CN109447151A - A kind of remotely-sensed data analysis method based on deep learning - Google Patents
A kind of remotely-sensed data analysis method based on deep learning Download PDFInfo
- Publication number
- CN109447151A CN109447151A CN201811260948.5A CN201811260948A CN109447151A CN 109447151 A CN109447151 A CN 109447151A CN 201811260948 A CN201811260948 A CN 201811260948A CN 109447151 A CN109447151 A CN 109447151A
- Authority
- CN
- China
- Prior art keywords
- sample
- classification
- marker samples
- learning
- data
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
Abstract
The invention discloses one kind can further increase classification accuracy, reduces the remotely-sensed data analysis method based on deep learning that member's model number influences classification accuracy.This method, comprising the following steps: 1) read the initial data of high-spectrum remote sensing data；2) marker samples are determined；And training sample and test sample are selected in marker samples；3) all training samples carry out unsupervised from coding depth e-learning；4) marker samples exercise supervision depth e-learning；5) unmarked sample input depth network obtains classification results；6) it uses in the Active Learning image classification method never marker samples sampled based on minimum difference and selects optimal sample；7) sample for choosing Active Learning is marked, and never deletes, is added in training sample in marker samples, updates training sample, obtains final classification device；9) classification results precision evaluation is carried out.Sampling this method can be improved classification accuracy, improve classification effectiveness.
Description
Technical field
The present invention relates to the analysis fields of remotely-sensed data, and in particular to a kind of remotely-sensed data analysis side based on deep learning
Method.
Background technique
Remotely-sensed data source is the object spectrum data different by a series of wavelength of actively or passively remote sensor acquisition.Institute
After the object spectrum data conversion of acquisition is at digital data, GIS-Geographic Information System can be used for.
The scanner of most of road resource satellites is typical spectral scanner, is a kind of linear array device.Scanning
Instrument obtains ground data by shuttle-scanning (such as LANDSAT) or pusher scanning (such as SPOT) mode.These scanners are passive
Ground records the solar spectrum of earth surface reflection, as multispectral scanner data can extract agrotype, distribution and growing way, ground
Shape, soil, the water surface and network of waterways information.And infrared scanner can record the heat radiation data that atural object is emitted, these heat radiation numbers
According to can be used to monitor fire, volcano and geothermal activity.Data acquired in heat radiation remote sensor or LONG WAVE INFRARED remote sensor can be used
To draw ocean temperature and study the changing rule of tide.In land, the variation of plant water content can be obtained by remotely-sensed data
To the delta data of tree crown temperature obtain.The temperature profile data that remotely-sensed data obtains is generally used to monitor city, industry
The distribution situation in area, the production center and farming region.
Existing remotely-sensed data analysis method, one kind as disclosed in Chinese Patent Application No. 201410851849.X are based on master
The Classifying Method in Remote Sensing Image of dynamic deep learning, including step 1: choosing remote sensing image data to be sorted；Step 2: using in advance
The algorithm first configured handles remote sensing image data；Step 3: in Active Learning Algorithm nEQB never marker samples U
Optimal sample B is selected in selection；Step 4: unmarked sample U and optimal sample B being done into subtraction, obtain new unmarked sample
Collect U`, marker samples L and optimal sample B are done into add operation, obtain new marker samples collection L`；Step 5: returning to step
2, above procedure is continued cycling through, until unmarked sample set U` for empty set or meets pre-set termination study index, circulation
Terminate, the matched classification accuracy of output phase and classification results figure.The invention has the benefit that passing through deep learning and active
Study effectively raises the nicety of grading of data the shortcomings that capable of overcoming respectively using unsupervised learning and supervised learning.
But the above method still deposits that classification accuracy is lower and member's model number is affected to classification accuracy
The disadvantages of.
Summary of the invention
Technical problem to be solved by the invention is to provide one kind can further increase classification accuracy, reduces member's mould
The remotely-sensed data analysis method based on deep learning that type number influences classification accuracy.
The technical solution adopted by the present invention to solve the technical problems is: a kind of remotely-sensed data analysis based on deep learning
Method, comprising the following steps:
1) initial data of high-spectrum remote sensing data is read；Data preprocessing operation is carried out, to reduce data dimension；Building
Substantially degree of deeply convinceing network model extracts the feature vector and spatial information of high-spectrum remote sensing data；
2) by the feature vector and spatial information of the high-spectrum remote sensing data extracted in step 1), marker samples are determined；And
Training sample and test sample are selected in marker samples；
3) it is carried out using all training samples of remote sensing image data unsupervised from coding depth e-learning；
4) it is then exercised supervision depth e-learning using pre-set marker samples；
5) classification results are obtained using pre-set unmarked sample input depth network；
6) it uses in the Active Learning image classification method never marker samples sampled based on minimum difference and selects best sample
This；
The Active Learning image classification method based on minimum difference sampling are as follows: be primarily based on adopting again for mark sample set
Sample result constructs decision-making committee, is not then marked using the difference of the probability value of ballot higher 2 classifications of probability to measure
The ballot inconsistency of each sample of sample set, the smallest sample of select probability difference transfer to human expert to mark；Such iteration
Update classifier；
7) unmarked sample U and optimal sample are done into subtraction, new unmarked sample set U` is obtained, by marker samples
L and optimal sample B do add operation, obtain new marker samples collection L`；
8) step 3 is returned to, above procedure is continued cycling through, until unmarked sample set U` is empty set or is met preparatory
The index of the termination study of setting, circulation terminate, the matched classification accuracy of output phase and classification results figure；
9) classification results precision evaluation is carried out.
Further, minimum difference sampling Active Learning image classification method comprising steps of
1) bagging is carried out using n training set of definition to mark sample set L first；
2) structure obtained based on bagging resampling technique, one decision committee member with each member's model of n of building
Meeting；For each sample not marked, its inconsistent metrology structure of voting is calculated by formula (1)；
Wherein, yi* is to not marking sample xiPrediction result, p (yi*=ω | xi) it is to xiIt is predicted as the general of classification ω
Rate, yi are the committees to sample xiThe category set of prediction, ω * are the classifications with maximum probability；
3) sample of most information content is selected according to the following formula；
4) sample class is manually marked, the sample that Active Learning is chosen is marked, and never deletes in marker samples
It removes, is added in training sample, update training sample, obtain final classification device.
The beneficial effects of the present invention are: a kind of remotely-sensed data analysis method based on deep learning of the present invention has
Following advantages:
A kind of remotely-sensed data analysis method based on deep learning of the present invention using based on minimum difference due to adopting
Optimal sample is selected in the Active Learning image classification method never marker samples of sample, thus considering the same of uncertain sampling
When, can effectively improve nicety of grading, can effectively improve classification effectiveness, and can initiative user to those most
Valuable data are marked；In the case where obtaining close accuracy rate, active samples selection can be with relative to random selection
Required sample number is reduced significantly, effectively raises the efficiency to data classification.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the remotely-sensed data analysis method based on deep learning in the embodiment of the present invention.
Specific embodiment
Present invention will be further explained below with reference to the attached drawings and examples.
As shown in Figure 1, remotely-sensed data analysis method of the kind based on deep learning, comprising the following steps:
1) initial data of high-spectrum remote sensing data is read；Data preprocessing operation is carried out, to reduce data dimension；Building
Substantially degree of deeply convinceing network model extracts the feature vector and spatial information of high-spectrum remote sensing data；
2) by the feature vector and spatial information of the high-spectrum remote sensing data extracted in step 1), marker samples are determined；And
Training sample and test sample are selected in marker samples；
3) it is carried out using all training samples of remote sensing image data unsupervised from coding depth e-learning；
4) it is then exercised supervision depth e-learning using pre-set marker samples；
5) classification results are obtained using pre-set unmarked sample input depth network；
6) it uses in the Active Learning image classification method never marker samples sampled based on minimum difference and selects best sample
This；
The Active Learning image classification method based on minimum difference sampling are as follows: be primarily based on adopting again for mark sample set
Sample result constructs decision-making committee, is not then marked using the difference of the probability value of ballot higher 2 classifications of probability to measure
The ballot inconsistency of each sample of sample set, the smallest sample of select probability difference transfer to human expert to mark；Such iteration
Update classifier；
7) unmarked sample U and optimal sample are done into subtraction, new unmarked sample set U` is obtained, by marker samples
L and optimal sample B do add operation, obtain new marker samples collection L`；
8) step 3 is returned to, above procedure is continued cycling through, until unmarked sample set U` is empty set or is met preparatory
The index of the termination study of setting, circulation terminate, the matched classification accuracy of output phase and classification results figure；
9) classification results precision evaluation is carried out.
Specifically, the Active Learning image classification method of minimum difference sampling comprising steps of
1) bagging is carried out using n training set of definition to mark sample set L first；
2) structure obtained based on bagging resampling technique, one decision committee member with each member's model of n of building
Meeting；For each sample not marked, its inconsistent metrology structure of voting is calculated by formula (1)；
Wherein, yi* is to not marking sample xiPrediction result, p (yi*=ω | xi) it is to xiIt is predicted as the general of classification ω
Rate, yi are the committees to sample xiThe category set of prediction, ω * are the classifications with maximum probability；
3) sample of most information content is selected according to the following formula；
4) sample class is manually marked, the sample that Active Learning is chosen is marked, and never deletes in marker samples
It removes, is added in training sample, update training sample, obtain final classification device.
Specifically, unsupervised from coding depth net using all training samples progress of remote sensing image data in step 3)
Network study；It can use sparse unsupervised self-editing to the progress of all training samples from encryption algorithm and support vector machines algorithm
Code depth e-learning.
In conclusion a kind of remotely-sensed data analysis method based on deep learning of the present invention is due to using based on most
Optimal sample is selected in the Active Learning image classification method never marker samples of small difference sampling, thus considering uncertainty
While sampling, nicety of grading can be effectively improved, can effectively improve classification effectiveness, and being capable of initiative user
The data of those most worthies are marked；In the case where obtaining close accuracy rate, active samples selection is relative to random
Selection can reduce required sample number significantly, effectively raise the efficiency to data classification.
Claims (2)
1. a kind of remotely-sensed data analysis method based on deep learning, which comprises the following steps:
1) initial data of high-spectrum remote sensing data is read；Data preprocessing operation is carried out, to reduce data dimension；Building is basic
Degree of deeply convinceing network model extracts the feature vector and spatial information of high-spectrum remote sensing data；
2) by the feature vector and spatial information of the high-spectrum remote sensing data extracted in step 1), marker samples are determined；And it is marking
Training sample and test sample are selected in note sample；
3) it is carried out using all training samples of remote sensing image data unsupervised from coding depth e-learning；
4) it is then exercised supervision depth e-learning using pre-set marker samples；
5) classification results are obtained using pre-set unmarked sample input depth network；
6) it uses in the Active Learning image classification method never marker samples sampled based on minimum difference and selects optimal sample；
The Active Learning image classification method based on minimum difference sampling are as follows: be primarily based on the resampling knot of mark sample set
Fruit constructs decision-making committee, does not then mark sample using the difference of the probability value of ballot higher 2 classifications of probability to measure
Collect the ballot inconsistency of each sample, the smallest sample of select probability difference transfers to human expert to mark；Such iteration updates
Classifier；
7) unmarked sample U and optimal sample are done into subtraction, obtain new unmarked sample set U`, by marker samples L with
Optimal sample does add operation, obtains new marker samples collection L`；
8) step 3 is returned to, above procedure is continued cycling through, until unmarked sample set U` is that empty set or satisfaction are preset
Termination study index, circulation terminates, the matched classification accuracy of output phase and classification results figure；
9) classification results precision evaluation is carried out.
2. a kind of remotely-sensed data analysis method based on deep learning as described in claim 1, it is characterised in that: the minimum
Difference sampling Active Learning image classification method comprising steps of
1) bagging is carried out using n training set of definition to mark sample set L first；
2) structure obtained based on bagging resampling technique, one decision-making committee with each member's model of n of building；Needle
To each sample not marked, its inconsistent metrology structure of voting is calculated by formula (1)；
Wherein, yi* is to not marking sample xiPrediction result, p (yi*=ω | xi) it is to xiIt is predicted as the probability of classification ω,
Yi is the committee to sample xiThe category set of prediction, ω * are the classifications with maximum probability；
3) sample of most information there is selected according to the following formula；
4) sample class is manually marked, the sample that Active Learning is chosen is marked, and never deletes, adds in marker samples
Enter into training sample, updates training sample, obtain final classification device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811260948.5A CN109447151A (en) | 2018-10-26 | 2018-10-26 | A kind of remotely-sensed data analysis method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811260948.5A CN109447151A (en) | 2018-10-26 | 2018-10-26 | A kind of remotely-sensed data analysis method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109447151A true CN109447151A (en) | 2019-03-08 |
Family
ID=65548555
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811260948.5A Pending CN109447151A (en) | 2018-10-26 | 2018-10-26 | A kind of remotely-sensed data analysis method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109447151A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175247A (en) * | 2019-03-13 | 2019-08-27 | 北京邮电大学 | A method of abnormality detection model of the optimization based on deep learning |
| CN111079147A (en) * | 2019-12-17 | 2020-04-28 | 厦门服云信息科技有限公司 | Virus detection method based on active learning, terminal equipment and storage medium |
| CN111369142A (en) * | 2020-03-04 | 2020-07-03 | 中国电子科技集团公司第五十四研究所 | Autonomous remote sensing satellite task generation method |
| CN111414942A (en) * | 2020-03-06 | 2020-07-14 | 重庆邮电大学 | Remote sensing image classification method based on active learning and convolutional neural network |
| CN111460966A (en) * | 2020-03-27 | 2020-07-28 | 中国地质大学（武汉） | Hyperspectral remote sensing image classification method based on metric learning and neighbor enhancement |
| CN112508092A (en) * | 2020-12-03 | 2021-03-16 | 上海云从企业发展有限公司 | Sample screening method, system, equipment and medium |
| CN112906666A (en) * | 2021-04-07 | 2021-06-04 | 中国农业大学 | Remote sensing identification method for agricultural planting structure |
| CN113158855A (en) * | 2021-04-08 | 2021-07-23 | 成都国星宇航科技有限公司 | Remote sensing image auxiliary processing method and device based on online learning |
| CN113470127A (en) * | 2021-09-06 | 2021-10-01 | 成都国星宇航科技有限公司 | Optical image effective compression method based on satellite-borne cloud detection |
| CN115730254A (en) * | 2022-12-06 | 2023-03-03 | 中电金信软件有限公司 | Method and device for expanding modeling sample data label |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
| CN104102929A (en) * | 2014-07-25 | 2014-10-15 | 哈尔滨工业大学 | Hyperspectral remote sensing data classification method based on deep learning |
| CN104484682A (en) * | 2014-12-31 | 2015-04-01 | 中国科学院遥感与数字地球研究所 | Remote sensing image classification method based on active deep learning |
| CN106056157A (en) * | 2016-06-01 | 2016-10-26 | 西北大学 | Hyperspectral image semi-supervised classification method based on space-spectral information |
-
2018
- 2018-10-26 CN CN201811260948.5A patent/CN109447151A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
| CN104102929A (en) * | 2014-07-25 | 2014-10-15 | 哈尔滨工业大学 | Hyperspectral remote sensing data classification method based on deep learning |
| CN104484682A (en) * | 2014-12-31 | 2015-04-01 | 中国科学院遥感与数字地球研究所 | Remote sensing image classification method based on active deep learning |
| CN106056157A (en) * | 2016-06-01 | 2016-10-26 | 西北大学 | Hyperspectral image semi-supervised classification method based on space-spectral information |
Non-Patent Citations (2)
| Title |
|---|
| 吴健等: "最小差异采样的主动学习图像分类方法", 《通信学报》 * |
| 王立国等: "融合主动学习的高光谱图像半监督分类", 《哈尔滨工程大学学报》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175247A (en) * | 2019-03-13 | 2019-08-27 | 北京邮电大学 | A method of abnormality detection model of the optimization based on deep learning |
| CN110175247B (en) * | 2019-03-13 | 2021-06-08 | 北京邮电大学 | Method for optimizing anomaly detection model based on deep learning |
| CN111079147A (en) * | 2019-12-17 | 2020-04-28 | 厦门服云信息科技有限公司 | Virus detection method based on active learning, terminal equipment and storage medium |
| CN111369142A (en) * | 2020-03-04 | 2020-07-03 | 中国电子科技集团公司第五十四研究所 | Autonomous remote sensing satellite task generation method |
| CN111369142B (en) * | 2020-03-04 | 2023-04-18 | 中国电子科技集团公司第五十四研究所 | Autonomous remote sensing satellite task generation method |
| CN111414942B (en) * | 2020-03-06 | 2022-05-03 | 重庆邮电大学 | Remote sensing image classification method based on active learning and convolutional neural network |
| CN111414942A (en) * | 2020-03-06 | 2020-07-14 | 重庆邮电大学 | Remote sensing image classification method based on active learning and convolutional neural network |
| CN111460966A (en) * | 2020-03-27 | 2020-07-28 | 中国地质大学（武汉） | Hyperspectral remote sensing image classification method based on metric learning and neighbor enhancement |
| CN111460966B (en) * | 2020-03-27 | 2024-02-02 | 中国地质大学（武汉） | Hyperspectral remote sensing image classification method based on metric learning and neighbor enhancement |
| CN112508092A (en) * | 2020-12-03 | 2021-03-16 | 上海云从企业发展有限公司 | Sample screening method, system, equipment and medium |
| CN112906666A (en) * | 2021-04-07 | 2021-06-04 | 中国农业大学 | Remote sensing identification method for agricultural planting structure |
| CN113158855A (en) * | 2021-04-08 | 2021-07-23 | 成都国星宇航科技有限公司 | Remote sensing image auxiliary processing method and device based on online learning |
| CN113470127A (en) * | 2021-09-06 | 2021-10-01 | 成都国星宇航科技有限公司 | Optical image effective compression method based on satellite-borne cloud detection |
| CN115730254A (en) * | 2022-12-06 | 2023-03-03 | 中电金信软件有限公司 | Method and device for expanding modeling sample data label |
| CN115730254B (en) * | 2022-12-06 | 2023-10-13 | 中电金信软件有限公司 | Method and device for expanding modeling sample data label |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109447151A (en) | A kind of remotely-sensed data analysis method based on deep learning | |
| Zhong et al. | Deep learning based multi-temporal crop classification | |
| Hargrove et al. | Potential of multivariate quantitative methods for delineation and visualization of ecoregions | |
| Koukoulas et al. | Mapping individual tree location, height and species in broadleaved deciduous forest using airborne LIDAR and multi‐spectral remotely sensed data | |
| Ha et al. | A review of downscaling methods for remote sensing-based irrigation management: Part I | |
| Reis et al. | Identification of hazelnut fields using spectral and Gabor textural features | |
| CN111598045B (en) | Remote sensing farmland change detection method based on object spectrum and mixed spectrum | |
| CN110399819A (en) | A kind of remote sensing image residential block extraction method based on deep learning | |
| Puente et al. | A genetic programming approach to estimate vegetation cover in the context of soil erosion assessment | |
| Kavzoglu et al. | Performance analysis of maximum likelihood and artificial neural network classifiers for training sets with mixed pixels | |
| CN106600595A (en) | Human body characteristic dimension automatic measuring method based on artificial intelligence algorithm | |
| CN112836725A (en) | Weak supervision LSTM recurrent neural network rice field identification method based on time sequence remote sensing data | |
| Holmgren et al. | Tree crown segmentation based on a geometric tree crown model for prediction of forest variables | |
| CN116580318A (en) | Soil nutrient inversion method, device, equipment and medium for crop coverage area | |
| Myint et al. | Examining lacunarity approaches in comparison with fractal and spatial autocorrelation techniques for urban mapping | |
| Meng et al. | A new digital soil mapping method with temporal-spatial-spectral information derived from multi-source satellite images | |
| Dahiya et al. | A comparative analysis of different land-use and land-cover classifiers using hyperspectral data | |
| Rasheed et al. | A framework base on deep neural network (DNN) for land use land cover (LULC) and rice crop classification without using survey data | |
| CN112508060A (en) | Landslide mass state judgment method and system based on graph convolution neural network | |
| Kaur et al. | Remote Image Classification Using Particle Swarm Optimization | |
| Jenicka | Land Cover Classification of Remotely Sensed Images | |
| Dahiya et al. | Quantitative analysis of different land-use and land-cover classifiers using hyperspectral dataset | |
| Shanmuganathan | A hybrid artificial neural network (ANN) approach to spatial and non-spatial attribute data mining: A case study experience | |
| Vibha et al. | Robust technique for segmentation and counting of trees from remotely sensed data | |
| Bigdeli et al. | DTM generation from the point cloud using a progressive geodesic morphology and a modified Particle Swarm Optimization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190308 |