CN103150567B - The remote sensing image cloud method of discrimination of integration mode identification and Context Knowledge - Google Patents
The remote sensing image cloud method of discrimination of integration mode identification and Context Knowledge Download PDFInfo
- Publication number
- CN103150567B CN103150567B CN201310043706.1A CN201310043706A CN103150567B CN 103150567 B CN103150567 B CN 103150567B CN 201310043706 A CN201310043706 A CN 201310043706A CN 103150567 B CN103150567 B CN 103150567B
- Authority
- CN
- China
- Prior art keywords
- tile
- block
- decision value
- piecemeal
- cloud
- 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
Links
Abstract
The invention discloses the remote sensing image cloud method of discrimination of a kind of integration mode identification and Context Knowledge, the drawback only utilizing the information of some piecemeals itself to carry out pattern-recognition to bring can be solved in prior art.First, adopt the square of L × L in remote sensing image, carry out the stepping of level and vertical direction, obtain piecemeal and be called Tile, each Tile is made up of 4 Block.Pattern-recognition is carried out to each Tile piecemeal, obtains Tile index matrix; Determine the decision value of Block according to Tile index matrix and finely tune; Utilize Context Knowledge, the decision value of Block is revised; The cloud position finally utilizing revised Block decision value to mark, carries out cloud rejecting to remote sensing image.
Description
Technical field
The present invention relates to a kind of remote sensing image cloud method of discrimination, particularly a kind of remote sensing image cloud method of discrimination integrating statistical-simulation spectrometry and Context Knowledge, belongs to image identification technical field.
Background technology
For remote sensing image, we interested are mostly terrain object, but be there is blocking of cloud under many circumstances, the part of such cloud is without any terrestrial information, but occupies storage space processing poweies a large amount of in disposal system and transmission bandwidth.Artificial screening and these data of rejecting require a great deal of time and energy usually, if these " inactive area " can by finding out someway, just this part data can be removed, and then save the workload of data space and data processing personnel, especially to the system on satellite, downlink bandwidth can also be saved to transmit prior data.
Therefore, a kind of effective automatic cloud is sentenced technology and is just become and be necessary very much.The mainly Land use models knowledge method for distinguishing that current cloud sentences method for distinguishing processes, and generally includes the sampling piecemeal to former figure, feature extraction, classification judgement and territory, cloud sector and rejects.These methods only utilize the information of some piecemeals itself usually, are generally used for processing in piecemeal effect when only having single decipher object better.But when processing the piecemeal in cloud and place, atural object boundary, because it contains two or more decipher objects, make the judgement deleterious of pattern-recognition, even, it is invalid to become.Such as, when a piecemeal medium cloud with atural object distribution is concentrated and content is respectively half (upper half piecemeal is cloud and lower half piecemeal is atural object), this piecemeal is inherently neither cloud neither land, so what meaning is the result of pattern-recognition just do not have.
Summary of the invention
In view of this, the invention provides the remote sensing image cloud method of discrimination of a kind of integration mode identification and Context Knowledge, the drawback only utilizing the information of some piecemeals itself to carry out pattern-recognition to bring can be solved in prior art.
In order to solve the problems of the technologies described above, the present invention is achieved in that
A remote sensing image cloud method of discrimination for integration mode identification and Context Knowledge, comprising:
The first step, samples and piecemeal to former figure: read in a width remote sensing image and to go forward side by side line sampling and piecemeal; Described piecemeal adopts overlapping partitioned mode, namely adopt the square of L × L in remote sensing image, carry out the stepping of level and vertical direction, stepping-in amount is L/2, obtain multiple piecemeal, be called Tile, every four adjacent overlapped Tile piecemeal can regard nine nonoverlapping sizes as is up and down that the little piecemeal of L/2 × L/2 is formed, and this little piecemeal is called Block;
Second step, pattern-recognition is carried out to each Tile piecemeal: Land use models identification is adjudicated each Tile piecemeal successively, and by decision value stored in Tile index matrix, decision value comprises the degree of confidence belonging to cloud, the degree of confidence belonging to atural object and uncertain, and be benchmark B by uncertain for decision value assignment here, confidence interval is D, the decision value of cloud should be B+mD, the decision value of corresponding atural object is B-nD, and wherein m and n gets positive integer, and the value of higher m or n of degree of confidence of cloud or atural object is larger;
3rd step, determine the threshold range of different object feature; Described different object comprises the cloud of different degree of confidence, the atural object of uncertain and different degree of confidence;
4th step, determine the decision value of Block according to Tile index matrix and finely tune:
Step S41) for each Tile piecemeal, calculate the gray average of 4 Block of this Tile piecemeal of composition respectively, threshold judgement method is adopted to carry out cloud judgement to Block based on the threshold range obtained in the 3rd step, decision value adopts form used in second step, namely be B time uncertain, B+mD when being judged to cloud, B-nD when being judged to atural object;
Step S42) utilize the decision value of Tile piecemeal and inner 4 Block thereof to calculate the fine setting factor Factor of Tile:
Wherein, Sum refers to be judged to cloud or atural object Block decision value sum in Tile, and n refers to be judged to uncertain Block number in Tile, and Predict refers to the decision value of Tile;
Step S43) for judgement in Tile for uncertain Block, using step S42) in the fine setting factor that obtains be added temporary decision values as this Block with the decision value of Tile;
5th step, each Block belongs to multiple Tile piecemeal, and therefore each Block has multiple temporary decision values, comprehensively determines the decision value of Block according to the temporary decision values of same Block, Block decision value composition Block index matrix; Utilize Context Knowledge, the decision value in Block index matrix is revised;
6th step, the cloud position utilizing revised Block decision value to mark, carries out cloud rejecting to remote sensing image.
Preferably, described 3rd step is: the threshold range utilizing the pattern recognition result determination different object feature of Tile; Be specially:
Step S31) utilize Tile index matrix, find the Tile that wherein decision value is identical wherein, and add up the eigenwert of different object in these Tile;
Step S32) feature of every type object all added up together, using the maximal value of these features and minimum value as its scope, the namely upper and lower bound of threshold value;
Step S33) due to the cloud and the atural object that have all degree of confidence in each scene can not be ensured, so the type lacked for those, the empirical thresholds that its threshold value adopts experimental summary to go out or the thresholding that training obtains.
Beneficial effect:
(1) the present invention adopts overlap partition, after pattern-recognition is carried out to Tile, recognition result is utilized to determine the decision value of Block, then the context relation between overlap partition is utilized to carry out auxiliary judgement, thus the false-alarm of the overwhelming majority can be weeded out, performance is greatly improved, the drawback only utilizing the information of some piecemeals itself to carry out pattern-recognition to bring can be solved in prior art.
(2) the present invention proposes to finely tune the factor, and its design concept is that the court verdict of Tile can be equivalent to the average of four Block court verdicts in this Tile.Utilizing this fine setting factor a part can be adjudicated inaccurate Block utilizes the logical relation between Tile and Block to revise its decision value, thus achieves fine setting, improves judgement effect.
(3) threshold value of the threshold classification that the present invention is used is the different characteristic of directly adding up the equal corresponding object of decision value in current scene, and the scope finding out these features is as threshold value.Because these threshold values are corresponding current scene gained, therefore compare the empirical thresholds that sums up through great many of experiments and by training the thresholding obtained to have stronger specific aim and adaptability.And the object classification added up is by second step given by pattern-recognition, therefore adjudicate the result of effect also closely pattern-recognition.
Accompanying drawing explanation
Fig. 1 is the process flow diagram that the present invention integrates the remote sensing image cloud method of discrimination of statistical-simulation spectrometry and Context Knowledge.
Fig. 2 is the method schematic diagram among the first step, whole figure being carried out to piecemeal in Fig. 1.
Embodiment
The method that solution background technology asks a question is exactly overlapping piecemeal, and the piecemeal of such diverse location increases, and same cloudlet is None-identified in a piecemeal, perhaps just can accurately identify in other piecemeals of its overlap.So, in turn introduce another problem: when the court verdict of the different piecemeals at lap place is not identical, which should be chosen as its recognition result.At this moment the context relation will used on a whole figure between overlap partition carrys out auxiliary judgement, thus the false-alarm of the overwhelming majority can be weeded out, and is greatly improved to performance.The invention provides the remote sensing image cloud method of discrimination of a kind of integration statistical-simulation spectrometry based on overlap partition and Context Knowledge for this reason.
To develop simultaneously embodiment below in conjunction with accompanying drawing, describe the present invention.
Fig. 1 shows the remote sensing image cloud method of discrimination process flow diagram of integration statistical-simulation spectrometry provided by the invention and Context Knowledge.As shown in Figure 1, the remote sensing image cloud method of discrimination of integration statistical-simulation spectrometry provided by the invention and Context Knowledge comprises the following step carried out in order:
The first step, former figure to be sampled and piecemeal:
After reading in a width remote sensing image, in order to the resolution requirement of adaptive algorithm process, first, need to sample to former figure, the resolution that the ratio that sampling adopts requires according to former figure resolution and algorithm is determined.Then, carry out piecemeal to whole figure after sampling, specifically adopting has overlapping partitioned mode.
As shown in Figure 2, in the first step stage, there is overlapping partitioned mode both can situation that effectively problem-solving pattern identification is fuzzy, the correlativity between piecemeal can be increased again, this correlativity is also contextual one side source, specific practice is: first, choosing according to whole figure the block size size will carrying out pattern recognition process is L × L, and making it be Tile, is L/2 with seasonal level and vertical stepping.Then, by this stepping-in amount, piecemeal is carried out to each row from left to right, finally, moved the piecemeal to full figure according to this stepping more from top to bottom with the partitioned mode of this row.So can draw, adjacent four overlapped Tile piecemeals (distribution of similar field word), also can regard the little piecemeal formation that nine nonoverlapping sizes are L/2 × L/2 as, make this little piecemeal be Block.Block is overlapping minimum unit, and each Tile can be divided into 4 Block, most Block all comprise by around 4 different Ti le.
Second step, pattern-recognition is carried out to each Tile piecemeal:
According to the piecemeal rule of previous step, Land use models identification is adjudicated each piecemeal Tile successively, and wherein the general characteristics of needs of pattern-recognition extracts, learns and identify this three step.And form a matrix after being arranged according to its spatial relation in whole figure by the decision value of each Tile, be referred to as the decision index matrix of Tile, be called for short Tile index matrix.The corresponding Tile piecemeal of element in Tile index matrix.The decision value of Tile comprises the degree of confidence belonging to cloud, the degree of confidence belonging to atural object and uncertain, here be benchmark B by uncertain for decision value assignment, confidence interval is D, the degree of confidence of cloud often increases once corresponding decision value and just on the basis of B, adds a D, and the decision value that the degree of confidence of atural object often increases once correspondence just deducts a D on the basis of B.Therefore the decision value of cloud should be B+mD, the decision value of corresponding atural object is B-nD, and wherein m and n gets positive integer.
3rd step, utilize the threshold range of the pattern recognition result determination different object feature of Tile.
Described different object comprises the cloud of different degree of confidence, the atural object of uncertain and different degree of confidence.
When utilizing threshold method to classify, wherein each threshold value is not through the empirical value that great many of experiments sums up, and obtains by training tagged object.Although the latter is more more objective relative to the former, what mark after all is limited to liking, and also can not have good adaptability to the scene residing for the object when pre-treatment.In the present embodiment, the different characteristic of the equal corresponding object of direct statistical decision value, and find the scope of these features as threshold value.Because these threshold values are corresponding current scene gained, therefore there is very strong specific aim and adaptability.And the object classification added up is by second step given by pattern-recognition, therefore adjudicate the result of effect also closely pattern-recognition.
Describedly determine that the threshold range of different object feature comprises the following step carried out in order:
Step S31) add up the feature of different object respectively: the index matrix utilizing Tile, find the Tile that wherein decision value is identical wherein, and add up the different characteristic value of these Tile.The eigenwert of Tile can be here extract, the coverage rate of the brightness and cloud of such as extracting Tile as feature, also can to utilize in second step the feature extracted.
Step S32) determine the scope of each feature: through step S31, the different characteristic of the object of every type all should be added up together, here can simply using the maximal value of these features and minimum value as its scope, the namely upper and lower bound of threshold value.
Step S33) supplement the threshold value lacked: owing to can not ensure to have in each scene cloud and the atural object of all degree of confidence, so for the threshold value of those types lacked, as a supplement, this process leaves it at that the thresholding that the empirical thresholds that experimental summary can be utilized to go out or training obtain.
4th step, determine the decision value of Block according to Tile index matrix and finely tune:
Because atural object is of a great variety, often kind of landforms have different forms, and cloud is different, and this just causes judgement is have certain error, therefore carries out finely tuning before context correction being extremely necessary.
Described method of finely tuning pattern recognition result comprises the following step carried out in order:
Step S41) utilize gray scale to carry out fine setting pre-service: according to the first step, each piecemeal Tile is made up of 4 overlapping minimum unit Block.Because gray average roughly can reflect the situation of piecemeal, here for each Tile piecemeal, calculate the gray average of 4 Block of this Tile piecemeal of composition respectively, threshold judgement method is adopted to carry out simple cloud judgement to Block based on the threshold value obtained in the 3rd step, decision value adopts form used in second step, is B time namely uncertain, B+mD when being judged to cloud, B-nD when being judged to atural object, and the value of higher m or n of the degree of confidence of cloud or atural object is larger.Owing to can not ensure that the threshold interval of adjacent confidence categories in the 3rd step can connect together (minimum value in larger interval is greater than the maximal value compared with minizone), its the possibility of result causes some Block not belong to any type above, can be uncertain by it is classified as this situation, assignment is B.
Step S42) calculate the fine setting factor of Tile: utilize the court verdict determined in Tile and inner 4 Block thereof to calculate the fine setting factor of Tile, formula is as follows:
Wherein, Sum refers to be judged to cloud or atural object Block decision value sum in Tile, and n refers to be judged to uncertain Block number in Tile, and Predict refers to the decision value of Tile, namely obtains the fine setting factor;
Principle used by the design fine setting factor is that the court verdict of Tile can be equivalent to the average of four Block court verdicts in this Tile.Because Tile and its 4 Block sums comprised inherently equivalent, and above-mentioned various different judging confidence value is just used to the analogue of expression and cloud, namely cloud in object proportion number, so Tile equals the mean value containing cloud ratio of Block containing the ratio of cloud.
When Block is wherein uncertain time, just it can be regarded as unknown number Predict-Factor, Factor is here the bias of Predict.Make Sum equal to be judged to cloud or atural object Block decision value sum in Tile, make Predict be the decision value of Tile, n refers to be judged to uncertain Block number in Tile, obtains according to relation above:
Solution take Factor as the equation of unknown quantity, just can obtain:
Known by upper surface analysis, utilize the fine setting factor part can be adjudicated inaccurate Block and utilize the logical relation between Tile and Block to revise its decision value, thus achieve fine setting, improve judgement effect.
Step S43) utilize the fine setting factor to finely tune: adjudicate as uncertain Block in Tile, the fine setting factor values obtained in previous step is added the temporary decision values as this Block with the decision value of Tile, this process leaves it at that.
5th step, context is utilized to carry out context correction to Block:
Each Block belongs to multiple Tile piecemeal, and therefore each Block has multiple temporary decision values, comprehensively determines the decision value of Block according to the temporary decision values of same Block, Block decision value composition Block index matrix.Utilize Context Knowledge, the decision value in Block index matrix is revised.
Wherein, comprehensively determine that the decision value of Block can adopt average method according to the temporary decision values of same Block.Utilize Context Knowledge to carry out correction to Block index matrix and can adopt existing any one context modification method, such as can utilize the method for (Markovrandomfield) markov random file: first, the object in some typical scenes is marked; Then, some cooccurrence relations of a tagged object and object adjacent around it in scene are utilized to set up probability model; Finally, this probability model just can have modified adjudicating uncertain object in the decision value index matrix of Block.
6th step, revised result is utilized to carry out cloud rejecting:
The cloud position utilizing revised Block decision value to mark, carries out cloud rejecting to remote sensing image, and namely all in the former figure of corresponding Block position pixel zero setting, all terminates to this whole process.
In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (1)
1. a remote sensing image cloud method of discrimination for integration mode identification and Context Knowledge, is characterized in that, comprising:
The first step, samples and piecemeal to former figure: read in a width remote sensing image and to go forward side by side line sampling and piecemeal; Described piecemeal adopts overlapping partitioned mode, namely adopt the square of L × L in remote sensing image, carry out the stepping of level and vertical direction, stepping-in amount is L/2, obtain multiple piecemeal, be called Tile, every four adjacent overlapped Tile piecemeal can regard nine nonoverlapping sizes as is up and down that the little piecemeal of L/2 × L/2 is formed, and this little piecemeal is called Block;
Second step, pattern-recognition is carried out to each Tile piecemeal: Land use models identification is adjudicated each Tile piecemeal successively, and by decision value stored in Tile index matrix, decision value comprises the degree of confidence belonging to cloud, the degree of confidence belonging to atural object and uncertain, and be benchmark B by uncertain for decision value assignment here, confidence interval is D, the decision value of cloud should be B+mD, the decision value of corresponding atural object is B-nD, and wherein m and n gets positive integer, and the value of higher m or n of degree of confidence of cloud or atural object is larger;
Wherein, form a matrix after being arranged according to its spatial relation in whole figure by the decision value of each Tile, be referred to as the decision index matrix of Tile, be called for short Tile index matrix;
3rd step, utilize the threshold range of the pattern recognition result determination different object feature of Tile; Described different object comprises the cloud of different degree of confidence, the atural object of uncertain and different degree of confidence;
Step S31) utilize Tile index matrix, find the Tile that wherein decision value is identical wherein, and add up the eigenwert of different object in these Tile;
Step S32) feature of every type object all added up together, using the maximal value of these features and minimum value as its scope, the namely upper and lower bound of threshold value;
Step S33) due to the cloud and the atural object that have all degree of confidence in each scene can not be ensured, so the type lacked for those, the empirical thresholds that its threshold value adopts experimental summary to go out or the thresholding that training obtains;
4th step, determine the decision value of Block according to Tile index matrix and finely tune:
Step S41) for each Tile piecemeal, calculate the gray average of 4 Block of this Tile piecemeal of composition respectively, threshold judgement method is adopted to carry out cloud judgement to Block based on the threshold range obtained in the 3rd step, decision value adopts form used in second step, namely be B time uncertain, B+mD when being judged to cloud, B-nD when being judged to atural object;
Step S42) utilize the decision value of Tile piecemeal and inner 4 Block thereof to calculate the fine setting factor Factor of Tile:
Wherein, Sum refers to be judged to cloud or atural object Block decision value sum in Tile, and n refers to be judged to uncertain Block number in Tile, and Predict refers to the decision value of Tile;
Step S43) for judgement in Tile for uncertain Block, using step S42) in the fine setting factor that obtains be added temporary decision values as this Block with the decision value of Tile;
5th step, each Block belongs to multiple Tile piecemeal, and therefore each Block has multiple temporary decision values, comprehensively determines the decision value of Block according to the temporary decision values of same Block, Block decision value composition Block index matrix; Utilize Context Knowledge, the decision value in Block index matrix is revised;
6th step, the cloud position utilizing revised Block decision value to mark, carries out cloud rejecting to remote sensing image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310043706.1A CN103150567B (en) | 2013-02-04 | 2013-02-04 | The remote sensing image cloud method of discrimination of integration mode identification and Context Knowledge |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310043706.1A CN103150567B (en) | 2013-02-04 | 2013-02-04 | The remote sensing image cloud method of discrimination of integration mode identification and Context Knowledge |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103150567A CN103150567A (en) | 2013-06-12 |
CN103150567B true CN103150567B (en) | 2016-01-20 |
Family
ID=48548630
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310043706.1A Active CN103150567B (en) | 2013-02-04 | 2013-02-04 | The remote sensing image cloud method of discrimination of integration mode identification and Context Knowledge |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103150567B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101694309B1 (en) * | 2013-10-02 | 2017-01-09 | 한국전자통신연구원 | Apparatus and method for recognizing object based sub region |
CN105913033B (en) * | 2016-04-15 | 2019-05-14 | 北京航空航天大学 | Multi layer cloud and single layer cloud-type compressive classification and knowledge method for distinguishing in remote sensing images |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101246545A (en) * | 2008-02-22 | 2008-08-20 | 华南师范大学 | Possion method for removing cloud from optical remote sensing image |
CN102663348A (en) * | 2012-03-21 | 2012-09-12 | 中国人民解放军国防科学技术大学 | Marine ship detection method in optical remote sensing image |
CN102750701A (en) * | 2012-06-15 | 2012-10-24 | 西安电子科技大学 | Method for detecting spissatus and spissatus shadow based on Landsat thematic mapper (TM) images and Landsat enhanced thematic mapper (ETM) images |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2008051207A2 (en) * | 2005-10-21 | 2008-05-02 | Carnegie Institution Of Washington | Remote sensing analysis of forest disturbances |
-
2013
- 2013-02-04 CN CN201310043706.1A patent/CN103150567B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101246545A (en) * | 2008-02-22 | 2008-08-20 | 华南师范大学 | Possion method for removing cloud from optical remote sensing image |
CN102663348A (en) * | 2012-03-21 | 2012-09-12 | 中国人民解放军国防科学技术大学 | Marine ship detection method in optical remote sensing image |
CN102750701A (en) * | 2012-06-15 | 2012-10-24 | 西安电子科技大学 | Method for detecting spissatus and spissatus shadow based on Landsat thematic mapper (TM) images and Landsat enhanced thematic mapper (ETM) images |
Non-Patent Citations (1)
Title |
---|
遥感图像的实时云判技术;闫宇松 等;《北京理工大学学报》;20100715;第30卷(第7期);817-821 * |
Also Published As
Publication number | Publication date |
---|---|
CN103150567A (en) | 2013-06-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109271991B (en) | License plate detection method based on deep learning | |
Erikson | Species classification of individually segmented tree crowns in high-resolution aerial images using radiometric and morphologic image measures | |
CN103049763B (en) | Context-constraint-based target identification method | |
CN103258213B (en) | A kind of for the dynamic vehicle model recognizing method in intelligent transportation system | |
CN111444821A (en) | Automatic identification method for urban road signs | |
CN103366367B (en) | Based on the FCM gray-scale image segmentation method of pixel count cluster | |
CN106022232A (en) | License plate detection method based on deep learning | |
CN110838126B (en) | Cell image segmentation method, cell image segmentation device, computer equipment and storage medium | |
CN103034863A (en) | Remote-sensing image road acquisition method combined with kernel Fisher and multi-scale extraction | |
CN115797798B (en) | Ecological restoration effect evaluation method based on abandoned mine remote sensing image | |
CN103366365B (en) | SAR image change detection method based on artificial immunity multi-object clustering | |
CN102236788A (en) | Kilowatt-hour meter image automatic identification method | |
CN105930794A (en) | Indoor scene identification method based on cloud computing | |
Xiang et al. | Lightweight fully convolutional network for license plate detection | |
CN104680193A (en) | Online target classification method and system based on fast similarity network fusion algorithm | |
Yang et al. | Toward country scale building detection with convolutional neural network using aerial images | |
CN115861738A (en) | Category semantic information guided remote sensing target detection active sampling method | |
CN102938069A (en) | Pure and mixed pixel automatic classification method based on information entropy | |
Han et al. | Vision-based crack detection of asphalt pavement using deep convolutional neural network | |
CN103150567B (en) | The remote sensing image cloud method of discrimination of integration mode identification and Context Knowledge | |
CN113177554B (en) | Thyroid nodule identification and segmentation method, system, storage medium and equipment | |
CN102005056A (en) | Target tracking method based on robust PCA (principal component analysis) subspace | |
Zhou et al. | SWDet: Anchor-based object detector for solid waste detection in aerial images | |
CN103093241B (en) | Based on the remote sensing image nonuniformity cloud layer method of discrimination of homogeneity process | |
Deepan et al. | Road recognition from remote sensing imagery using machine learning |
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 | ||
CB03 | Change of inventor or designer information |
Inventor after: Long Teng Inventor after: Yang Zhu Inventor after: Liu Feng Inventor after: Pang Fengjian Inventor after: Bi Fukun Inventor after: Chen Liang Inventor before: Long Teng Inventor before: Liu Feng Inventor before: Pang Fengjian Inventor before: Bi Fukun Inventor before: Chen Liang |
|
CB03 | Change of inventor or designer information |