CN106056609A - Method based on DBNMI model for realizing automatic annotation of remote sensing image - Google Patents

Method based on DBNMI model for realizing automatic annotation of remote sensing image Download PDF

Info

Publication number
CN106056609A
CN106056609A CN201610383452.1A CN201610383452A CN106056609A CN 106056609 A CN106056609 A CN 106056609A CN 201610383452 A CN201610383452 A CN 201610383452A CN 106056609 A CN106056609 A CN 106056609A
Authority
CN
China
Prior art keywords
remote sensing
sensing image
dbnmi
semantic
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201610383452.1A
Other languages
Chinese (zh)
Other versions
CN106056609B (en
Inventor
黄冬梅
许琼琼
杜艳玲
贺琪
朱建钢
田为民
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.)
Shanghai Maritime University
Shanghai Ocean University
Original Assignee
Shanghai Maritime 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 Shanghai Maritime University filed Critical Shanghai Maritime University
Priority to CN201610383452.1A priority Critical patent/CN106056609B/en
Publication of CN106056609A publication Critical patent/CN106056609A/en
Application granted granted Critical
Publication of CN106056609B publication Critical patent/CN106056609B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20004—Adaptive image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation

Abstract

The invention relates to a method based on a DBNMI model for realizing automatic annotation of a remote sensing image. The method comprises the steps of: carrying out self-adaptive segmentation on an initial input remote sensing image, and dividing the remote sensing image into a background region and an object region in a coarse granularity manner; representing the divided object region of the remote sensing image by utilizing a multi-instance multi-tag technology; carrying out automatic modeling based on the DBNMI model for the relation between low-level visual characteristics and high-level semantic concepts of the remote sensing image; and calculating co-occurrence and opposite semantic relations between annotation words in a quantitative manner. By adopting the method based on the DBNMI model for realizing automatic annotation of the remote sensing image, a DBNMI-based remote sensing image semantic automatic annotation model is provided for the fact that the region of interest of ocean remote sensing image semantic annotation task only occupies a small part of the image, the precision of image automatic annotation is effectively increased, and the method has wider application range.

Description

Method based on DBNMI model realization remote sensing image automatic marking
Technical field
The present invention relates to technical field of image processing, particularly relating to ocean remote sensing affects label technology field, specifically refers to A kind of method based on DBNMI model realization remote sensing image automatic marking.
Background technology
Remote sensing image meaning automatic marking is to allow machine by learning the image example of known semantic label, then makes term Justice representation of concept one pair does not marks the content of remote sensing image.Along with many ocean remote sensing satellite transmitting (as SeaStar, The satellite such as GEOSAT, ERS-n, HY-1 to HY-3), ocean remote sensing image quantity explosive increase.Marked by remote sensing image semanteme Note, it is possible to achieve remote sensing image data analysis, index and the task such as retrieval, efficiently utilizes remote sensing image.The method of artificial mark There is subjectivity and inexactness, waste time and energy, can not meet current " big data " epoch.Sending out of machine learning scheduling theory Exhibition so that remote sensing image meaning automatic marking becomes a big research direction.
Existing remote sensing image semanteme marking method is set about the most from the following aspect: feature extraction, disaggregated model, high-level semantic Divide level etc..Bratasanu et al. proposes a kind of mapping mode pixel groups of identical semantic concept being combined together, and completes Map the mapping of the high-level semantic arriving human intelligible again to spectrum semantic indexing without semantic low-level image feature.Li é nou et al. is to each Concept gives a training set, utilizes maximum likelihood method to unknown images allocation probability, distributes mould based on potential Di Li Cray Type (Latent Dirichlet Allocation is called for short LDA) model learns.Chen et al. utilizes at ground floor and shows more Example multi-tag study (multi-instance multi-label learning, MIML) framework carries out high-definition remote sensing shadow Image scale is noted, and the second layer uses multiple dimensioned layering semantic information simulation example in the spatial relationship of MIML, and third layer utilizes nonparametric Probabilistic model merges possible class.Yang Yang et al. proposes the degree of depth learning model improved, and the markup information of image is considered as image Classification information, to characteristics of image pay close attention to deficiency.Due to the existence of semantic gap, the automatic marking performance of image semanteme urgently carries High.
Remote sensing image has abundant spatial structural form and geographical feature information, and ocean remote sensing image semantic tagger is appointed Region interested of being engaged in only accounts for image very small part.The contribution that Semantic Similarity is measured by the zones of different of one secondary remote sensing image Degree is different, and in ocean remote sensing image, " subject area " is more much bigger than the contribution of " background area ", and work on hand is distant to ocean The complexity of sense image structure considers few.Diversified semantic hierarchies and pass is there is between remote sensing image mark vocabulary System, including close, oppose, containing etc..Such as, after a secondary remote sensing image is noted as the vocabulary such as " steamer ", " sea water ", " port Mouthful " as this image mark vocabulary probability will promote accordingly.
Summary of the invention
It is an object of the invention to the shortcoming overcoming above-mentioned prior art, it is provided that one is capable of for ocean remote sensing Image semantic tagger task region interested only accounts for image very small part, improves the essence of ocean remote sensing image automatic marking The method based on DBNMI model realization remote sensing image automatic marking of degree.
To achieve these goals, the present invention has a following composition:
Being somebody's turn to do method based on DBNMI model realization remote sensing image automatic marking, it is mainly characterized by, and described method includes Following steps:
(1) initial input remote sensing image carrying out adaptivenon-uniform sampling, coarseness divides remote sensing image background area and object Region；
(2) for the subject area of the remote sensing image marked off, many examples multi-tag technology is utilized to represent；
(3) for relation between the Low Level Vision feature of subject area of remote sensing image and high level semantic-concept, based on DBNMI Model carries out automatic modeling；
(4) co-occurrence and the semantic relation of opposition between quantitative Analysis mark word.
It is preferred that described remote sensing image is ocean remote sensing image.
It is preferred that described step (1), comprise the following steps:
(1-1) wavelet transformation is utilized to produce the pyramid form of multiresolution image；
(1-2) self adaptation builds the image information hierarchical organization of different scale；
(1-3) select suitably to split yardstick according to video scenery complexity, remote sensing image is split.
It is preferred that described step (2), comprise the following steps:
(2-1) (X is usedi,Li) represent width remote sensing image, wherein an Xi={ xi 1,xi 2,…,xi mRepresent instance space, m It is example number, Li={ li 1,li 2,…,li nRepresent tally set, n label number；
(2-2) every width remote sensing image is as a bag in different spaces yardstick, and relevant cut zone is considered as example, its In i-th layer with bag Bi={ xi1,xi2,…,ximRepresent；
(2-3) threshold value t is set, calculates background area and subject area similarity s, it is judged that the magnitude relationship of s and t, if s During more than predetermined threshold value t, then i-th layer of bag needs to continue to divide, and otherwise i-th layer of bag needs not continue to divide.
It is preferred that described step (3), comprise the following steps:
(3-1) subject area of remote sensing image is carried out feature extraction, using the feature that extracts as the first of DBN model Layer, i.e. input layer；
(3-2) DBN model carries out successively eigentransformation to input data, is transformed in the character representation in former space by sample One new feature space；
(3-3) obtain, by limited Boltzmann machine, the weights that expression is optimal from low-level image feature to high-level semantic, utilize anti- To communication network, degree of depth belief network is carried out tuning.
It is preferred that described step (4), comprise the following steps:
(4-1) as F (Bi, L)=1 time,Represent cooccurrence relation, wherein B between mark wordiFor i-th Bag, L represents semantic label collection, represents bag BiIn containing label l, then during this bag can select semantic label collection L belonging to this label Other labels；
(4-2) as F (Bi,L1,L2During)=1,Count(Bi,l1)≥1,Δ(Bi,l2)=0, represents Antagonistic relations, wherein semantic label collection L between mark word1,L2Belong to antagonistic relations, wrap BiIn containing label l1, then this bag can select Select semantic label collection L belonging to this label1In other labels, but semantic label collection L can not be selected2In label.
It is preferred that described method is carried out on Matlab experiment porch.
Have employed the method based on DBNMI model realization remote sensing image automatic marking in this invention, for ocean remote sensing Image semantic tagger task region interested only accounts for image very small part, proposes a kind of based on DBNMI (degree of depth belief network Many examples) remote sensing image meaning automatic marking model, initial input remote sensing image is carried out adaptivenon-uniform sampling, coarseness divides Ocean remote sensing image background area and subject area, between the Low Level Vision feature in imaged object region and high level semantic-concept Relation, degree of depth belief network model based on limited Boltzmann machine carries out automatic modeling, between quantitative Analysis mark word co-occurrence and The semantic relation of opposition, is effectively improved the precision of image automatic marking, has wider application scope.
Accompanying drawing explanation
Fig. 1 is the schematic diagram of each cut zone of the present invention.
Fig. 2 is that the counterpropagation network that utilizes of the present invention carries out the flow chart of tuning to degree of depth belief network.
Fig. 3 be the present invention by study to marking model complete the flow chart of semantic tagger of new image.
Fig. 4 is the flow chart of the method for the remote sensing image automatic marking of the present invention.
Detailed description of the invention
In order to more clearly describe the technology contents of the present invention, carry out further below in conjunction with specific embodiment Describe.
The present invention solves the problems referred to above and be employed technical scheme comprise that, specifically comprises the following steps that
Step one, carries out adaptivenon-uniform sampling to initial input ocean remote sensing image, and coarseness divides remote sensing image background area Territory and subject area；
Step 2, for the subject area of the ocean remote sensing image marked off, utilizes many examples to represent；
Step 3, for relation between the Low Level Vision feature of remote sensing image subject area and high level semantic-concept, based on DBNMI model carries out automatic modeling；
Step 4, co-occurrence and the semantic relation of opposition between quantitative Analysis mark word.
In one preferably embodiment, step one utilize wavelet transformation produce the pyramid of multiresolution image Formula, self adaptation builds the image information hierarchical organization of different scale, it is achieved remote sensing image information biography between different scale layer Pass.Its implementation is as follows,
Selecting suitably to split yardstick according to video scenery complexity, simple scenario selects big yardstick, and complex scene selects Select little yardstick；
Wavelet transformation is provided that image expression under different resolution, it is achieved image according to different scale and displacement point Solve.
In one preferably embodiment, step 2 utilize many examples multi-tag (MIML) technology represent ocean remote sensing Imaged object region, its implementation is as follows,
Use (Xi,Li) represent width remote sensing image, wherein an Xi={ xi 1,xi 2,…,xi mRepresenting instance space, m is example Number, Li={ li 1,li 2,…,li nRepresent tally set, n label number；
Every width image is as a bag (bag) in different spaces yardstick, and relevant cut zone is considered as example (instances), wherein i-th layer with bag Bi={ xi1,xi2,…,xim}.Self adaptation arranges threshold value t, calculates background area with right As Regional Similarity s, if s is more than predetermined threshold value t, i-th layer of bag needs to continue to divide, and then need not during less than threshold value t.
And, for relation between the Low Level Vision feature in imaged object region and high level semantic-concept in step 3, based on DBNMI model carries out automatic modeling, and its implementation is as follows,
The remote sensing image subject area generated by wavelet transformation is carried out feature extraction, using the feature that extracts as The ground floor of DBN, i.e. input layer；
DBN model carries out successively eigentransformation to input data, and the sample character representation in former space is transformed to one New feature space；
Obtained by limited Boltzmann machine from low-level image feature to high-level semantic, express optimal weights, utilize back propagation Network carries out tuning to degree of depth belief network, it is achieved degree of depth belief network model automatic marking based on limited Boltzmann machine.
In one preferably embodiment, in step 4, quantitative Analysis marks co-occurrence and the semantic relation of opposition between word, Its implementation is as follows,
As F (Bi, L)=1 time,Represent cooccurrence relation between mark word.Wherein BiFor i-th bag, L Represent semantic label collection.Represent bag BiIn containing label l, then its during this bag can select semantic label collection L belonging to this label His label；
As F (Bi,L1,L2During)=1,Count(Bi,l1)≥1,Δ(Bi,l2)=0, represents mark word Between antagonistic relations.Wherein semantic label collection L1,L2Belong to antagonistic relations, wrap BiIn containing label l1, then this bag can select this mark Semantic label collection L belonging to signing1In other labels, but semantic label collection L can not be selected2In label.
It is further described the present invention below with a specific embodiment:
The present invention realizes on Matlab2013a experiment porch, mainly includes three steps, specific as follows:
Every width remote sensing image utilizes wavelet pyramid method to carry out 3 layers of resolution pretreatment, obtains different resolution yardstick Under image；
On different spaces yardstick, remote sensing image being carried out coarseness segmentation, image is considered as bag, and relevant cut zone regards For example, each bag is made up of 2-10 example, as shown in Figure 1.Wherein i-th layer with bag Bi={ xi1,xi2,…,ximRepresent. Arranging threshold value t, calculate background area and subject area similarity s, if s is more than predetermined threshold value t, i-th layer of bag needs to continue to draw Point, then need not during less than threshold value t.
Training set each width remote sensing image (Xi,Li) represent, wherein Xi={ xi 1,xi 2,…,xi mRepresent instance space, m It is example number, Li={ li 1,li 2,…,li nRepresent tally set, n label number.It is intended to from training set, obtain a function: fMIML:2x→2l, input here is the instance space X that a width remote sensing image producesi, output is the mark that this width remote sensing image is corresponding Sign collection Li。
The characteristic extracted by each width remote sensing image, is combined into one-dimensional vector.All training set image feature numbers According to being combined into matrix, every a line represents the feature of the extraction of a secondary remote sensing image.Matrix all values is normalized to [0,1] by row Scope, as the ground floor of DBN model, i.e. input layer.
When the state of given visible layer node, it is separate between the state of activation of each hidden layer node, it may be assumed thatWhen the state of given hidden layer node, it is mutual between the state of activation of each visible layer node Independent, it may be assumed thatLimited Boltzmann machine visible layer input for v ∈ 0,1} binary variable, it is assumed that Each neuron follows Bernoulli Jacob's distribution, and the energy function defining this model is as follows: Wherein θ={ a, b, W} are the parameters of model.It is available that (v, joint probability distribution h) is P (v, h)=e-E(v,h)/ Z, wherein, Z For normaliztion constant.
Input data are carried out successively eigentransformation, the sample character representation in former space is transformed to a new feature sky Between；Obtained by limited Boltzmann machine from low-level image feature to high-level semantic, express optimal weights, utilize counterpropagation network Degree of depth belief network is carried out tuning, uses the mode of iteration to be trained, aim at and learn parameter θ={ a, b, W}'s Value, as shown in Figure 2.
Probability any two mark word occur simultaneously carries out quantitative Analysis, uses BiRepresenting i-th bag, L represents semantic Tally set, by mark F (Bi, L)=1, represent there is cooccurrence relation, i.e. between mark wordRepresent bag Bi In containing label l, then other labels during this bag can select semantic label collection L belonging to this label；
The probability that any two mark word occur will not simultaneously carries out quantitative Analysis, uses BiRepresent i-th bag, semantic mark Sign collection L1,L2Belong to antagonistic relations, by mark F (Bi,L1,L2)=1, represents have antagonistic relations, i.e. between mark wordCount(Bi,l1)≥1,Δ(Bi,l2)=0, wraps BiIn containing label l1, then this bag can select this label Affiliated semantic label collection L1In other labels, but semantic label collection L can not be selected2In label.
The present invention proposes a kind of remote sensing image meaning automatic marking mould based on the many examples of degree of depth belief network (DBNMI) Type, it is intended to improve the precision of mark.Realize process as it is shown on figure 3, utilize training set image to Low Level Vision feature and high-rise language Justice conceptual relation carries out automatic modeling, by study to marking model complete the semantic tagger of new image.
Have employed the method based on DBNMI model realization remote sensing image automatic marking in this invention, for ocean remote sensing Image semantic tagger task region interested only accounts for image very small part, proposes a kind of based on the many examples of degree of depth belief network Remote sensing image meaning automatic marking model, initial input remote sensing image is carried out adaptivenon-uniform sampling, it is distant that coarseness divides ocean Sense image background area and subject area, for relation between the Low Level Vision feature in imaged object region and high level semantic-concept, Degree of depth belief network model based on limited Boltzmann machine carries out automatic modeling, co-occurrence and opposition between quantitative Analysis mark word Semantic relation, is effectively improved the precision of image automatic marking, has wider application scope.
In this description, the present invention is described with reference to its specific embodiment.But it is clear that still may be made that Various modifications and alterations are without departing from the spirit and scope of the present invention.Therefore, specification and drawings is considered as illustrative And it is nonrestrictive.

Claims (7)

1. a method based on DBNMI model realization remote sensing image automatic marking, it is characterised in that described method include with Lower step:
(1) initial input remote sensing image carrying out adaptivenon-uniform sampling, coarseness divides remote sensing image background area and subject area；
(2) for the subject area of the remote sensing image marked off, many examples multi-tag technology is utilized to represent；
(3) for relation between the Low Level Vision feature of subject area of remote sensing image and high level semantic-concept, based on DBNMI model Carry out automatic modeling；
(4) co-occurrence and the semantic relation of opposition between quantitative Analysis mark word.
Method based on DBNMI model realization remote sensing image automatic marking the most according to claim 1, it is characterised in that Described remote sensing image is ocean remote sensing image.
Method based on DBNMI model realization remote sensing image automatic marking the most according to claim 1, it is characterised in that Described step (1), comprises the following steps:
(1-1) wavelet transformation is utilized to produce the pyramid form of multiresolution image；
(1-2) self adaptation builds the image information hierarchical organization of different scale；
(1-3) select suitably to split yardstick according to video scenery complexity, remote sensing image is split.
Method based on DBNMI model realization remote sensing image automatic marking the most according to claim 1, it is characterised in that Described step (2), comprises the following steps:
(2-1) (X is usedi,Li) represent width remote sensing image, wherein an Xi={ xi 1,xi 2,…,xi mRepresenting instance space, m is to show Example number, Li={ li 1,li 2,…,li nRepresent tally set, n label number；
(2-2) every width remote sensing image is as a bag in different spaces yardstick, and relevant cut zone is considered as example, Qi Zhong I layer bag Bi={ xi1,xi2,…,ximRepresent；
(2-3) threshold value t is set, calculates background area and subject area similarity s, it is judged that the magnitude relationship of s and t, if s is more than During predetermined threshold value t, then i-th layer of bag needs to continue to divide, and otherwise i-th layer of bag needs not continue to divide.
Method based on DBNMI model realization remote sensing image automatic marking the most according to claim 1, it is characterised in that Described step (3), comprises the following steps:
(3-1) subject area of remote sensing image is carried out feature extraction, using the feature that extracts as the ground floor of DBN model, I.e. input layer；
(3-2) DBN model carries out successively eigentransformation to input data, and the sample character representation in former space is transformed to one New feature space；
(3-3) obtain, by limited Boltzmann machine, the weights that expression is optimal from low-level image feature to high-level semantic, utilize and reversely pass Broadcast network and degree of depth belief network is carried out tuning.
Method based on DBNMI model realization remote sensing image automatic marking the most according to claim 1, it is characterised in that Described step (4), comprises the following steps:
(4-1) as F (Bi, L)=1 time,Represent cooccurrence relation, wherein B between mark wordiFor i-th bag, L Represent semantic label collection, represent bag BiIn containing label l, then its during this bag can select semantic label collection L belonging to this label His label；
(4-2) as F (Bi,L1,L2During)=1,Count(Bi,l1)≥1,Δ(Bi,l2)=0, represents mark word Between antagonistic relations, wherein semantic label collection L1,L2Belong to antagonistic relations, wrap BiIn containing label l1, then this bag can select this mark Semantic label collection L belonging to signing1In other labels, but semantic label collection L can not be selected2In label.
Method based on DBNMI model realization remote sensing image automatic marking the most according to claim 1, it is characterised in that Described method is carried out on Matlab experiment porch.
CN201610383452.1A 2016-06-02 2016-06-02 Method based on DBNMI model realization remote sensing image automatic markings Active CN106056609B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610383452.1A CN106056609B (en) 2016-06-02 2016-06-02 Method based on DBNMI model realization remote sensing image automatic markings

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610383452.1A CN106056609B (en) 2016-06-02 2016-06-02 Method based on DBNMI model realization remote sensing image automatic markings

Publications (2)

Publication Number Publication Date
CN106056609A true CN106056609A (en) 2016-10-26
CN106056609B CN106056609B (en) 2018-11-06

Family

ID=57172795

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610383452.1A Active CN106056609B (en) 2016-06-02 2016-06-02 Method based on DBNMI model realization remote sensing image automatic markings

Country Status (1)

Country Link
CN (1) CN106056609B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107169051A (en) * 2017-04-26 2017-09-15 山东师范大学 Based on semantic related method for searching three-dimension model and system between body
CN108319972A (en) * 2018-01-18 2018-07-24 南京师范大学 A kind of end-to-end difference online learning methods for image, semantic segmentation
CN110704549A (en) * 2019-10-09 2020-01-17 中国石油大学(华东) Method, system, medium and device for selecting and constructing marine environment data service granularity
CN111080625A (en) * 2019-12-18 2020-04-28 北京推想科技有限公司 Training method and training device for lung image strip and rope detection model
CN113761971A (en) * 2020-06-02 2021-12-07 中国人民解放军战略支援部队信息工程大学 Method and device for constructing target knowledge graph of remote sensing image

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103823845A (en) * 2014-01-28 2014-05-28 浙江大学 Method for automatically annotating remote sensing images on basis of deep learning

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103823845A (en) * 2014-01-28 2014-05-28 浙江大学 Method for automatically annotating remote sensing images on basis of deep learning

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
MINH N. DO ETAL: "Rotation Invariant Texture Characterization and Retrieval Using Steerable Wavelet-Domain Hidden Markov Models", 《IEEE TRANSACTIONS ON MULTIMEDIA》 *
XIAOCHUN CAO ETAL.: "SLED: Semantic Label Embedding Dictionary Representation for Multilabel Image Annotation", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 *
吕启 等: "基于DBN模型的遥感图像分类", 《计算机研究与发展》 *
张滢 等: "基于特征提取和多示例学习的图像区域标注", 《电子测量与仪器学报》 *
杨阳 等: "基于深度学习的图像自动标注算法", 《数据采集与处理》 *
葛永 等: "基于多示例学习的遥感图像检索", 《中国科学技术大学学报》 *
黄志坚 等: "基于视觉复杂度的自适应尺度遥感影像分割", 《电子与信息学报》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107169051A (en) * 2017-04-26 2017-09-15 山东师范大学 Based on semantic related method for searching three-dimension model and system between body
CN107169051B (en) * 2017-04-26 2019-09-24 山东师范大学 Based on relevant method for searching three-dimension model semantic between ontology and system
CN108319972A (en) * 2018-01-18 2018-07-24 南京师范大学 A kind of end-to-end difference online learning methods for image, semantic segmentation
CN108319972B (en) * 2018-01-18 2021-11-02 南京师范大学 End-to-end difference network learning method for image semantic segmentation
CN110704549A (en) * 2019-10-09 2020-01-17 中国石油大学(华东) Method, system, medium and device for selecting and constructing marine environment data service granularity
CN111080625A (en) * 2019-12-18 2020-04-28 北京推想科技有限公司 Training method and training device for lung image strip and rope detection model
CN111080625B (en) * 2019-12-18 2020-12-29 推想医疗科技股份有限公司 Training method and training device for lung image strip and rope detection model
CN113761971A (en) * 2020-06-02 2021-12-07 中国人民解放军战略支援部队信息工程大学 Method and device for constructing target knowledge graph of remote sensing image
CN113761971B (en) * 2020-06-02 2023-06-20 中国人民解放军战略支援部队信息工程大学 Remote sensing image target knowledge graph construction method and device

Also Published As

Publication number Publication date
CN106056609B (en) 2018-11-06

Similar Documents

Publication Publication Date Title
Zhu et al. Adaptive deep sparse semantic modeling framework for high spatial resolution image scene classification
Zhong et al. SatCNN: Satellite image dataset classification using agile convolutional neural networks
Shi et al. Road detection from remote sensing images by generative adversarial networks
Văduva et al. Latent Dirichlet allocation for spatial analysis of satellite images
CN102902821B (en) The image high-level semantics mark of much-talked-about topic Network Based, search method and device
CN106991382A (en) A kind of remote sensing scene classification method
CN107704877A (en) A kind of image privacy cognitive method based on deep learning
CN106056609A (en) Method based on DBNMI model for realizing automatic annotation of remote sensing image
CN107247938A (en) A kind of method of high-resolution remote sensing image City Building function classification
CN101763440B (en) Method for filtering searched images
Wang et al. Evolutionary computation
Kim et al. CityCraft: 3D virtual city creation from a single image
CN106599051A (en) Method for automatically annotating image on the basis of generation of image annotation library
CN103745201B (en) A kind of program identification method and device
Ye et al. Parallel multi-stage features fusion of deep convolutional neural networks for aerial scene classification
CN104346814B (en) Based on the SAR image segmentation method that level vision is semantic
Ning et al. Choosing an appropriate training set size when using existing data to train neural networks for land cover segmentation
CN104751175A (en) Multi-label scene classification method of SAR (Synthetic Aperture Radar) image based on incremental support vector machine
Zhang et al. City2vec: Urban knowledge discovery based on population mobile network
CN108897750A (en) Merge the personalized location recommendation method and equipment of polynary contextual information
Tu et al. Scale effect on fusing remote sensing and human sensing to portray urban functions
CN102945372B (en) Classifying method based on multi-label constraint support vector machine
Li et al. An aerial image segmentation approach based on enhanced multi-scale convolutional neural network
CN104484347A (en) Geographic information based hierarchical visual feature extracting method
CN113052121B (en) Multi-level network map intelligent generation method based on remote sensing image

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant