CN110321528A - A kind of Hyperspectral imaging heavy metal-polluted soil concentration evaluation method based on semi-supervised geographical space regression analysis - Google Patents
A kind of Hyperspectral imaging heavy metal-polluted soil concentration evaluation method based on semi-supervised geographical space regression analysis Download PDFInfo
- Publication number
- CN110321528A CN110321528A CN201910624568.3A CN201910624568A CN110321528A CN 110321528 A CN110321528 A CN 110321528A CN 201910624568 A CN201910624568 A CN 201910624568A CN 110321528 A CN110321528 A CN 110321528A
- Authority
- CN
- China
- Prior art keywords
- model
- heavy metal
- supervised
- semi
- geographical space
- 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
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Operations Research (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Immunology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Algebra (AREA)
- Spectroscopy & Molecular Physics (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biochemistry (AREA)
- Image Analysis (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention belongs to soil environment monitorings and evaluation areas, and in particular to the Hyperspectral imaging heavy metal-polluted soil concentration evaluation based on semi-supervised geographical space regression analysis.The method establishes model between heavy metal concentration and imaging airborne-remote sensing using semi-supervised geographical space regression analysis;The semi-supervised geographical space regression analysis is the method combined using co-training Semi-Supervised Regression method and geographical space homing method.The present invention establishes model by the semi-supervised method of co-training and carries out the estimation of heavy metal-polluted soil concentration, realization has label data information and the comprehensive utilization without label data information, part unlabeled exemplars exemplar has been converted into, greatly it is augmented with exemplar collection quantity, so that the comprehensive heavy metal distribution character of model and a large amount of unlabeled exemplars that final training obtains are that assessment models establish the studying space provided, so that the accuracy and reliability of model be made to get a promotion.
Description
Technical field
The present invention relates to heavy metal in soil concentration prediction fields, and in particular to one kind is returned based on semi-supervised geographical space
The Hyperspectral imaging heavy metal-polluted soil concentration evaluation method of analysis.
Background technique
In recent years, China's problem of environmental pollution continues seriously, in order to more efficiently carry out to heavy metal in soil concentration
Monitoring and evaluation is monitored heavy metal-polluted soil concentration by new technologies necessary.Conventional soil heavy metal concentration
Monitoring method is that field soil sampling carries out laboratory assay measurement, and this method excessively takes time and effort, and the soil obtained
Heavy metal is dotted information, confidence level higher planar information is unable to get through geography spatial interpolation methods, no
Geographical space continuous in region can be analyzed and determined.
Common heavy metal content in soil estimation modeling algorithm has Partial Least Squares Regression (PLS), multiple linear regression
(MLR) etc..Wherein offset minimum binary is most popular algorithm in heavy metal content in soil estimation modeling.Currently, soil weight
Metal correlative study is more to be analyzed for ground experiment room spectrum, and carries out heavy metal-polluted soil based on imaging high-spectrum remote-sensing
Content estimation research is few, the reason is that aerial imagery high-spectral data feature itself and ground experiment room spectroscopic data property difference
It is larger, prevent analysis and method that indoor spectral obtains are studied from simply migrating to imaging EO-1 hyperion, correlative study and analysis
The lab analysis stage is remained in, no matter from aviation or space flight EO-1 hyperion, there is no realize large scale geographical space model
The application for enclosing the estimation of heavy metal-polluted soil concentration, estimates to limit high-spectrum remote-sensing to a certain degree in heavy metal content in soil
It calculates and the application in drawing.
It is retrieved, relevant application case exists in the prior art, such as Chinese Patent Application No. 201710900422.8, application
Day is that the application case on the 28th of September in 2017 is disclosed based on unmanned plane EO-1 hyperion inverting heavy metal in soil pollution monitoring method,
Specific step is as follows for its monitoring method: spot sampling;Sample preprocessing;Use the weight of x-ray fluorescence analyzer collecting sample
The content of the main research element in metallic pollution source;Utilize the laboratory hyper spectral reflectance of field spectroradiometer collecting sample;It is right
Original spectrum reflectivity data carries out data processing respectively;To the master for using Partial Least Squares Regression algorithm that will measure respectively
The content for studying element is micro- with laboratory EO-1 hyperion primary reflection rate data, inverse, logarithm, first differential and second order respectively
Divided data carry out correlation analysis simultaneously verifying optimization is carried out to model, obtain optimal transform method, using carry EO-1 hyperion at
As unmanned plane acquisition research area's hyper spectral reflectance data of spectrometer are as testing data, large area inverting content of beary metal.
Data analysis is much not achieved in the accuracy for the heavy metal-polluted soil concentration that X ray fluorescence analysis instrument is measured in the above method
Required precision, meanwhile, although this method can serve the drawing of heavy metal-polluted soil spatial distribution to a certain extent, it is based on rotor
Unmanned plane high-spectral data acquisition capability it is lower, can not widespread adoption;And its data analysing method is excessively traditional, does not also give
Case verification result out.There are non-linear space relationship between the characteristic wave bands of heavy metal, traditional deflected secondary air exists
It is the modeling process using Numerical accuracy as judgement when to the estimation of heavy metal-polluted soil concentration, ignores the most basic ground of geography
The problem of spatial continuity of object feature, does not have space characteristics.Therefore generated space when a wide range of survey region is faced
Heterogeneity cannot effectively overcome.
Therefore, it based on the defect of the prior art, needs to invent one kind and can effectively overcome and a wide range of survey region is predicted
When generated special heterogeneity problem, to improve the modeling method of model accuracy and reliability.
Summary of the invention
1. to solve the problems, such as
For existing Detection Techniques when estimating heavy metal-polluted soil concentration, founding mathematical models are considered merely and are ignored
The problem of spatial continuity of the most basic characters of ground object of geography, space when a wide range of survey region cannot be overcome to predict are different
Matter problem, the present invention establish model by the semi-supervised method of co-training and carry out the estimation of heavy metal-polluted soil concentration, and realization has
Label and the comprehensive utilization without label data, have been converted to label data without label data for part, have greatly been augmented with label sample
This collection quantity, so that the comprehensive heavy metal spatial distribution attribute of model and a large amount of unlabeled exemplars that final training obtains provide
Space is practised, so that the accuracy and reliability of model be made to get a promotion.
2. technical solution
To solve the above-mentioned problems, the technical solution adopted in the present invention is as follows:
The present invention provides a kind of Hyperspectral imaging heavy metal-polluted soil concentration based on semi-supervised geographical space regression analysis
Appraisal procedure, the method is using semi-supervised geographical space regression analysis in heavy metal concentration and imaging Hyperspectral imaging number
Model is established between;The semi-supervised geographical space regression analysis is to utilize co-training Semi-Supervised Regression method
The method combined with geographical space homing method.
As further improvement of the present invention, the semi-supervised geographical space regression analysis establishes the specific of model
Step are as follows:
2-1) by heavy metal concentration data imaging airborne-remote sensing corresponding with sampled point and sampled point geographical coordinate
Data combination, is formed with label data collection;By near sampled point spectrum image data and and its corresponding pixel at geographical sit
Data screening combination is marked, is formed without label data collection;
To 2-2) have label data collection to be divided into has label training dataset and has label Verification data set, described to have label instruction
Practice data set and is used for model training, it is described to have label Verification data set for proof-tested in model precision;
2-3) establish model: setting geography Spatial Regression Model I and model II, respectively to the data of label training dataset
Subset A and B are trained, and establish rudimentary model, are recycled co-training training and are established circulation, realize two models
It learns from each other, selects final mask using proof-tested in model precision result.
As further improvement of the present invention, the described method comprises the following steps:
Soil sample 3-1) is acquired, the heavy metal concentration of soil sample is measured;
It 3-2) obtains the imaging spectrometer image data in research area and realizes pretreatment;
3-3) model is established using semi-supervised geographical space regression analysis;
3-4) by the pretreated imaging spectrometer image data input model of 3-2), the estimation of heavy metal-polluted soil concentration is obtained
Figure.
As further improvement of the present invention, the step 2-3) specifically:
S1) by setting initial parameter, two groups of different geographical space regression models, model I and model II are generated, simultaneously
Two groups of data subset A and B having in label training dataset are randomly choosed, by model I corresponding data subset A, II pairs of model
Answer subset B;
S2) its corresponding subset is trained respectively using model I and model II, establishes rudimentary model;
S3) carry out co-training training and establish repeatedly circulation: will have label data collection and without label data collection according to
Different proportion stochastic inputs model I and model II realize two to the confidence level that no label data is predicted according to model I and model II
A model is learnt from each other;
S4 the accuracy test for having label Verification data set to carry out model I and model II, selection) are utilized after circulation terminates every time
The good model of precision.
As further improvement of the present invention, the process for establishing model specifically: the step S1) in, two groups
The initial parameter of the setting of model is different, the S3) in two models learn from each other that detailed process is as follows: as model I
When realizing prediction to some unlabeled exemplars i, and achieving higher confidence level, sample i input model II is carried out pre-
Its confidence level is surveyed and assesses, when model I and model II obtains high confidence to some sample, by sample i from no label
It is deleted in data set, has been put into label training dataset.
As further improvement of the present invention, the step S1) in data subset A and B sample size be no more than
Total have exemplar quantity 1/4.
As further improvement of the present invention, the step S3) described in have label data collection and no label data collection
Ratio be 1:1,1:3,1:5,1:8.
As further improvement of the present invention, the no label data collection include near sampled point 10 pixels away from
With a distance from, 30 pixels and 50 distance lengths are the spectrum image data and geographic coordinate data in the buffer area of radius.
As further improvement of the present invention, the heavy metal includes As and Cr.
3. beneficial effect
Compared with the prior art, the invention has the benefit that
(1) the Hyperspectral imaging heavy metal-polluted soil concentration evaluation side of the invention based on semi-supervised geographical space regression analysis
Semi-supervised geographical space regression analysis is applied to imaging Hyperspectral imaging heavy metal-polluted soil and predicts field, to imaging by method
The minute quantity that high-spectral data possesses has the characteristic spectrum of exemplar (heavy metal-polluted soil concentration) and huge amount unlabeled exemplars
Complex linear relationship between wave band is fitted, and should utilizing in the process First Law of Geography, (geographical object or attribute are in sky
Between it is related each other in distribution), in conjunction with the spatial position attribute of sample, establish based on heavy metal concentration spatial autocorrelation constraint, same
When the heavy metal-polluted soil concentration evaluation model of study is optimized in conjunction with no label data, which has excellent capability of fitting
With transfer ability, the high-precision forecast of the heavy metal concentration spatial distribution in a wide range of geographical space may be implemented.
(2) the Hyperspectral imaging heavy metal-polluted soil concentration evaluation side of the invention based on semi-supervised geographical space regression analysis
Method, by the organic knot of geographical space homing method in the classical semi-supervised learning model in machine learning field and Geographical Analysis
It closes, both allows and to realize having complementary advantages, can overcome traditional using Numerical accuracy as the modeling process of judgement, comprehensively consider geography
Space Elements, while having label and the comprehensive utilization without label data by the semi-supervised method realization of co-training, by part
No label data has been converted to label data, is greatly augmented with exemplar collection quantity, provides study for a large amount of unlabeled exemplars
Space, therefore the accuracy and reliability of the finally obtained model of method of the invention is more excellent.
(3) the Hyperspectral imaging heavy metal-polluted soil concentration evaluation side of the invention based on semi-supervised geographical space regression analysis
Method, with traditional supporting vector machine model (SVM), geographical space regression model (GWR) and nearest neighbour analysis method (k-NN)
It is compared as assessment models, according to the overall merit (R of validation data set2And RMSE) as a result, assessment models of the invention have
Higher precision.
Detailed description of the invention
Fig. 1 is the basic schematic diagram that semi-supervised geographical space of the invention returns;
Fig. 2 is that the Yitong Jilin of the method for the present invention analysis prediction studies area's heavy metal-polluted soil As concentration evaluation figure;
Fig. 3 is that the Yitong Jilin of the method for the present invention analysis prediction studies area's heavy metal-polluted soil Cr concentration evaluation figure.
Specific embodiment
Embodiment
The present invention is further described below combined with specific embodiments below.
The present embodiment, which is described in detail to return based on semi-supervised geographical space by taking her Tong County research area, Jilin Province as an example, to be divided
The application of the Hyperspectral imaging heavy metal-polluted soil concentration evaluation method of analysis is respectively adopted following methods and carries out research area's soil huge sum of money
Belong to the concentration mensuration of As.
1) research area's soil sample acquisition: research area selects the blackland region in China, heavy industry area, northeast, is studying
The uniformly distributed sampled point of checkerboard type method is pressed in area, the determination of each sampling point position needs to combine image spatial resolution and adopt
Topography and geomorphology at sampling point comprehensively considers, and selectively the relatively simple region of Table Properties determines soil sampling point position to Ying Jinliang,
Corresponding spectrum at sampled point is accurately obtained in image in this way convenient for the later period;Specific sampling includes: according to research area's topography and geomorphology
And the area that surface soil attribute is consistent is greater than determining sample in the region of 3*3 image spatial resolution unit
It sets;Every place's sampled point acquires the soil sample of 4~6 earth's surface 2cm thickness according to quincuncial pile formula method;Research area is total to collecting sample
95;Coordinate record is determined by the way that real time dynamic differential localization method (RTK) is accurate at each sampled point, is passed through in practical operation
China surveys RTK base station-movement station mode and realizes, base station is erected at research area's height above sea level highest point mountain top, and mountain top have no occluder.
2) the elements laboratory chemical such as heavy metal concentration and organic matter measures in soil sample: passing through inductively coupled plasma
Constitution spectrum (ICP-MS) method and national standard operation carry out chemical examination measurement to the concentration of heavy metal-polluted soil As and Cr;
3) research area's imaging spectrometer data is obtained using airborne platform and imaging spectrometer: passing through lens focus, row
High setting, flight course planning etc., adjustment imaging spectrometer system fetched data spatial resolution, imaging spectrometer data analysis space
The too low then mixed pixel problem of resolution ratio is serious, is unfavorable for later period modeling analysis, by adjusting lens focus and airborne platform
It is 4.5m that flight flying height, which determines that the present invention uses data spatial resolution,;Imaging spectral flying quality obtain same day 10:30 extremely
It carries out, is required with guaranteeing that imaging data spectral radiance quality meets analysis, flight band is set by 30% sidelapping during 14:00
It sets, it is final to obtain research 8 strip datas of area's imaging spectral;
Airborne Hyperspectral data prediction, including geometric correction, radiation calibration, atmospheric correction and strips mosaic etc.;Aviation
The data that airborne platform obtains successively are converted imaging space coordinate system by airborne direction and location system without geospatial coordinates
Geometric correction is completed to geospatial coordinates system, so that each image picture element has geospatial coordinates;In geometric correction
On the basis of by imaging spectrometer obtain digital signal, according to radiation calibration Parameter Switch be entrance pupil at spectral radiance, make
Obtaining image data has spectrum physics meaning;The meteorological related data obtained again by early period, passes through Mondtran atmospheric radiation
Mode completes atmospheric correction;After pretreatment, initial data is converted to research area imaging reflectivity data;
4) modeling of heavy metal-polluted soil concentration optical spectrum is carried out using semi-supervised geographical space regression analysis, obtains fitting energy
Power and the strong heavy metal-polluted soil appraising model of transfer ability:
4-1) by spectroscopic data, the spectroscopic data of sampling point position corresponding position in sampled point heavy metal concentration data and image
The coordinate data of position combines, and label data collection is formed with, by 10 pixel radiuses, 20 pictures near sampled point
Screening in first radius and 30 pixel radius buffer areas without the coordinate data of label spectroscopic data and its corresponding position is combined
It is formed without label data collection, entire data collection is by having label data collection and constituting without label data collection, each sample data
There is the spatial coordinated information of every spectrum position;To have label data collection according to 2:1 ratio cut partition be have label instruction
Practice data set and has label Verification data set;
4-2) heavy metal concentration value estimation steps, the step specifically:
A) spectrum characteristic selection: according to pearson correlation analysis, selection analysis step 4) has label data collection
The wave band quantity of the middle higher wave band of correlation, selection is no more than 5;
B) Spatial weight matrix is generated: according to heavy metal-polluted soil concentration data is measured in step 2), according to every kind of soil weight
Concentration of the metal at sampling point position establishes Spatial weight matrix, the space of matrices resolution ratio and Airborne Hyperspectral image data
It is identical
C) semi-supervised geographical space regression analysis uses co-training training pattern
Semi-supervised geographical space regression analysis is by the classical semi-supervised learning model in machine learning field and geographical credit
Geographical space homing method in analysis organically combines, and allows the two to realize and has complementary advantages.Concrete principle simplified summary are as follows: be based on co-
2 geographical space regression models of training under training semi-supervised learning frame, are realized in geographical space and semi-supervised learning 2
The modeling of a dimension.The present invention is in co-training training pattern, 2 geographical space regression models of training, and 2 geographical empty
Between regression model initial parameter it is different.Specific training process as shown in Figure 1, specific steps are as follows:
S1: 2 groups of random selection has label training dataset data subset A and B, sample size in the data subset A and B
It is no more than total have exemplar quantity 1/4.
S2: being randomly provided 2 geographical space regression models, model I and model II, and initial parameter setting is different
(and differing greatly), its object is to realize the observational learning of Different Cognitive dimension, so that more comprehensive learning outcome is obtained,
Model I and model II respectively corresponds data subset A and B first and is trained, establish rudimentary model, model I and model II and
The Spatial weight matrix established in step b) is used in the later period specific training process of all models;
S3: starting co-training training, vertical 500 circulations of building together, according to having label data collection and no label data collection
Ratio be 1:1,1:3,1:5,1:8, by the model I and model II in no label data collection stochastic inputs step S2;Described
Model I or model II can be realized the prediction to unlabeled exemplars, and the model I and model II mutually learn;Process
It is as follows: as model I realizes prediction to certain unlabeled exemplars i, and to achieve higher confidence level, at this time input i sample
Progressive die type II is predicted and is assessed its confidence level, will when model I and model II obtains high confidence to some sample
I sample is concentrated from no label data and is deleted, and label training dataset has been put into;Increase when not obtaining expected confidence level and changes
Generation number comes back to no label data collection, every time after circulation terminates using there is label Verification data set to carry out model I and model
The accuracy test of II, while having label training data that sample size is concentrated to be increased;There is label training data to concentrate sample number
The increase of amount can significantly strengthen model learning effect, further increase the precision for finally obtaining model;
S4: until circulation terminates, label training dataset sample is effectively enhanced by step S3, model I and mould
The study predictive ability of type II increases mutually, finally in model I and model II selection to have label Verification data set performance most
Good model is as final assessment models.
Using R2With RMSE as model accuracy evaluation index.R2The referred to as coefficient of determination, R2Value is between 0~1, R2More
Illustrate that models fitting precision is higher close to 1, wherein reaching 0.5 or more then illustrates that models fitting precision is higher and has certain
It is credible;Reach 0.6 or more and then illustrates that credibility is strong.
Fig. 2 is to study area's heavy metal-polluted soil As concentration profile using the Yitong Jilin of the method for the present invention analysis prediction;Fig. 3
To study area's heavy metal-polluted soil Cr concentration profile using the Yitong Jilin of the method for the present invention analysis prediction.According to fig. 2, Fig. 3 can
Know, the result trend one that Prediction of Soil Heavy Metal concentration distribution trend of the present invention and the traditional interpolation fitting method in sample region obtain
It causes, it was demonstrated that the reliability of this method;Compared with existing model prediction method, prediction result of the invention is more in line with soil weight
The distribution character of metal concentration, and heavy metal-polluted soil can be fast implemented by Airborne Hyperspectral image data
The continuously distributed assessment of a wide range of geographical space, more efficiently.
Comparative example 1
This comparative example is substantially identical as embodiment, the difference is that: using k-NN as base learner side in step c
Method carries out the modeling of heavy metal-polluted soil concentration optical spectrum.The reason of being modeled using k-NN is the learning strategy of co-training
Be to multiple view mode of learning it is a kind of it is classical realize, it is higher in semi-supervised research field influence power, mode of learning it is specific
Realization is different for the selection of base learner, and researcher has different strategies, and it is k-NN that selection is the most universal and classics.
Comparative example 2
This comparative example is substantially identical as comparative example 1, the difference is that: support vector machines (SVM) is used in step 4)
Method carries out the modeling of heavy metal-polluted soil concentration optical spectrum, and SVM method needs to adjust ginseng.Compared with embodiment and comparative example 1, the party
The non-Semi-Supervised Regression learning method of method, SVM method are classical statistical machine learning methods, and predictive ability is strong.
Comparative example 3
This comparative example is substantially identical as comparative example 1, the difference is that: geographical space homing method is used in step 4)
Carry out the modeling of heavy metal-polluted soil concentration optical spectrum.
Semi-supervised geographical space regression analysis is referred to as co-training (GWR);Semi-supervised nearest neighbour analysis method
Referred to as co-training (k-NN);Geographical space homing method is referred to as GWR;The Hyperspectral imaging soil weight of different models
Metal appraising model precision evaluation comparison is as shown in table 1.
The Hyperspectral imaging heavy metal-polluted soil appraising model precision evaluation of the different models of table 1
According to verify data overall merit (R2And RMSE) as a result, method of the invention and existing SVM and GWR model are pre-
Survey method is compared, and has highest accuracy value.And method of the invention and co-training of the k-NN method as base learner
Model is compared, and higher accuracy value is also embodied.Since in classification problem, the core concept of k-NN algorithm is that a sample exists
Most of in k in feature space most adjacent samples belong to some classification, then the sample also belongs to this classification.k-
NN method is only related with minimal amount of adjacent sample in classification decision.It therefore is then around neighbour's sample on regression problem
This average value or distance weighting combined value.Therefore compared with co-training method of the invention, classical k-NN method
It cannot preferably realize regression analysis of the heavy metal-polluted soil concentration based on geographical space.
Claims (9)
1. a kind of Hyperspectral imaging heavy metal-polluted soil concentration evaluation method based on semi-supervised geographical space regression analysis, feature
Be: the method using semi-supervised geographical space regression analysis heavy metal concentration and imaging airborne-remote sensing it
Between establish model;The semi-supervised geographical space regression analysis is to utilize co-training Semi-Supervised Regression method and ground
The method that reason space homing method combines.
2. the Hyperspectral imaging heavy metal-polluted soil concentration according to claim 1 based on semi-supervised geographical space regression analysis
Appraisal procedure, it is characterised in that: the semi-supervised geographical space regression analysis establishes the specific steps of model are as follows:
2-1) by heavy metal concentration data imaging airborne-remote sensing corresponding with sampled point and sampled point geographic coordinate data
Combination, is formed with label data collection;By near sampled point spectrum image data and and its corresponding pixel at geographical coordinate number
It combines, is formed without label data collection according to screening;
To 2-2) have label data collection to be divided into has label training dataset and has label Verification data set, described to have label training number
It is used for model training according to collection, it is described to have label Verification data set for proof-tested in model precision;
2-3) establish model: two groups of geographical space regression models of setting, respectively model I and model II, respectively to label training
The data subset A and B of data set are trained, and establish rudimentary model, are recycled co-training training and are established circulation, real
Existing two models are learnt from each other, and select final mask using proof-tested in model precision result.
3. the Hyperspectral imaging heavy metal-polluted soil according to claim 1 or 2 based on semi-supervised geographical space regression analysis
Concentration evaluation method, it is characterised in that: the described method comprises the following steps:
Soil sample 3-1) is acquired, the heavy metal concentration of soil sample is measured;
It 3-2) obtains the imaging spectrometer image data in research area and realizes pretreatment;
3-3) model is established using semi-supervised geographical space regression analysis;
3-4) by the pretreated imaging spectrometer image data input model of 3-2), heavy metal-polluted soil concentration estimation figure is obtained.
4. the Hyperspectral imaging heavy metal-polluted soil concentration according to claim 1 based on semi-supervised geographical space regression analysis
Appraisal procedure, it is characterised in that: the step 2-3) specifically:
S1) by setting initial parameter, two groups of different geographical space regression models, model I and model II are generated, while random
Two groups of data subset A and B having in label training dataset are selected, by model I corresponding data subset A, the corresponding son of model II
Collect B;
S2) its corresponding subset is trained respectively using model I and model II, establishes rudimentary model;
S3 it) carries out co-training training and establishes repeatedly circulation: will have label data collection and without label data collection according to difference
Ratio stochastic inputs model I and model II realize two moulds to the confidence level that no label data is predicted according to model I and model II
Type is learnt from each other;
S4 the accuracy test for having label Verification data set to carry out model I and model II, choice accuracy) are utilized after circulation terminates every time
Good model.
5. the Hyperspectral imaging heavy metal-polluted soil concentration according to claim 4 based on semi-supervised geographical space regression analysis
Appraisal procedure, it is characterised in that: the process for establishing model specifically:
The step S1) in, the initial parameter of the setting of two group models is different, the S3) in two models learn from each other
Detailed process is as follows:, will when model I realizes prediction to some unlabeled exemplars i, and achieves higher confidence level
Sample i input model II is predicted and is assessed its confidence level, when model I and model II obtains higher confidence to some sample
When spending, sample i is concentrated from no label data and is deleted, label training dataset has been put into.
6. the Hyperspectral imaging heavy metal-polluted soil concentration according to claim 5 based on semi-supervised geographical space regression analysis
Appraisal procedure, it is characterised in that: the step S1) in data subset A and B sample size be no more than and total have exemplar
The 1/4 of quantity.
7. the Hyperspectral imaging heavy metal-polluted soil according to claim 4 or 5 based on semi-supervised geographical space regression analysis
Concentration evaluation method, it is characterised in that: the step S3) described in there is the ratio of label data collection and no label data collection to be 1:
1、1:3、1:5、1:8。
8. the Hyperspectral imaging heavy metal-polluted soil concentration according to claim 7 based on semi-supervised geographical space regression analysis
Appraisal procedure, it is characterised in that: the no label data collection includes 10 pixel distances, 30 pixel distances near sampled point
It is the spectrum image data and geographic coordinate data in the buffer area of radius with 50 distance lengths.
9. the Hyperspectral imaging heavy metal-polluted soil concentration according to claim 8 based on semi-supervised geographical space regression analysis
Appraisal procedure, it is characterised in that: the heavy metal includes As and Cr.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910624568.3A CN110321528B (en) | 2019-07-11 | 2019-07-11 | Hyperspectral image soil heavy metal concentration assessment method based on semi-supervised geospatial regression analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910624568.3A CN110321528B (en) | 2019-07-11 | 2019-07-11 | Hyperspectral image soil heavy metal concentration assessment method based on semi-supervised geospatial regression analysis |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110321528A true CN110321528A (en) | 2019-10-11 |
CN110321528B CN110321528B (en) | 2022-11-11 |
Family
ID=68121932
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910624568.3A Active CN110321528B (en) | 2019-07-11 | 2019-07-11 | Hyperspectral image soil heavy metal concentration assessment method based on semi-supervised geospatial regression analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110321528B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112749841A (en) * | 2020-12-30 | 2021-05-04 | 科大国创云网科技有限公司 | User public praise prediction method and system based on self-training learning |
CN114782801A (en) * | 2022-03-01 | 2022-07-22 | 泰瑞数创科技(北京)有限公司 | Machine learning-based automatic control point layout method and device |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102129571A (en) * | 2011-01-31 | 2011-07-20 | 重庆大学 | Method for classifying multi-spectral remote sensing data land use based on semi-supervisor manifold learning |
US20140085133A1 (en) * | 2012-09-27 | 2014-03-27 | Magnetrol International, Incorporated | Time domain reflectometry based method for emulsion detection and profiling |
CN104764868A (en) * | 2015-04-02 | 2015-07-08 | 中国科学院南京土壤研究所 | Soil organic carbon predication method based on geographically weighted regression |
WO2017158438A2 (en) * | 2016-03-15 | 2017-09-21 | Ipq Pty Ltd | Systems and methods for imputing groundwater usage in a groundwater basin |
CN107421894A (en) * | 2017-09-28 | 2017-12-01 | 威海五洲卫星导航科技有限公司 | Based on unmanned plane EO-1 hyperion inverting heavy metal in soil pollution monitoring method |
CN107729293A (en) * | 2017-09-27 | 2018-02-23 | 中南大学 | A kind of geographical space method for detecting abnormal based on Multivariate adaptive regression splines |
CN109784392A (en) * | 2019-01-07 | 2019-05-21 | 华南理工大学 | A kind of high spectrum image semisupervised classification method based on comprehensive confidence |
-
2019
- 2019-07-11 CN CN201910624568.3A patent/CN110321528B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102129571A (en) * | 2011-01-31 | 2011-07-20 | 重庆大学 | Method for classifying multi-spectral remote sensing data land use based on semi-supervisor manifold learning |
US20140085133A1 (en) * | 2012-09-27 | 2014-03-27 | Magnetrol International, Incorporated | Time domain reflectometry based method for emulsion detection and profiling |
CN104764868A (en) * | 2015-04-02 | 2015-07-08 | 中国科学院南京土壤研究所 | Soil organic carbon predication method based on geographically weighted regression |
WO2017158438A2 (en) * | 2016-03-15 | 2017-09-21 | Ipq Pty Ltd | Systems and methods for imputing groundwater usage in a groundwater basin |
CN107729293A (en) * | 2017-09-27 | 2018-02-23 | 中南大学 | A kind of geographical space method for detecting abnormal based on Multivariate adaptive regression splines |
CN107421894A (en) * | 2017-09-28 | 2017-12-01 | 威海五洲卫星导航科技有限公司 | Based on unmanned plane EO-1 hyperion inverting heavy metal in soil pollution monitoring method |
CN109784392A (en) * | 2019-01-07 | 2019-05-21 | 华南理工大学 | A kind of high spectrum image semisupervised classification method based on comprehensive confidence |
Non-Patent Citations (4)
Title |
---|
KUN TAN等: "Random forest–based estimation of heavy metal concentration in agricultural soils with hyperspectral sensor data", 《ENVIRON MONIT ASSESS》 * |
YI YANG等: "An Extended Semi-Supervised Regression Approach with Co-Training and Geographical Weighted Regression: A Case Study of Housing Prices in Beijing", 《ISPRS INTERNATIONAL JOURNAL OF FEO-INFORMATION》 * |
江振蓝等: "GWR模型在土壤重金属高光谱预测中的应用", 《地理学报》 * |
马伟波: "基于HyMAP-C航空高光谱影像土壤重金属浓度估算", 《中国优秀硕士学位论文全文数据库 工程科技I辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112749841A (en) * | 2020-12-30 | 2021-05-04 | 科大国创云网科技有限公司 | User public praise prediction method and system based on self-training learning |
CN114782801A (en) * | 2022-03-01 | 2022-07-22 | 泰瑞数创科技(北京)有限公司 | Machine learning-based automatic control point layout method and device |
Also Published As
Publication number | Publication date |
---|---|
CN110321528B (en) | 2022-11-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110174359A (en) | A kind of Airborne Hyperspectral image heavy metal-polluted soil concentration evaluation method returned based on Gaussian process | |
CN107230197B (en) | Tropical cyclone objective strength determination method based on satellite cloud image and RVM | |
CN107271405A (en) | A kind of water quality parameter retrieving concentration method and apparatus of multi-model collaboration | |
CN103258330B (en) | A kind of method of estimation of abundance of hyperspectral image end member | |
CN110186851A (en) | It is a kind of based on the semi-supervised Hyperspectral imaging heavy metal-polluted soil concentration evaluation method from Coded Analysis | |
CN112099110A (en) | Ocean internal wave forecasting method based on machine learning and remote sensing data | |
CN110536257B (en) | Indoor positioning method based on depth adaptive network | |
CN108875794A (en) | Image visibility detection method based on transfer learning | |
CN111783555A (en) | Deep learning satellite data cloud detection algorithm using unified samples | |
CN107632964A (en) | A kind of plane GEOMAGNETIC FIELD downward continuation recurrence cosine transform method | |
CN110321528A (en) | A kind of Hyperspectral imaging heavy metal-polluted soil concentration evaluation method based on semi-supervised geographical space regression analysis | |
CN115311502A (en) | Remote sensing image small sample scene classification method based on multi-scale double-flow architecture | |
Zhang et al. | Water meter pointer reading recognition method based on target-key point detection | |
CN113486581A (en) | Machine learning earth surface temperature estimation method, system, storage medium and equipment based on knowledge driving and logical reasoning | |
CN115494007A (en) | Random forest based high-precision rapid detection method and device for soil organic matters | |
CN114332534B (en) | Hyperspectral image small sample classification method | |
CN113901348A (en) | Oncomelania snail distribution influence factor identification and prediction method based on mathematical model | |
CN117710508A (en) | Near-surface temperature inversion method and device for generating countermeasure network based on improved condition | |
Zhang et al. | Modeling of spatial distributions of farmland density and its temporal change using geographically weighted regression model | |
CN116611561A (en) | Air quality prediction method and system | |
Tian et al. | Improving MODIS aerosol estimates over land with the surface BRDF reflectances using the 3-D discrete cosine transform and RossThick-LiSparse models | |
CN113553708B (en) | Method and device for tracing key influence factors of simulation model | |
CN112966710B (en) | FY-3D infrared hyperspectral cloud detection method based on linear discriminant analysis | |
Yin et al. | A Neural Network Method for Inversion of Turbulence Strength | |
CN112954633B (en) | Parameter constraint-based dual-network architecture indoor positioning method |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |