CN104573816B - The neural network clustering method of microwave radiometer remote sensing atmosphere parameter - Google Patents
The neural network clustering method of microwave radiometer remote sensing atmosphere parameter Download PDFInfo
- Publication number
- CN104573816B CN104573816B CN201410802336.XA CN201410802336A CN104573816B CN 104573816 B CN104573816 B CN 104573816B CN 201410802336 A CN201410802336 A CN 201410802336A CN 104573816 B CN104573816 B CN 104573816B
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- correlation
- microwave radiometer
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Radiation Pyrometers (AREA)
Abstract
The present invention relates to a kind of neural network clustering method of microwave radiometer remote sensing atmosphere parameter, it is characterised in that:Comprise the following steps, first by history sounding data, quantitative calculating is carried out to the correlation between air different layers knot, the big layer knot of coefficient correlation is elected, then artificial neural network training is substituted into by cluster, inverting exports the atmospheric parameter of different clusters respectively.Artificial neural network algorithm is by constantly adjusting the weight and deviation of network, to reduce by the deviation between the output water-vapo(u)r density vector that input vector is calculated and the training objective output water-vapo(u)r density vector of reality.After training is completed, the weights and deviation of network also it is determined that, therefore in microwave radiometer actual observation, it is only necessary to call its weight matrix and deviation, its arithmetic speed comparatively fast and stably, has higher inversion accuracy.
Description
Technical field
The present invention relates to a kind of Atmospheric Microwave remote sensing fields, more particularly to a kind of Neural Network Inversion skill of atmospheric parameter
Art.
Background technology
Artificial neural network is at a kind of novel information proposed based on human brain tissue structure, the Preliminary study of active mechanism
Reason system, there is the functions such as associative memory, Nonlinear Mapping, classification and identification, optimization calculating.For setting for artificial neural network
Meter and parameter setting include:Number of training, initial weight, network structure, network training and test etc..Artificial neuron
Network method is widely used in microwave radiometer Inverting Terrestrial Atmospheric Parameters, it is common practice that microwave radiometer is observed into bright temperature and substitutes into god
Through network, by below 10km atmospheric parameter overall outputs, because the characteristic of each layer of knot of air is different, the correlation between each layer
Also it is different, therefore the inner link rule between input and output is different, overall inverting output may result in inversion error
It is larger.
The content of the invention
The problem to be solved in the present invention is to provide a kind of nerve net of the smaller microwave radiometer remote sensing atmosphere parameter of error
Network clustering method.
In order to solve the above technical problems, the neural network clustering method of the microwave radiometer remote sensing atmosphere parameter of the present invention,
It is characterized in that:Comprise the following steps, first by history sounding data, the correlation between air different layers knot is determined
Amount is calculated, and the big layer knot of coefficient correlation is elected, and then substitutes into artificial neural network training by cluster, and inverting exports respectively
The atmospheric parameter of difference cluster.
The history sounding data is the atmospheric radiation bright temperature data that microwave radiometer receives.
The history sounding data for the temperature and humidity pressure and microwave radiometer on ground the bright temperature of measurement, by the temperature and humidity pressure on ground and
The bright temperature of measurement of microwave radiometer is as input parameter, using atmosphere vapour density profile as output parameter.Input vector X length
Spend for L, output vector Y length is M.The output function of any single neuron is:
In formula:F is the output valve of neuron, and ω is the input weight matrix of neuron, and p is the input value of neuron, and b is
The bias vector of neuron.Using logarithm tangent type function:
The set of frequency of the microwave radiometer:From 22.24 to steam and the K-band of liquid water sensitive, 23.04,
23.84th, 25.44,26.24,27.84,30,31.4GHz and to the 51.26 of temperature and the v wave bands of liquid water sensitive, 52.28,
53.86th, 54.94,55.5,56.66,57.3,58GHz and infrared thermometer (wavelength is 8~14 microns).The observation elevation angle is set
Put:90 degree.Neutral net input is set:The bright temperature and infrared temperature of ground temperature and humidity pressure and 16 passages.Neutral net exports
Vertical resolution:100 meters one layer between 0~2km, 250 meters one layer between 2~10km, amount to 53 layers plus ground floor,
The water-vapo(u)r density correlation of the different layers knot of air is analyzed using history sounding data, coefficient R
Calculation formula is:
In formula, ρ1、ρ2Respectively the water-vapo(u)r density of the different layers knot of air, C are covariance matrix, are calculated by following formula:
C(ρ1,ρ2)=E ((ρ1-μ1)(ρ2-μ2))
In formula:μ1、μ2Respectively air ρ1、ρ2Average value, E represent expectation computing.By the way that conclusion is calculated:Air
Layer knot distance is nearer, and its coefficient correlation is bigger, i.e., correlation is stronger;
According to the characteristics of neutral net, the strong data input neutral net of correlation can be more accurately fitted defeated
Enter the relation of output, then choose three groups of higher layer knots of coefficient correlation, such data are put together and cluster output, Ran Houtong
Cross neutral net to be trained respectively, obtain three groups of inverting coefficient matrixes, be combined into unified coefficient matrix:
ω=[ω1;ω2;ω3]
In formula:ω1、ω2、ω3The inverting coefficient matrix that respectively three group cluster data are obtained by neural computing,
During actual observation, by the unified output of three system number inversion results, so that it may obtain the water-vapo(u)r density profile under 10km height.
Artificial neural network algorithm is calculated by constantly adjusting the weight and deviation of network to reduce by input vector
Deviation between the output water-vapo(u)r density vector and actual training objective output water-vapo(u)r density vector that arrive.Although neutral net is calculated
Method iterates for a long time, but after training is completed during training, the weights and deviation of network also it is determined that,
Therefore in microwave radiometer actual observation, it is only necessary to call its weight matrix and deviation, its arithmetic speed it is very fast and
It is stable, there is higher inversion accuracy.Clustering neural network method proposed by the present invention, be first by quantitative calculating air not
With the correlation between layer knot, the big layer knot of coefficient correlation is found out, neural network structure is then substituted into respectively, obtains more accurately
Inverting coefficient matrix.
Brief description of the drawings
The present invention is further detailed explanation with reference to the accompanying drawings and detailed description.
Fig. 1 is neutral net vector model.
Fig. 2 is the coefficient correlation of atmosphere first layer knot (ground floor) and other 52 atmospheres, different atmospheric stratification steam
The coefficient correlation of density and surface vapor concentration.
Fig. 3 is the simulation result that error calculation is carried out using multiple checking samples, and Neural Network Inversion water-vapo(u)r density is square
Root error.
Fig. 4 is that the profile of Neural Network Inversion water-vapo(u)r density compares.
Fig. 5 is neural network clustering inverting flow process figure.
Embodiment
Using the bright temperature of the measurement of the temperature and humidity pressure on ground and microwave radiometer as input parameter, atmosphere vapour density profile is made
For output parameter.Input vector X length is L, and output vector Y length is M.The output function of any single neuron is:
In formula:F is the output valve of neuron, and ω is the input weight matrix of neuron, and p is the input value of neuron, and b is
The bias vector of neuron.Using logarithm tangent type function:
Specifically, the set of frequency of the microwave radiometer:From 22.24 to steam and the K-band of liquid water sensitive,
23.04th, 23.84,25.44,26.24,27.84,30,31.4GHz and to the 51.26 of temperature and the v wave bands of liquid water sensitive,
52.28th, 53.86,54.94,55.5,56.66,57.3,58GHz and infrared thermometer (wavelength is 8~14 microns).Observation is faced upward
Angle is set:90 degree.Neutral net input is set:The bright temperature and infrared temperature of ground temperature and humidity pressure and 16 passages.Neutral net
Export vertical resolution:100 meters one layer between 0~2km, 250 meters one layer between 2~10km, amount to 53 layers plus ground floor,
The water-vapo(u)r density correlation of the different layers knot of air is analyzed using history sounding data, coefficient R
Calculation formula is:
In formula, ρ1、ρ2Respectively the water-vapo(u)r density of the different layers knot of air, C are covariance matrix, are calculated by following formula:
C(ρ1,ρ2)=E ((ρ1-μ1)(ρ2-μ2))
In formula:μ1、μ2Respectively air ρ1、ρ2Average value, E represent expectation computing.By the way that conclusion is calculated:Air
Layer knot distance is nearer, and its coefficient correlation is bigger, i.e., correlation is stronger;
According to the characteristics of neutral net, the strong data input neutral net of correlation can be more accurately fitted defeated
Enter the relation of output, then choose three groups of higher layer knots of coefficient correlation, such data are put together and cluster output, Ran Houtong
Cross neutral net to be trained respectively, obtain three groups of inverting coefficient matrixes, be combined into unified coefficient matrix:
ω=[ω1;ω2;ω3]
In formula:ω1、ω2、ω3The inverting coefficient matrix that respectively three group cluster data are obtained by neural computing,
During actual observation, by the unified output of three system number inversion results, so that it may obtain the water-vapo(u)r density profile under 10km height.
The water-vapo(u)r density correlation of air different layers knot is analyzed using the somewhere history sounding data of 10 years, schemed
2 be the coefficient correlation of atmosphere first layer knot (ground floor) and other 52 atmospheres, as can be seen from Figure, atmospheric stratification distance
Nearer, its coefficient correlation is bigger, i.e., correlation is stronger.Therefore according to the characteristics of neutral net, it have chosen the high layer of coefficient correlation
Knot, the cluster that such data can be put together output.Cluster standard will be made a concrete analysis of for particular problem, choose 1 here
~5 layers, 6~15 layers, 16~53 layers as an example.Then it is respectively trained using neutral net, obtains three groups of inverting coefficients
Matrix.
Fig. 3 is the simulation result that error calculation is carried out using multiple checking samples, it can be seen that neutral net overall output
The inversion error of water-vapo(u)r density is maximum in the middle low layer (1~2km) of air, and clustering neural network method centering lower atmosphere layer
Inversion accuracy has preferable improvement.Bright temperature data is surveyed using QFW6000 microwave radiometers inverting is carried out to air, obtain two groups
Water-vapo(u)r density profile, and contrasted with sounding data, as shown in Figure 4, it can be seen that the inversion result of clustering neural network is in
Low layer more coincide with sounding, suitable with holistic approach precision in other layers of knot.
Claims (1)
- A kind of 1. neural network clustering method of microwave radiometer remote sensing atmosphere parameter, it is characterised in that:Comprise the following steps, it is first First usage history sounding data, quantitative calculating is carried out to the correlation between air different layers knot, by the big layer knot of coefficient correlation Elect, then substitute into artificial neural network training by cluster, inverting exports the atmospheric parameter of different clusters respectively;It is described to go through History sounding data is the atmospheric radiation bright temperature data that microwave radiometer receives;The history sounding data for ground temperature and humidity pressure and The bright temperature of measurement of microwave radiometer, using the bright temperature of the measurement of the temperature and humidity pressure on ground and microwave radiometer as input parameter, by air For water-vapo(u)r density profile as output parameter, input vector X length is L, and output vector Y length is M, any single neuron Output function be:<mrow> <mi>f</mi> <mo>=</mo> <mi>S</mi> <mrow> <mo>(</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>L</mi> </munderover> <msub> <mi>&omega;</mi> <mi>i</mi> </msub> <msub> <mi>P</mi> <mi>i</mi> </msub> <mo>+</mo> <mi>b</mi> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>In formula:F is the output valve of neuron, and ω is the input weight matrix of neuron, and p is the input value of neuron, and b is nerve The bias vector of member, using logarithm tangent type function:<mrow> <mi>S</mi> <mrow> <mo>(</mo> <mi>n</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mi>n</mi> </mrow> </msup> <mo>+</mo> <mn>1</mn> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>The set of frequency of the microwave radiometer:From 22.24 to steam and the K-band of liquid water sensitive, 23.04,23.84, 25.44th, 26.24,27.84,30,31.4GHz and to the 51.26 of temperature and the v wave bands of liquid water sensitive, 52.28,53.86, 54.94th, 55.5,56.66,57.3,58GHz and infrared thermometer, wavelength are 8~14 microns, and the observation elevation angle is set:90 degree, Neutral net input is set:The bright temperature and infrared temperature of ground temperature and humidity pressure and 16 passages, neutral net output is vertical to differentiate Rate:100 meters one layer between 0~2km, 250 meters one layer between 2~10km, amount to 53 layers plus ground floor,The water-vapo(u)r density correlation of the different layers knot of air is analyzed using history sounding data, the calculating of coefficient R Formula is:<mrow> <mi>R</mi> <mrow> <mo>(</mo> <msub> <mi>&rho;</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>&rho;</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>C</mi> <mrow> <mo>(</mo> <msub> <mi>&rho;</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>&rho;</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> </mrow> <msqrt> <mrow> <mi>C</mi> <mrow> <mo>(</mo> <msub> <mi>&rho;</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>&rho;</mi> <mn>1</mn> </msub> <mo>)</mo> </mrow> <mi>C</mi> <mrow> <mo>(</mo> <msub> <mi>&rho;</mi> <mn>2</mn> </msub> <mo>,</mo> <msub> <mi>&rho;</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> </mrow> </msqrt> </mfrac> </mrow>In formula, ρ1、ρ2Respectively the water-vapo(u)r density of the different layers knot of air, C are covariance matrix, are calculated by following formula:C(ρ1,ρ2)=E((ρ1-μ1)(ρ2-μ2))In formula:μ1、μ2Respectively air ρ1、ρ2Average value, E represent expectation computing, by the way that conclusion is calculated:Atmospheric stratification away from From nearer, its coefficient correlation is bigger, i.e., correlation is stronger;According to the characteristics of neutral net, by the strong data input neutral net of correlation, it is defeated can be more accurately fitted input The relation gone out, three groups of higher layer knots of coefficient correlation are then chosen, such data are put together and cluster output, then pass through god It is trained respectively through network, obtains three groups of inverting coefficient matrixes, be combined into unified coefficient matrix:ω=[ω1;ω2;ω3]In formula:ω1、ω2、ω3The inverting coefficient matrix that respectively three group cluster data are obtained by neural computing, it is actual During observation, by the unified output of three system number inversion results, so that it may obtain the water-vapo(u)r density profile under 10km height.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410802336.XA CN104573816B (en) | 2014-12-18 | 2014-12-18 | The neural network clustering method of microwave radiometer remote sensing atmosphere parameter |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410802336.XA CN104573816B (en) | 2014-12-18 | 2014-12-18 | The neural network clustering method of microwave radiometer remote sensing atmosphere parameter |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104573816A CN104573816A (en) | 2015-04-29 |
CN104573816B true CN104573816B (en) | 2017-12-26 |
Family
ID=53089833
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410802336.XA Active CN104573816B (en) | 2014-12-18 | 2014-12-18 | The neural network clustering method of microwave radiometer remote sensing atmosphere parameter |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104573816B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107703554B (en) * | 2017-08-24 | 2020-04-14 | 安徽四创电子股份有限公司 | Multichannel millimeter wave radiometer temperature and humidity profile inversion system and inversion method thereof |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3486675B2 (en) * | 2001-06-07 | 2004-01-13 | 独立行政法人通信総合研究所 | Method and apparatus for predicting electromagnetic disturbances in the terrestrial magnetosphere |
CN102998663A (en) * | 2012-12-27 | 2013-03-27 | 中国电子科技集团公司第二十二研究所 | Calibration method for microwave radiometer |
CN104007486A (en) * | 2014-06-05 | 2014-08-27 | 中国气象局气象探测中心 | Atmospheric temperature and humidity profile processing method and system with active and passive remote sensing combined |
CN104181507A (en) * | 2013-05-23 | 2014-12-03 | 中国科学院空间科学与应用研究中心 | Data processing and controlling apparatus used for ground-based microwave radiometer |
-
2014
- 2014-12-18 CN CN201410802336.XA patent/CN104573816B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3486675B2 (en) * | 2001-06-07 | 2004-01-13 | 独立行政法人通信総合研究所 | Method and apparatus for predicting electromagnetic disturbances in the terrestrial magnetosphere |
CN102998663A (en) * | 2012-12-27 | 2013-03-27 | 中国电子科技集团公司第二十二研究所 | Calibration method for microwave radiometer |
CN104181507A (en) * | 2013-05-23 | 2014-12-03 | 中国科学院空间科学与应用研究中心 | Data processing and controlling apparatus used for ground-based microwave radiometer |
CN104007486A (en) * | 2014-06-05 | 2014-08-27 | 中国气象局气象探测中心 | Atmospheric temperature and humidity profile processing method and system with active and passive remote sensing combined |
Non-Patent Citations (7)
Title |
---|
"A new calibration technique for Ground-Based Microwave Radiometer";Han Hengmin, Li Jiangman etal.;《IEEE》;20121231;第765-768页 * |
"Research on retrieving the liquid content of clouds by Ground-Based Microwave Radiometer";Han Hengmin, Li Jiangman etal.;《IEEE》;20121231;第971-974页 * |
"地基多通道微波辐射计反演大气温湿廓线的研究";周玉驰;《中国优秀硕士学位论文全文数据库 基础科学辑》;20110415(第04期);全文 * |
"地基微波辐射计大气环境遥感研究";王波;《中国优秀硕士学位论文全文数据库 基础科学辑》;20090315(第03期);正文第24-57页 * |
"地基微波辐射计遥感大气廓线的BP神经";刘亚亚 等;《高原气象》;20101231;第29卷(第6期);第1514-1522页 * |
"地基微波辐射计遥感阳江大气温湿廓线的研究";张曦;《中国优秀硕士学位论文全文数据库 基础科学辑》;20120915(第09期);全文 * |
"微波辐射计与GPS联合反演大气折射率";李江漫 等;《微波学报》;20130831;第29卷(第4期);第43-47页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104573816A (en) | 2015-04-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103792538B (en) | A kind of atmospheric profile retrieval method based on ground EO-1 hyperion microwave radiometer | |
Chen et al. | Partitioning of daily evapotranspiration using a modified shuttleworth-wallace model, random Forest and support vector regression, for a cabbage farmland | |
CN108508442A (en) | A kind of Atmosphere and humidity profiles inversion method based on ground multi-channel microwave radiometer | |
CN109212505B (en) | A kind of forest stand characteristics inversion method based on the multispectral high degree of overlapping image of unmanned plane | |
CN111060992B (en) | Equal-weight conjugate precipitation detection method and system for satellite-borne microwave dioxygen detection channel | |
Wenzhi et al. | Estimation of sunflower seed yield using partial least squares regression and artificial neural network models | |
CN110398466A (en) | Crop growth state monitoring method based on remote-sensing inversion | |
CN101738620A (en) | Method by utilizing passive microwave remote sensing data AMSR-E (Advanced Microwave Scanning Radiometer-EOS ) to invert surface temperature | |
CN108875794A (en) | Image visibility detection method based on transfer learning | |
CN110222903A (en) | A kind of Rice Yield Prediction method and system based on unmanned aerial vehicle remote sensing | |
CN112906300B (en) | Polarization SAR soil humidity inversion method based on double-channel convolutional neural network | |
CN109186811A (en) | A kind of FBG temperature calibration method based on BP neural network | |
CN113011372B (en) | Automatic monitoring and identifying method for saline-alkali soil | |
CN104020135A (en) | Calibration model establishing method based on near infrared spectrum | |
CN111160680A (en) | Agricultural drought assessment method based on information assimilation and fusion | |
CN115308386B (en) | Soil salinity inversion method and system based on CYGNSS satellite data | |
CN113486581A (en) | Machine learning earth surface temperature estimation method, system, storage medium and equipment based on knowledge driving and logical reasoning | |
CN110569605A (en) | Non-glutinous rice leaf nitrogen content inversion model method based on NSGA2-ELM | |
Liu et al. | Estimating leaf chlorophyll contents by combining multiple spectral indices with an artificial neural network | |
CN105095652B (en) | Sample component assay method based on stack limitation learning machine | |
CN114880933A (en) | Atmospheric temperature and humidity profile inversion method and system for non-exploration-site foundation microwave radiometer based on reanalysis data | |
CN112766161A (en) | Hyperspectral target detection method based on integrated constraint multi-example learning | |
Cui et al. | Predicting carbon and water vapor fluxes using machine learning and novel feature ranking algorithms | |
CN115049026A (en) | Regression analysis method of space non-stationarity relation based on GSNNR | |
Lou et al. | An effective method for canopy chlorophyll content estimation of marsh vegetation based on multiscale remote sensing data |
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 |