CN105005789A - Vision lexicon based remote sensing image terrain classification method - Google Patents
Vision lexicon based remote sensing image terrain classification method Download PDFInfo
- Publication number
- CN105005789A CN105005789A CN201510379234.6A CN201510379234A CN105005789A CN 105005789 A CN105005789 A CN 105005789A CN 201510379234 A CN201510379234 A CN 201510379234A CN 105005789 A CN105005789 A CN 105005789A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- slice map
- word
- images
- preliminary
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a vision lexicon based remote sensing image terrain classification method. The method comprises the following steps of: firstly classifying all remote sensing images into a training set and a test set; obtaining preliminary slice images by cutting each remote sensing image at a fixed size; and extracting a preliminary slice image that comprises a target; generating a multi-layer Gaussian space pyramid for the preliminary slice image that comprises the target by means of Gaussian blur and sampling; performing both SIFT feature extraction and LBP feature extraction on images of each layer; performing clustering on remote sensing words of all the remote sensing images in the training set and the test set, to obtain a plurality of cluster centers, wherein all the cluster centers constitute a remote sensing dictionary; and setting different radii; establishing a frequency histogram for remote sensing words in different radii of each preliminary slice image: performing training for the frequency histogram of remote sensing words in the training set by using a support vector machine RBF-SVM, and then performing terrain classification on remote sensing images in the test set by using the trained RBF-SVM.
Description
Technical field
The invention belongs to technical field of remote sensing image processing, relate to a kind of terrain classification method based on high-resolution remote sensing image.
Background technology
Along with developing rapidly of remote sensing satellite, the resolution of the satellite-remote-sensing image obtained is also more and more higher, the processing requirements for remote sensing images caused thus is also more and more higher, wherein, object Classification and Identification for High Resolution Remote Sensing Satellites is an important research direction, especially city scientific allocation, agricultural planting distribution planning and the classification of military sensitive target are extracted all significant.The high resolving power of High Resolution Remote Sensing Satellites also means that details is abundanter simultaneously, and background is also larger on the impact of target, and this brings challenge with regard to the classification for image.
In remote sensing image classification field, mainly comprising two class sorting techniques: a class is non-supervisory method, by using the thought of cluster to realize, comprising k-means, the methods such as isodata.But have inter-class variance to reduce along with remote sensing images resolution is more and more higher, variance within clusters increases, thus cause between class overlapping, occur the phenomenon of the different spectrum of jljl and same object different images, cause classification error; Another kind of is have measure of supervision, also be the sorting technique generally used now, adopt the thought of learning and training, comprise BP neural network model, genetic model and support vector machine etc., the method for supervision possesses certain increasing amount adjustment, but when remote sensing images resolution constantly increases, for the selection of training sample and be all a challenge for the training speed of every class sample, especially the selection of training sample is the process of an artificial selection, brings larger time cost.
The simultaneously development of high-resolution remote sensing image means that the kind of feature selecting increases, that set up based on the low-level image features such as color, texture, shape with relation that is sample high layer information, there will be generalization difference, the shortcoming that relevance grade is not high.In order to overcome the wide gap of low-level image feature and the high-level semantic information of sample, the derivation model be born thus between based on sample intermediate features and the high-level semantic information of sample.BOV (visual word bag) model, as the representative of intermediate features, has had some to be in progress in OBIA (object-based remote-sensingimage analysis).But traditional self defectiveness of BOV model method, the full detail utilizing image that can not be good, such as spatial level information etc., and BOV category of model is mainly used in the Classification and Identification of natural forms, does not utilize very well for remote sensing images terrain classification.
One of research direction that remote sensing images terrain classification become remote sensing images field important of how using the thought of word bag to solve.But because remote sensing images ground species is varied, single Feature Words bag method and single remote sensing word frequencies histogram are difficult to represent all remote sensing images kinds.
Summary of the invention
In view of this, the invention provides a kind of remote sensing images terrain classification method of view-based access control model vocabulary, solve the problem of remote sensing images terrain classification accuracy rate deficiency.
In order to achieve the above object, technical scheme of the present invention comprises the steps:
Step one, all remote sensing images are divided into training set and test set, for every width remote sensing images, with fixing size, preliminary slice map are obtained to its cutting, extract the preliminary slice map comprising target.
Step 2, in step one extract the preliminary slice map comprising target by Gaussian Blur with sampling generate multilayer Gaussian spatial pyramid.
Step 3, all carry out feature extraction to the image of the every one deck in Gaussian spatial pyramid, feature extraction comprises local feature and SIFT feature is extracted and image texture characteristic and LBP feature extraction.
When carrying out SIFT feature and extracting, only carry out one deck gaussian filtering, then by the mode of sliding window, image is divided into the segmentation slice map of 16 × 16, carries out according to the graded of each segmentation slice map the SIFT feature vector that SIFT special medical treatment vector extracts 128 dimensions obtained about this segmentation slice map; The LBP proper vector about this segmentation slice map is determined according to the radius of setting and sampling number in each segmentation slice map.
The SIFT feature vector sum LBP combination of eigenvectors that then each segmentation slice map is corresponding forms a remote sensing word; Generate the remote sensing word about these remote sensing images thus.
Wherein top Gaussian spatial pyramid to meet in its preliminary slice map and correspondingly obtains remote sensing word number and be at least 1/2 of the quantity of remote sensing dictionary.
The mode of following steps four ~ five is taked to process for the preliminary slice map in each layer of gaussian pyramid:
Step 4, cluster is carried out to the remote sensing word of remote sensing images all in training set and test set, obtain multiple cluster centre, all cluster centre composition remote sensing dictionaries.
In cluster process, adopt PCA principal component analysis (PCA) to carry out dimensionality reduction to remote sensing word, retain and maximum dimension is contributed to covariance, and setting data Loss Rate threshold value disratio, make data loss rate be no more than disratio.
Step 5, setting different radii value, all frequency histogram is set up to the remote sensing word in each preliminary slice map different radii value:
This histogrammic horizontal ordinate is each remote sensing word, ordinate is the frequency occurred in the preliminary slice map of remote sensing word in current radius value, the computing method of this frequency are: frequency values is initially 0, calculate the Euclidean distance between two between remote sensing word in the preliminary slice map in current radius value, if the Euclidean distance between current remote sensing word A to another remote sensing word B is the minimum value of other all remote sensing words to the Euclidean distance of B, then the frequency values of A increases by 1.
Step 6, for remote sensing word in training set frequency histogram use support vector machine RBF-SVM train, then use training after RBF-SVM terrain classification is carried out to remote sensing images in test set.
If the number of plies of the gaussian pyramid belonging to remote sensing word is at the bottom, then distribute the maximum weight of sorter in the support vector machine of its correspondence, more up, the weights of corresponding sorter are less for the number of plies.
Further, when image being divided into the segmentation slice map of 16*16 by the mode of sliding window, the step-length of sliding window is set between 1 to 8.
Further, the Selection of kernel function histogram intersectionkernel in RBF-SVM support vector machine.
Further, in step 2, during cluster, adopt K-means clustering method.
Beneficial effect:
1, first remote sensing images collection is divided into test set training set two parts by the method, then feature point extraction is carried out to the section of all remote sensing images, obtain the word of every width slice map, carry out cluster for the unique point of training set afterwards and obtain dictionary, the word being finally directed to dictionary and test set builds frequency histogram training, use the word histogram of test set to test, can classify to remote sensing image atural object more accurately, classification results is more accurate.
2, adopt the method for multilayer gaussian pyramid in this method when carrying out Gaussian Blur and sampling to slice map, therefore, it is possible to utilize the spatial information of remote sensing images better, thus make classification more accurate.
3, this method only comprises direction in 16*16 slice map and gradient information traditional due to SIFT feature vector, the i.e. information of stable point in slice map, do not possess the texture feature information of whole slice map, therefore on the basis of SIFT feature, add the even LBP feature of invariable rotary, compensate for the texture feature information do not possessed in SIFT feature, all information in remote sensing images can be expressed in all directions.
4, in this method when generated frequency histogram, choose multiple different radius, ensure that selected radius should make inside comprise enough remote sensing word quantity simultaneously, to the remote sensing word in one deck slice map different radii every in pyramid and remote sensing dictionary generated frequency histogram, like this can on the basis of the histogrammic demand of satisfied generation remote sensing word frequencies, utilize the spatial information in remote sensing images from center to surrounding simultaneously, thus the target can expressed better in remote sensing images, improve classify accuracy.
5, this method final utilization multi-categorizer adds decision-making device and obtains the result of decision, classify for the remote sensing word frequencies histogram of every one deck in the empty pyramid of Gauss, weights are distributed to different sorters, we think the information maximum weight of gaussian pyramid bottom, more up information weights are less, the generic of remote sensing slice map is obtained through decision-making, weights are distributed by giving gaussian pyramid classification results each time, do not trust merely the classification results of a certain tomographic image, the target in remote sensing images is jointly determined by the classification results of different levels gaussian pyramid, improve classify accuracy.
Accompanying drawing explanation
Fig. 1 is implementing procedure figure of the present invention.
Fig. 2 is the product process figure of remote sensing word of the present invention.
Fig. 3 is the histogrammic product process figure of remote sensing word frequencies of the present invention.
Fig. 4 is the design of sorter of the present invention.
Embodiment
To develop simultaneously embodiment below in conjunction with accompanying drawing, describe the present invention.
Step one, all remote sensing images are divided into training set and test set, for every width remote sensing images, with fixing size, preliminary slice map are obtained to its cutting, extract the preliminary slice map comprising target;
Step 2, for a kind of extraction of step the preliminary slice map comprising target by Gaussian Blur with sampling generate multilayer Gaussian spatial pyramid, the pyramidal number of plies of Gaussian spatial is determined according to the size of preliminary slice map;
Accompanying drawing 2 is the product process figure of remote sensing word, comprises Gaussian spatial pyramid and generates and multi-feature extraction two parts.
First multilayer Gaussian spatial pyramid is generated for remote sensing slice map by Gaussian Blur and sampling, be directed to the slice map of different size, the Gaussian spatial pyramid of the different number of plies can be generated, top pyramid require to meet at least obtain remote sensing dictionary quantity 1/2 with the remote sensing word frequencies histogram after forming, be 16 stack features values in the present embodiment, can ensure that generated frequency histogram is meaningful like this.
The Gaussian spatial pyramid of multilayer can utilize the spatial information of remote sensing images better, thus makes classification more accurate.
Secondly, carry out multiple features (SIFT feature and LBP feature) for the every piece image in Gaussian spatial pyramid and extract.In order to keep the consistance of every width picture frequency histogram quantity, SIFT feature is changed, only carry out one deck gaussian filtering, by the mode of sliding window, image is divided into segmentation slice map, it is the slice map of 16*16 in the present embodiment, the step-length of sliding window can be set to 1 to 8 according to actual needs, obtains the SIFT feature vector of 128 dimensions according to the graded of the segmentation slice map of each 16*16.The SIFT feature being directed to each sub-picture is like this counted unanimously, the statistics with histogram after being convenient to.
The direction in 16*16 slice map and gradient information is only comprised because SIFT feature is vectorial, the i.e. information of stable point in slice map, do not possess the texture feature information of whole slice map, therefore on the basis of SIFT feature, add the even LBP feature of invariable rotary to make up the sign for image, all information in remote sensing images can be expressed in all directions.In the slice map of each 16*16, determine LBP feature according to radius and sampling number (artificially setting), owing to have employed the LBP feature of invariable rotary, make the dimension of representative image textural characteristics less.
The SIFT feature of every width image being got up with the even constant LBP integrate features of rotation is exactly the remote sensing word (the corresponding remote sensing word of slice map of each 16*16) representing this width image, and the remote sensing word quantity generated from every width slice map is consistent.
Step 2, carry out cluster based on remote sensing words all in training set and test set, obtain multiple cluster centre, all cluster centres just constitute remote sensing dictionary.
Due to the process that cluster is an off-line, along with increasing of cluster data, the overall partition clustering method of ordinary meaning can produce the computing cost of geometry multiple.Therefore the thought of dimensionality reduction cluster processes extensive and ultra-large remote sensing dictionary data.
Here, PCA principal component analysis (PCA) is adopted to carry out dimensionality reduction to remote sensing word, retain and maximum dimension is contributed to covariance, but it should be noted that, dimensionality reduction number needs to ensure not lose useful data in a large number, therefore, definition data loss rate disratio, ensures that remaining data retains most of useful information.
The method of cluster adopts the K-means clustering method of time efficiency and space efficiency equilibrium, and K-means cluster directly can obtain cluster centre, and cluster centre is the mean value of all properties in class, can well represent Lei Nei center.The number of cluster centre needs to determine according to the number of classification, and need the classification number of classification more, the cluster centre of needs is more, also just means that the word number in remote sensing dictionary is more, could meet the demand of multi-class targets classification accuracy.
Step 3, setting different radii value, all frequency histogram is set up to the remote sensing word in each preliminary slice map different radii value:
This histogrammic horizontal ordinate is each remote sensing word, ordinate is the frequency occurred in the preliminary slice map of remote sensing word in current radius value, the computing method of this frequency are: frequency values is initially 0, calculate the Euclidean distance between two between remote sensing word in the preliminary slice map in current radius value, if the Euclidean distance between current remote sensing word A to another remote sensing word B is the minimum value of other all remote sensing words to the Euclidean distance of B, then the frequency values of A increases by 1.
Accompanying drawing 3 is the histogrammic product process figure of remote sensing word frequencies of the present invention, here in order to effectively utilize the spatial positional information in picture, be directed to the spatial pyramid of different levels, choose suitable radius, to the remote sensing word in one deck slice map different radii every in pyramid and remote sensing dictionary generated frequency histogram, the remote sensing word frequencies histogram all about this slice map is combined, the representatively remote sensing word frequencies histogram of this preliminary slice map.
When choosing suitable radius, its inside should be met and comprise enough remote sensing word quantity, thus meet the histogrammic demand of generation remote sensing word frequencies.Select the benefit of multiple radius to be the spatial information utilized in remote sensing images from center to surrounding, thus the target in remote sensing images can be expressed better, improve classify accuracy.
Step 4, use RBF-SVM support vector machine to train for the remote sensing frequency histogram of the remote sensing word frequencies histogram of training set and test set, Selection of kernel function has the histogram intersection kernel of good behaviour in statistics field.
Accompanying drawing 4 is the formation of sorter, use multi-categorizer to add decision-making device and obtain the result of decision, classify for the remote sensing word frequencies histogram of every one deck in the empty pyramid of Gauss, weights are distributed to different sorters, we think the information maximum weight of gaussian pyramid bottom, more up information weights are less, obtain the generic of remote sensing slice map through decision-making.By distributing weights to gaussian pyramid classification results each time, not trusting merely the classification results of a certain tomographic image, jointly being determined the target in remote sensing images by the classification results of different levels gaussian pyramid, improving classify accuracy.
Since then, the terrain classification of remote sensing images is just completed.
To sum up, 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 (4)
1. a remote sensing images terrain classification method for view-based access control model vocabulary, is characterized in that, comprise the steps:
Step one, all remote sensing images are divided into training set and test set, for every width remote sensing images, with fixing size, preliminary slice map are obtained to its cutting, extract the preliminary slice map comprising target;
Step 2, in step one extract the preliminary slice map comprising target by Gaussian Blur with sampling generate multilayer Gaussian spatial pyramid;
Step 3, all carry out feature extraction to the image of the every one deck in described Gaussian spatial pyramid, described feature extraction comprises local feature and SIFT feature is extracted and image texture characteristic and LBP feature extraction;
When carrying out SIFT feature and extracting, only carry out one deck gaussian filtering, then by the mode of sliding window, image is divided into the segmentation slice map of 16 × 16, carries out according to the graded of each segmentation slice map the SIFT feature vector that SIFT special medical treatment vector extracts 128 dimensions obtained about this segmentation slice map; The LBP proper vector about this segmentation slice map is determined according to the radius of setting and sampling number in each segmentation slice map;
The SIFT feature vector sum LBP combination of eigenvectors that then each segmentation slice map is corresponding forms a remote sensing word; Generate the remote sensing word about these remote sensing images thus;
Wherein top Gaussian spatial pyramid to meet in its preliminary slice map and correspondingly obtains remote sensing word number and be at least 1/2 of the quantity of remote sensing dictionary;
The mode of following steps four ~ five is taked to process for the preliminary slice map in each layer of gaussian pyramid:
Step 4, cluster is carried out to the remote sensing word of remote sensing images all in training set and test set, obtain multiple cluster centre, all cluster centre composition remote sensing dictionaries;
In cluster process, adopt PCA principal component analysis (PCA) to carry out dimensionality reduction to remote sensing word, retain and maximum dimension is contributed to covariance, and setting data Loss Rate threshold value disratio, make data loss rate be no more than disratio;
Step 5, setting different radii value, all frequency histogram is set up to the remote sensing word in each preliminary slice map different radii value:
This histogrammic horizontal ordinate is each remote sensing word, ordinate is the frequency occurred in the preliminary slice map of remote sensing word in current radius value, the computing method of this frequency are: frequency values is initially 0, calculate the Euclidean distance between two between remote sensing word in the preliminary slice map in current radius value, if the Euclidean distance between current remote sensing word A to another remote sensing word B is the minimum value of other all remote sensing words to the Euclidean distance of B, then the frequency values of A increases by 1;
Step 6, for remote sensing word in training set frequency histogram use support vector machine RBF-SVM train, then use training after RBF-SVM terrain classification is carried out to remote sensing images in test set;
If the number of plies of the gaussian pyramid belonging to remote sensing word is at the bottom, then distribute the maximum weight of sorter in the support vector machine of its correspondence, more up, the weights of corresponding sorter are less for the number of plies.
2. the remote sensing images terrain classification method of a kind of view-based access control model vocabulary as claimed in claim 1, is characterized in that, when image is divided into the segmentation slice map of 16*16 by the described mode by sliding window, the step-length of sliding window is set between 1 to 8.
3. the remote sensing images terrain classification method of a kind of view-based access control model vocabulary as claimed in claim 1, is characterized in that, the Selection of kernel function histogram intersectionkernel in described RBF-SVM support vector machine.
4. the remote sensing images terrain classification method of a kind of view-based access control model vocabulary as claimed in claim 1, is characterized in that, in step 2, adopts K-means clustering method during cluster.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510379234.6A CN105005789B (en) | 2015-07-01 | 2015-07-01 | A kind of remote sensing images terrain classification method of view-based access control model vocabulary |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510379234.6A CN105005789B (en) | 2015-07-01 | 2015-07-01 | A kind of remote sensing images terrain classification method of view-based access control model vocabulary |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105005789A true CN105005789A (en) | 2015-10-28 |
| CN105005789B CN105005789B (en) | 2018-04-03 |
Family
ID=54378454
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510379234.6A Active CN105005789B (en) | 2015-07-01 | 2015-07-01 | A kind of remote sensing images terrain classification method of view-based access control model vocabulary |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105005789B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105654122A (en) * | 2015-12-28 | 2016-06-08 | 江南大学 | Spatial pyramid object identification method based on kernel function matching |
| CN106778494A (en) * | 2016-11-21 | 2017-05-31 | 河海大学 | A kind of target in hyperspectral remotely sensed image feature extracting method based on SIFT LPP |
| CN108230240A (en) * | 2017-12-31 | 2018-06-29 | 厦门大学 | It is a kind of that the method for position and posture in image city scope is obtained based on deep learning |
| CN108604303A (en) * | 2016-02-09 | 2018-09-28 | 赫尔实验室有限公司 | General image feature from bottom to top and the from top to bottom system and method for entity classification are merged for precise image/video scene classification |
| CN109325501A (en) * | 2018-08-14 | 2019-02-12 | 王斌 | Material Identification method, apparatus and readable storage medium storing program for executing based on guitar backboard image |
| WO2019080430A1 (en) * | 2017-10-27 | 2019-05-02 | 平安科技（深圳）有限公司 | Electronic apparatus, disordered sample arrangement method and computer-readable storage medium |
| CN111563937A (en) * | 2020-07-14 | 2020-08-21 | 成都四方伟业软件股份有限公司 | Picture color extraction method and device |
| CN111860615A (en) * | 2020-06-30 | 2020-10-30 | 河海大学 | Remote sensing water body target extraction method based on feature dictionary fusion |
| CN113012049A (en) * | 2021-04-15 | 2021-06-22 | 山东新一代信息产业技术研究院有限公司 | Remote sensing data privacy protection method based on GAN network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101777125A (en) * | 2010-02-03 | 2010-07-14 | 武汉大学 | Method for supervising and classifying complex class of high-resolution remote sensing image |
| CN102208038A (en) * | 2011-06-27 | 2011-10-05 | 清华大学 | Image classification method based on visual dictionary |
| CN102622607A (en) * | 2012-02-24 | 2012-08-01 | 河海大学 | Remote sensing image classification method based on multi-feature fusion |
-
2015
- 2015-07-01 CN CN201510379234.6A patent/CN105005789B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101777125A (en) * | 2010-02-03 | 2010-07-14 | 武汉大学 | Method for supervising and classifying complex class of high-resolution remote sensing image |
| CN102208038A (en) * | 2011-06-27 | 2011-10-05 | 清华大学 | Image classification method based on visual dictionary |
| CN102622607A (en) * | 2012-02-24 | 2012-08-01 | 河海大学 | Remote sensing image classification method based on multi-feature fusion |
Non-Patent Citations (4)
| Title |
|---|
| WANG YUXIN等: "Bag of spatial visual words model for scene classification", 《COMPUTER SCIENCE》 * |
| 周宇谷等: "基于视觉词袋模型的遥感图像分类方法", 《重庆理工大学学报（自然科学）》 * |
| 王莹: "基于BoW模型的图像分类方法研究", 《万方数据》 * |
| 罗会兰等: "一种基于多级空间视觉词典集体的图像分类方法", 《电子学报》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105654122A (en) * | 2015-12-28 | 2016-06-08 | 江南大学 | Spatial pyramid object identification method based on kernel function matching |
| CN108604303B (en) * | 2016-02-09 | 2022-09-30 | 赫尔实验室有限公司 | System, method, and computer-readable medium for scene classification |
| CN108604303A (en) * | 2016-02-09 | 2018-09-28 | 赫尔实验室有限公司 | General image feature from bottom to top and the from top to bottom system and method for entity classification are merged for precise image/video scene classification |
| CN106778494A (en) * | 2016-11-21 | 2017-05-31 | 河海大学 | A kind of target in hyperspectral remotely sensed image feature extracting method based on SIFT LPP |
| WO2019080430A1 (en) * | 2017-10-27 | 2019-05-02 | 平安科技（深圳）有限公司 | Electronic apparatus, disordered sample arrangement method and computer-readable storage medium |
| CN108230240B (en) * | 2017-12-31 | 2020-07-31 | 厦门大学 | Method for obtaining position and posture in image city range based on deep learning |
| CN108230240A (en) * | 2017-12-31 | 2018-06-29 | 厦门大学 | It is a kind of that the method for position and posture in image city scope is obtained based on deep learning |
| CN109325501A (en) * | 2018-08-14 | 2019-02-12 | 王斌 | Material Identification method, apparatus and readable storage medium storing program for executing based on guitar backboard image |
| CN109325501B (en) * | 2018-08-14 | 2021-12-03 | 王斌 | Guitar backboard image-based material identification method and device and readable storage medium |
| CN111860615A (en) * | 2020-06-30 | 2020-10-30 | 河海大学 | Remote sensing water body target extraction method based on feature dictionary fusion |
| CN111563937A (en) * | 2020-07-14 | 2020-08-21 | 成都四方伟业软件股份有限公司 | Picture color extraction method and device |
| CN113012049A (en) * | 2021-04-15 | 2021-06-22 | 山东新一代信息产业技术研究院有限公司 | Remote sensing data privacy protection method based on GAN network |
| CN113012049B (en) * | 2021-04-15 | 2022-08-02 | 山东新一代信息产业技术研究院有限公司 | Remote sensing data privacy protection method based on GAN network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105005789B (en) | 2018-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105005789B (en) | A kind of remote sensing images terrain classification method of view-based access control model vocabulary | |
| CN104091321B (en) | It is applicable to the extracting method of the multi-level point set feature of ground laser radar point cloud classifications | |
| CN102622607B (en) | Remote sensing image classification method based on multi-feature fusion | |
| CN103955702B (en) | SAR image terrain classification method based on depth RBF network | |
| CN109255790A (en) | A kind of automatic image marking method of Weakly supervised semantic segmentation | |
| CN103971123B (en) | Hyperspectral image classification method based on linear regression Fisher discrimination dictionary learning (LRFDDL) | |
| CN106897679A (en) | A kind of semantic change detecting method and system based on improvement fuzzy C-means clustering | |
| CN103942564B (en) | High-resolution remote sensing image scene classifying method based on unsupervised feature learning | |
| CN104680173B (en) | A kind of remote sensing images scene classification method | |
| CN106991382A (en) | A kind of remote sensing scene classification method | |
| Liu et al. | Multimorphological superpixel model for hyperspectral image classification | |
| CN103984953A (en) | Cityscape image semantic segmentation method based on multi-feature fusion and Boosting decision forest | |
| CN107784657A (en) | A kind of unmanned aerial vehicle remote sensing image partition method based on color space classification | |
| CN102005034A (en) | Remote sensing image segmentation method based on region clustering | |
| CN109657610A (en) | A kind of land use change survey detection method of high-resolution multi-source Remote Sensing Images | |
| CN111339947A (en) | Method and system for extracting remote sensing image fuzzy boundary ground object, storage medium and equipment | |
| CN105046673B (en) | High spectrum image and visual image fusion sorting technique based on self study | |
| CN102982338A (en) | Polarization synthetic aperture radar (SAR) image classification method based on spectral clustering | |
| CN107480620A (en) | Remote sensing images automatic target recognition method based on heterogeneous characteristic fusion | |
| CN107292336A (en) | A kind of Classification of Polarimetric SAR Image method based on DCGAN | |
| CN105303195A (en) | Bag-of-word image classification method | |
| CN105069478A (en) | Hyperspectral remote sensing surface feature classification method based on superpixel-tensor sparse coding | |
| CN110060273A (en) | Remote sensing image landslide plotting method based on deep neural network | |
| CN101196564A (en) | Laplace regularization least square synthetic aperture radar automatic target recognition method | |
| CN105989336A (en) | Scene identification method based on deconvolution deep network learning with weight |
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 |