08-02 | 2016-09-21 | 南京理工大学 | A kind of method extracting image local grain roughness |
| CN104346800B (en) * | 2013-08-02 | 2016-12-28 | 南京理工大学 | A kind of twilight image object detection method based on texture significance |
| CN103761522B (en) * | 2013-12-27 | 2017-02-08 | 河海大学 | SAR image river channel extracting method based on minimum circumscribed rectangle window river channel segmentation model |
| CN104281854B (en) * | 2014-09-28 | 2018-09-25 | 中国人民解放军海军航空工程学院 | High-resolution polarization SAR Ship Targets and jamming target discrimination method |
| CN104794495B (en) * | 2015-05-04 | 2018-02-06 | 福建师范大学 | A kind of large format remote sensing image territorial classification method based on straight line statistical nature |
| CN105740871A (en) * | 2016-02-01 | 2016-07-06 | 福建师范大学 | Method for detecting polygonal ground object by using line section space relationship |
| CN106557740B (en) * | 2016-10-19 | 2019-05-03 | 华中科技大学 | The recognition methods of oil depot target in a kind of remote sensing images |
| CN106599886B (en) * | 2016-11-07 | 2020-07-28 | 纵目科技（上海）股份有限公司 | Mobile device, storage number identification method and system, and character identification method |
| CN108287845A (en) * | 2017-01-09 | 2018-07-17 | 北京四维图新科技股份有限公司 | A kind of Automatic extraction method for road information and device and hybrid navigation system |
| CN107664778B (en) * | 2017-08-24 | 2019-11-08 | 中国石油天然气股份有限公司 | The determination method and apparatus of earthquake data quality |
| CN107944351B (en) * | 2017-11-07 | 2020-08-04 | 深圳市易成自动驾驶技术有限公司 | Image recognition method, image recognition device and computer-readable storage medium |
| CN109481248A (en) * | 2018-12-26 | 2019-03-19 | 浙江师范大学 | A kind of smart guide glasses |
| CN111079596A (en) * | 2019-12-05 | 2020-04-28 | 国家海洋环境监测中心 | System and method for identifying typical marine artificial target of high-resolution remote sensing image |
| CN111932451B (en) * | 2020-06-17 | 2023-11-10 | 珠海格力电器股份有限公司 | Method and device for evaluating repositioning effect, electronic equipment and storage medium |
| CN114581781B (en) * | 2022-05-05 | 2022-08-09 | 之江实验室 | Target detection method and device for high-resolution remote sensing image |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101470806A (en) * | 2007-12-27 | 2009-07-01 | 东软集团股份有限公司 | Vehicle lamp detection method and apparatus, interested region splitting method and apparatus |
| CN101814144A (en) * | 2010-03-25 | 2010-08-25 | 西安电子科技大学 | Water-free bridge target identification method in remote sensing image |
-
2010
- 2010-11-26 CN CN2010105623195A patent/CN102043958B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101470806A (en) * | 2007-12-27 | 2009-07-01 | 东软集团股份有限公司 | Vehicle lamp detection method and apparatus, interested region splitting method and apparatus |
| CN101814144A (en) * | 2010-03-25 | 2010-08-25 | 西安电子科技大学 | Water-free bridge target identification method in remote sensing image |
Non-Patent Citations (5)
| Title |
|---|
| Acquisition,Tracking,Pointing,and Laser Systems Technologies XXIV * |
| and Optical Pattern Recognition XXI》.2010, * |
| yuehuan Wang,et.al.Multi-class target recognition based on adaptive feature selection.《 Conference：Automatic Target Recognition XX * |
| 左峥嵘等.基于兴趣区检测的地面目标识别方法研究.《红外技术》.2004,第26卷(第3期),第33-34页. * |
| 王亮.形状提取与识别.《CNKI优秀硕士学位论文全文库》.2003,第11-12页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102043958A (en) | 2011-05-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102043958B (en) | High-definition remote sensing image multi-class target detection and identification method | |
| CN108510467B (en) | SAR image target identification method based on depth deformable convolution neural network | |
| CN102622607B (en) | Remote sensing image classification method based on multi-feature fusion | |
| CN107346420A (en) | Text detection localization method under a kind of natural scene based on deep learning | |
| CN102542295B (en) | Method for detecting landslip from remotely sensed image by adopting image classification technology | |
| Bremer et al. | Eigenvalue and graph-based object extraction from mobile laser scanning point clouds | |
| CN103839065A (en) | Extraction method for dynamic crowd gathering characteristics | |
| CN103116744A (en) | Fake fingerprint detection method based on markov random field (MRF) and support vector machine-k nearest neighbor (SVM-KNN) classification | |
| CN103578110A (en) | Multi-band high-resolution remote sensing image segmentation method based on gray scale co-occurrence matrix | |
| CN102867195B (en) | Method for detecting and identifying a plurality of types of objects in remote sensing image | |
| Yuan et al. | Learning to count buildings in diverse aerial scenes | |
| CN105512622B (en) | A kind of visible remote sensing image sea land dividing method based on figure segmentation and supervised learning | |
| CN102842044A (en) | Method for detecting variation of remote-sensing image of high-resolution visible light | |
| CN104282026A (en) | Distribution uniformity assessment method based on watershed algorithm and minimum spanning tree | |
| CN106845458B (en) | Rapid traffic sign detection method based on nuclear overrun learning machine | |
| CN102129559B (en) | SAR (Synthetic Aperture Radar) image object detection method based on Primal Sketch algorithm | |
| Liang et al. | An extraction and classification algorithm for concrete cracks based on machine vision | |
| Kumar et al. | An automated road roughness detection from mobile laser scanning data | |
| CN101980295A (en) | Method for detecting mainstream line of Yellow River based on skewness analysis | |
| CN104331708B (en) | A kind of zebra crossing automatic detection analysis method and system | |
| CN103065296B (en) | High-resolution remote sensing image residential area extraction method based on edge feature | |
| CN104794723A (en) | Remote-sensing image building location detection method based on probability | |
| CN102013097A (en) | Yellow River main humping line detection method based on spectrum similarity and space continuity | |
| CN108447045B (en) | SAR remote sensing image water area detection method based on SAT integral graph | |
| Sirmacek et al. | Road detection from remotely sensed images using color features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20121121 Termination date: 20151126 |