CN107230197A - Tropical cyclone based on satellite cloud picture and RVM is objective to determine strong method - Google Patents
Tropical cyclone based on satellite cloud picture and RVM is objective to determine strong method Download PDFInfo
- Publication number
- CN107230197A CN107230197A CN201710406752.1A CN201710406752A CN107230197A CN 107230197 A CN107230197 A CN 107230197A CN 201710406752 A CN201710406752 A CN 201710406752A CN 107230197 A CN107230197 A CN 107230197A
- Authority
- CN
- China
- Prior art keywords
- strong
- rvm
- objective
- gradient
- deviation
- 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
- 238000000034 method Methods 0.000 title claims abstract description 28
- 239000011159 matrix material Substances 0.000 claims abstract description 60
- 230000004927 fusion Effects 0.000 claims abstract description 45
- 238000012512 characterization method Methods 0.000 claims abstract description 36
- 238000002474 experimental method Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 7
- 238000002156 mixing Methods 0.000 claims description 6
- 230000031068 symbiosis, encompassing mutualism through parasitism Effects 0.000 claims description 3
- 241000208340 Araliaceae Species 0.000 claims description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 2
- 230000008859 change Effects 0.000 claims description 2
- 235000008434 ginseng Nutrition 0.000 claims description 2
- 238000010801 machine learning Methods 0.000 claims 1
- 238000005516 engineering process Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 238000012417 linear regression Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241001300078 Vitrea Species 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000000116 mitigating effect Effects 0.000 description 1
- 230000003389 potentiating effect Effects 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
- 238000013316 zoning Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
- G06T7/45—Analysis of texture based on statistical description of texture using co-occurrence matrix computation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30192—Weather; Meteorology
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geometry (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Processing (AREA)
- Radiation Pyrometers (AREA)
- Image Analysis (AREA)
Abstract
The present invention builds the objective fixed strong model of tropical cyclone (Tropical Cyclone, TC) based on fixed statellite cloud atlas and Method Using Relevance Vector Machine (Relevance Vector Machine, RVM).Mainly comprising following two aspects：(1) first, infrared and vapor channel cloud atlas is merged using laplacian pyramid algorithm.It is as a reference point with TC centers, construct angle of deviation gradient co-occurrence matrix.The present invention sets up the objective fixed strong models of TC using the information structurings such as the multiple statistical parameter combination TC kernels yardsticks and center latitude in co-occurrence matrix and the closely related characterization factor of TC intensity using RVM.(2) it is as a reference point successively with every bit based on fusion satellite cloud picture, angle of deviation gradient co-occurrence matrix is constructed, minimum value, intermediate value and the average of co-occurrence matrix statistical parameter battle array is calculated.The present invention is using multiple statistical parameters in co-occurrence matrix parametric array and combines the information structuring such as TC kernels yardstick and center latitude and the closely related characterization factor of TC intensity, and the objective fixed strong models of TC are set up using RVM.
Description
Technical field
The invention belongs to image processing techniques and weather prognosis field.It particularly relates to surely strong to improve tropical cyclone
The tropical cyclone based on satellite cloud picture and Method Using Relevance Vector Machine (Relevance Vector Machine, RVM) for the purpose of precision
It is objective to determine strong method.
Background technology
Tropical cyclone (Tropical Cyclone, TC) is to endanger a kind of calamity common in Chinese numerous natural calamities
Evil, its activity is along with blast, heavy rain and storm tide, or even can cause the natural geologic hazards such as landslide and mud-rock flow.TC intensity
With the accurate forecast in path, influence most important to preventing and mitigating its disaster brought.
In recent years, the development of all kinds of observation methods and Numerical Forecast Technology has promoted TC course guidances level not both at home and abroad
Disconnected to improve, still, the prediction ability of TC intensity is but in progress very slow.On the one hand its main cause is the surely strong scarce capacities of TC,
On the other hand be to influence TC intensity mechanism it is still unclear, and these two aspects close ties.There is scholar to carry out correlative study
Draw：Lower accuracy surely strong TC hinders the progress of TC forecast of intensity.In 1973, Dvorak proposed to use infrared satellite cloud
Figure determines TC intensity, and it is surely strong theoretical to form in 1975 the TC based on satellite cloud picture.Dvorak successively in 1977 and
TC intensity estimation technique was improved twice in 1984.After 1987, the aircraft method for detecting of northwest Pacific by
It is deactivated in cost issues.Because the subjectivity of Dvorak technologies is larger, in 1989, Zehr, which proposes one kind, can improve TC
Fixed the objective of strong precision determines strong method.In 1998, the scholar such as Velden proposed the visitor based on fixed statellite infrared digital image
Fixed strong scheme is seen to estimate TC intensity.Original Dvorak technologies are often relied on subjective experience when surely strong, and easily TC clouds shape is missed
Sentence, cause the precision reduction that TC is surely strong.In 2007, Olander and Velden were improved Dvorak technologies, proposed profit
Determine strong method with the TC of fixed statellite infrared cloud image is objective.
Current TC is surely strong main including being based on polar-orbiting satellite Microwave Data and fixed statellite cloud atlas data.It is micro- in polar-orbiting satellite
In terms of ripple data, application studies of such as scholar Lu Yi using TRMM/TMI data in the estimation of TC intensity.Neeru Jaiswal
Data mining technology is combined with QuickScat satellite ocean wind scatterometer data acquireds, TC intensity is predicted.Although
Microwave Data can detect the structure of the TC below cloud top, but be due to easily by precipitation interference and polar-orbiting satellite when
Between resolution ratio it is not high the reason for, tend not to comprehensively catch TC Strength Changes and its correlation internal convection structural evolution.And it is quiet
Only satellite data temporal resolution is high, at present by more surely strong applied to TC.There are many scholars to do the research work of correlation, it is main
Concentrate on and utilize visible ray and infrared channel data surely strong, wherein to utilize single channel data, especially infrared channel data
At most, such as Dvorak technologies.Pineros etc. is changed using infrared cloud picture of satellite to TC structure and intensity to be studied.
The scholars such as Fetanat have studied a kind of objective fixed strong technologies of TC based on infrared cloud image in 2013, and this method is needed in TC
The heart extracts azimuthal bright temperature information.The scholars such as Knaff proposed that the TC based on TC information and infrared cloud picture of satellite was strong in 2015
Spend the improved method of estimation.The scholars such as Zhao proposed the multivariate linear model based on infrared cloud picture of satellite in 2016, to northwest
Pacific TC intensity estimated, test result indicates that：The model has a preferably fixed strong precision for violent typhoon, but for compared with
Weak TC, then have larger fixed strong error.Gholamreza and Abdollah is from the bright temperature data in TC center extractions orientation, based on K
Class mean algorithm carries out objective estimation to TC intensity.Miguel etc. estimates that TC is strong using deviation angular variance and Sigmoid functions
Degree.
When only considering infrared channel, positioned only when TC, which has, compares clearly Vitrea eye or obvious spiral rainband feature in TC
The heart is reliable, but when the region cloud top where TC is covered by cirrus and can not observe above architectural feature, in positioning TC
The heart is relatively difficult.By contrast, infrared channel combination vapor channel data can clearly provide typhoon eye and more rich cloud
It is detailed information, is favorably improved the precision of TC centralized positionings, and to directly affect follow-up TC surely strong for the precision of TC centralized positionings
Quality.In addition, vapor channel data to TC clouds type, steam feature etc. to portray (these are closely related with TC intensity) more clear
Chu, enables in particular to protrude the feature in strong convection region.Recently have a few studies work and combine the infrared and bright temperature difference of vapor channel
Data is surely strong for TC, achieves preferably fixed potent fruit.As Olander and Velden is combined infrared with vapor channel data
Use, TC intensity is estimated.Zhuge Xiaoyong etc. proposed a kind of infrared and water based on fixed statellite in 2015
The TC intensity methods of estimation that vapour image is combined.However, such method still has a larger room for improvement at three aspects, first,
Existing method analyzed area is fixed as the kernel area of radii fixus size around TC centers, but different TC kernel area yardsticks
It is change, surely strong precision and the scope of application will certainly be influenceed using unified kernel yardstick.Second, the fixed strong technology is used
Linear regression is modeled, and is all generally more complicated non-linear relation between TC intensity and its factor of influence, letter
The non-linear relation between TC intensity and its factor of influence singly can not be built well using linear regression method modeling.3rd,
In existing research, seldom influence of the center latitude to TC intensity simultaneously.Correlative study shows that the Position Latitude at TC centers is to TC
Intensity has considerable influence, and only considers bright temperature data and determine TC good general and can cause certain error.
The content of the invention
Strong method is determined it is an object of the invention to provide a kind of tropical cyclone based on satellite cloud picture and RVM is objective.First, it is sharp
Infrared and vapor channel cloud atlas is merged with laplacian pyramid blending algorithm, so as to obtain fusion satellite cloud picture.
Then it is as a reference point with TC centers, construct the angle of deviation-gradient co-occurrence matrix.The present invention utilizes multiple statistics in co-occurrence matrix
The information structuring such as parameter combination TC kernels yardstick and center latitude and the closely related characterization factor of TC intensity, are set up using RVM
The objective fixed strong models of TC.It is as a reference point successively with every bit based on fusion satellite cloud picture herein on basis, construct deviation
Angle-gradient co-occurrence matrix parametric array, calculates minimum value, intermediate value and the average of the statistical parameter battle array of co-occurrence matrix.The present invention is utilized
Multiple statistical parameters (minimum value, intermediate value and average) and the combination information such as TC kernels and center latitude in co-occurrence matrix parametric array
Construction and the closely related characterization factor of TC intensity, the objective fixed strong models of TC are set up using RVM.
The tropical cyclone based on satellite cloud picture and RVM is objective determines strong method by the present invention, comprises the following steps that：
Step 1 is based on Laplacian-pyramid image blending algorithm to the infrared 1 passage cloud atlas and steam in satellite cloud picture
Passage cloud atlas carries out fusion treatment, so as to obtain a width fusion cloud image；
The yearbook data that step 2 is provided using Shanghai Institute of Typhoon determines TC center, then using TC centers as
The center of circle is outwards expanded interception and melted with radially every 50 kilometers from TC centers in away from the kilometer range of centre distance 200 for interval
Close cloud atlas；
Step 3 calculates the bright temperature gradient matrix of the fusion cloud image of interception, and Zai Yi TC centers are as a reference point, calculates and obtains
Angle of deviation battle array；
Step 4 builds the fusion cloud image angle of deviation-gradient co-occurrence matrix；
Step 5 is configured to characterize the best features factor of TC intensity, and is tested based on the best features factor in optimal TC
Core yardstick；
Step 6 is based on RVM under optimal kernel yardstick, using characterization factor and builds the objective fixed strong models of TC, to TC intensity
Carry out objective estimation；
Step 7, using characterization factor and combination TC centers latitude, builds TC objective under optimal kernel yardstick based on RVM
Fixed strong model, objective estimation is carried out to TC intensity；
Step 8 on the basis of step 2, calculate interception fusion cloud image bright temperature gradient matrix, then using every bit as
Reference point, calculates and obtains angle of deviation battle array；
Step 9 builds the fusion cloud image angle of deviation-gradient co-occurrence matrix parametric array, then calculates each parameter correspondence minimum
Value, intermediate value and average；
The best features factor that step 10 is constructed based on step 5, construct relevant parameter can characterize TC intensity most
Good characterization factor, and and optimal TC kernels yardstick is tested based on the best features factor；
Step 11 is based on RVM under optimal kernel yardstick, using characterization factor and builds the objective fixed strong models of TC, strong to TC
Degree carries out objective estimation；
Step 12, using characterization factor and combination TC centers latitude, TC visitors is built based on RVM under optimal kernel yardstick
Fixed strong model is seen, objective estimation is carried out to TC intensity；
1. the cloud atlas data used in the present invention has 5 passages from the meteorological fixed statellite FY-2 of China：Infrared 1 leads to
Road, infrared 2 passage, aqueous vapor passage, infrared 4 passage, visible channel.In the step 1 based on laplacian pyramid figure
Picture blending algorithm carries out fusion treatment to infrared 1 passage satellite cloud picture and vapor channel cloud atlas, and its step is as follows：
Step 1 decomposes original image by gaussian pyramid；
Step 2 decomposes the decomposition result of gaussian pyramid using Laplacian pyramid algorithm, so that it is general to obtain drawing
Lars pyramid；
Step 3, in the different scale fusion rule different with the characteristic design of resolution ratio, is obtained according to laplacian pyramid
Laplacian pyramid after to fusion；
Step 4 carries out inverse transformation to it, that is, reconstructs, and finally gives the image after fusion.
In step 3, the present invention is taken greatly to the fusion rule of top layer images using region gradient, and other layers of fusion rule are adopted
Taken with region energy greatly, specific method is as follows：
Average gradient represents the definition of image, also reflects the feature of image minor detail and texture variations.First, count
Calculate the zone leveling gradient of M × N sizes in top layer images centered on each pixel：
Wherein, IxIt is the first-order differences of pixel f (x, y) in the x direction, and IyThe jumps of pixel f (x, y) in y-direction
Point, expression formula is as follows：
ΔIx=f (x, y)-f (x-1, y) (2)
ΔIy=f (x, y)-f (x, y-1) (3)
Therefore, G (i, j) is exactly each pixel L in top layer imagesNZone leveling gradient corresponding to (i, j), wherein LlFor
L tomographic images (0≤l≤N) after Laplacian pyramid.So, top layer images fusion results are：
In formula, LFN(i, j) is top layer fused image, LAlAnd LBlIt is that source images A and B pass through the golden word of Laplce respectively
L tomographic images after tower decomposition.
For other l tomographic images (0≤l ＜ N), zoning energy after Laplacian pyramid：
In formula, p=1, q=1,
So, other tomographic image fusion results are：
Obtain the fused images LF of each layer of laplacian pyramidlAfter (0≤l≤N), by reconstruct, it can finally be melted
Close image.
2. described in intercept fusion cloud image under different kernel yardsticks in step 2, carried first with Shanghai Institute of Typhoon
The TC yearbooks data of confession is positioned to TC centers, then by the center of circle of TC centers in away from the kilometer range of centre distance 200, from
TC sets out at center and outwards expands interception fusion cloud image with radially every 50 kilometers for interval.
3. the structure fusion cloud image angle of deviation-gradient co-occurrence matrix in above-mentioned steps 4, the TC of mature shape
Similar circle, i.e. zhou duicheng tuxing.For arbitrary graphic, to determine whether zhou duicheng tuxing, the gradient that can be put from certain
The angle of direction and the point and the RADIAL of reference point judges that the gradient direction of the point and the angle of RADIAL be the angle of deviation.
If figure is closer to zhou duicheng tuxing, then, the probability that the angle of deviation is intended to 0 ° is bigger.TC from initial stage to the maturity period,
To last extinction, with not increasing for TC intensity, whole cloud system is gradually intended to zhou duicheng tuxing, is especially reached most in TC intensity
The big stage is the most obvious.
1984, flood proposed Gray Level-Gradient Co-occurrence Matrix after light, has 15 statistical parameters.It is common based on Gray Level-Gradient
Raw matrix principle, the present invention proposes the angle of deviation-gradient co-occurrence matrix.The element definition of the angle of deviation-gradient co-occurrence matrix is to return
It is common that there is the total pixel number that the angle of deviation is i and gradient is j in one angle of deviation matrix and normalized gradient image changed.It is right
The following angle of deviation-gradient co-occurrence matrix is normalized：
15 statistical parameters of 1 angle of deviation of table-gradient co-occurrence matrix
4. being configured in above-mentioned steps 5 characterizes the best features factor of TC intensity, due to the angle of deviation-gradient symbiosis square
Have 15 statistical parameters in battle array, tested by modeling error can find out the feature that is best suited for building the objective fixed strong models of TC because
Son, its step is as follows：
Step 1 is utilized respectively each statistical parameter and TC centers wind speed (the TC yearbook data that Shanghai Institute of Typhoon is provided
In can obtain), the surely strong models of TC are set up based on linear regression method；
Step 2 analyzes the fixed strong error of each characterization factor, as shown in figure 3, and carry out sequence from small to large, successively
For：T6、T15、T1、T11、T14、T3、T13、T10、T5、T8、T2、T7、T4、T12、T9；
Step 3 gradually increases characterization factor dimension since the minimum parameter of error, increases to 15 dimensions from 1 dimension, respectively
The surely strong models of 15 TC are set up using RVM；
Step 4 analyzes the error results of 15 fixed strong models, as shown in figure 4, and determine modeling characterization factor it is optimal
Dimension be 9 dimension, comprising statistical parameter be：T6、T15、T1、T11、T14、T3、T13、T10、T5.
5. in above-mentioned steps 5-7, based on 9 characterization factors obtained by step 4 experiment, based on RVM respectively in different kernels
The surely strong models of TC are built under yardstick, it is 200km finally to measure the minimum kernel yardstick of error, as shown in table 2, and in 9 spies
Levy on the basis of the factor, with reference to TC centers latitude, the objective fixed strong models of TC are built based on RVM.
The fixed strong error of different kernel radial dimensions of the table 2 by reference point of TC central points
6. it is as a reference point successively with every bit based on the satellite cloud picture after fusion in above-mentioned steps 8-12, calculate deviation
Angle battle array, so as to construct the angle of deviation-gradient co-occurrence matrix statistical parameter battle array, calculate the minimum value of each statistical parameter battle array, intermediate value and
Average, then parameter T6 minimum value, intermediate value and average are based on RVM with center wind speed respectively and build fixed strong model, error result
As shown in table 3.When use the average of co-occurrence matrix parametric array as modeling characterization factor when, mean absolute error surely strong TC
All it is minimum with average relative error, it can be seen that, the average of co-occurrence matrix parametric array is more suitable for building the objective fixed strong moulds of TC
Type.Then, the surely strong models of TC are built under different kernel yardsticks based on RVM respectively, finally measures the minimum kernel yardstick of error
For 200km, as shown in table 4, and on the basis of 9 characterization factors, with reference to TC centers latitude, TC is built based on RVM objective
Fixed strong model.
Error result of the statistical parameter of table 3 battle array T6 minimum value, intermediate value and the average respectively as the TC strength characteristic factors
The fixed strong error of different radial direction kernel yardsticks of the table 4 by reference point of each point
Brief description of the drawings
Fig. 1 is based on satellite cloud picture and RVM with the objective fixed strong models of TC centers TC as a reference point；
Fig. 2 is based on satellite cloud picture and RVM with every bit successively objective fixed strong models of TC as a reference point；
Fig. 3 difference statistical parameter as the TC strength characteristic factors error curve；
Fig. 4 difference dimension statistical parameter as the TC strength characteristic factors error curve；
The error block diagram of the surely strong models of RVM when Fig. 5 kernel functions are Gauss；
The error block diagram of the surely strong models of RVM when Fig. 6 kernel functions are Cauchy；
The error block diagram of the surely strong models of RVM when Fig. 7 kernel functions are Cauchy；
The error block diagram of the surely strong models of RVM when Fig. 8 kernel functions are Gauss
Embodiment
132 TC during the present invention is used 2005 to 2014 obtained by the scanning of FY-2 satellites, including the torrid zone
Storm, severe tropical storm, typhoon, violent typhoon and Super Typhoon.Due to yearbook data be at interval of 3 hours or 6 hours, so through
Cross to select and finally obtain the vapor channel cloud atlas of the 2744 infrared 1 passage cloud atlas and 2744 width that have yearbook data in the same time.This
Invention determines strong method, including two aspects there is provided a kind of TC based on satellite cloud picture and RVM is objective：It is as a reference point with TC centers
It is as a reference point successively with every bit, perform following steps successively as depicted in figs. 1 and 2.Step performed by Fig. 1：
Step 1 is based on Laplacian-pyramid image blending algorithm to the infrared 1 passage cloud atlas and steam in satellite cloud picture
Passage cloud atlas carries out fusion treatment, so as to obtain a width fusion cloud image；
The yearbook data that step 2 is provided using Shanghai Institute of Typhoon determines TC center, then using TC centers as
The center of circle is outwards expanded interception and melted with radially every 50 kilometers from TC centers in away from the kilometer range of centre distance 200 for interval
Close cloud atlas；
Step 3 calculates the bright temperature gradient matrix of the fusion cloud image of interception, and Zai Yi TC centers are as a reference point, calculates and obtains
Angle of deviation battle array；
Step 4 builds the fusion cloud image angle of deviation-gradient co-occurrence matrix；
Step 5 is configured to characterize the best features factor of TC intensity, and is tested based on the best features factor in optimal TC
Core yardstick；
Step 6 is based on RVM under optimal kernel yardstick, using characterization factor and builds the objective fixed strong models of TC, to TC intensity
Carry out objective estimation；
Step 7, using characterization factor and combination TC centers latitude, builds TC objective under optimal kernel yardstick based on RVM
Fixed strong model, objective estimation is carried out to TC intensity；
Step performed by Fig. 2：
Step 1 calculates the bright temperature gradient matrix of the fusion cloud image of interception on the basis of Fig. 1 steps 2, then with every bit
It is as a reference point, calculate and obtain angle of deviation battle array；
Step 2 builds the fusion cloud image angle of deviation-gradient co-occurrence matrix parametric array, then calculates each parameter correspondence minimum
Value, intermediate value and average；
Step 3 calculates each parameter correspondence minimum value, intermediate value and average, and the best features constructed based on step 5 because
Son, construct relevant parameter can characterize the best features factor of TC intensity, and and optimal based on the test of the best features factor
TC kernel yardsticks；
Step 4 is based on RVM under optimal kernel yardstick, using characterization factor and builds the objective fixed strong models of TC, to TC intensity
Carry out objective estimation；
Step 5, using characterization factor and combination TC centers latitude, builds TC objective under optimal kernel yardstick based on RVM
Fixed strong model, objective estimation is carried out to TC intensity；
Carry out the performance of labor this method below by four groups of experiments：
Experiment 1：As a reference point with TC centers, the present invention builds 9 statistical parameters and center in co-occurrence matrix based on RVM
The fixed strong model of wind speed, as shown in table 5, when RVM chooses Gauss kernel functions, TC is fixed for the experimental results of different kernel functions
Strong absolute error and relative error histogram results are as shown in Figure 5.
The application condition of the surely strong models of RVM under the different kernel functions of table 5
Experiment 2：As a reference point with TC centers, the present invention is based on 9 statistical parameters in RVM structure co-occurrence matrixs and combines
The fixed strong model of center latitude and center wind speed, center latitude data derives from yearbook data, the experiment test of different kernel functions
As a result as shown in table 6, when RVM chooses Cauchy kernel functions, absolute error surely strong TC and relative error histogram results are such as
Shown in Fig. 6.
The application condition of the surely strong models of RVM under the different kernel functions of table 6
Experiment 3：As a reference point successively with every bit, the present invention builds 9 statistics in co-occurrence matrix parametric array based on RVM
Parameter and the fixed strong model of center wind speed, the experimental results of different kernel functions are as shown in table 7, when RVM chooses Cauchy cores
During function, absolute error and relative error histogram results surely strong TC is as shown in Figure 7.
The application condition of the surely strong models of RVM under the different kernel functions of table 7
Experiment 4：As a reference point successively with every bit, the present invention builds 9 statistics in co-occurrence matrix parametric array based on RVM
Parameter and the fixed strong model for combining center latitude and center wind speed, center latitude data derive from yearbook data, different kernel functions
Experimental results as shown in table 8, when RVM chooses Gauss kernel functions, surely strong TC absolute error and relative error column
Figure result is as shown in Figure 8.
The application condition of the surely strong models of RVM under the different kernel functions of table 8
It is as a reference point with TC centers, based on fusion cloud image and the objective fixed strong models of RVM constructions TC.First, it is general using drawing
Lars pyramid blending algorithm is merged to infrared and vapor channel cloud atlas, obtains fusion satellite cloud picture.Then with TC
The heart be the center of circle to away from centre distance 200km in the range of, from TC centers so that radially often 50km outwards expands interception as interval and melted
Close cloud atlas.The bright temperature gradient matrix of the fusion cloud image of interception is calculated, it is as a reference point with TC centers, angle of deviation matrix is calculated, so
The angle of deviation-gradient co-occurrence matrix is constructed afterwards.The present invention is using multiple statistical parameters in co-occurrence matrix and combines TC kernels with
The information structurings such as heart latitude and the closely related characterization factor of TC intensity, the objective fixed strong models of TC are set up using RVM.Experimental result
Show, contrast the fixed strong error of each kernel yardstick, optimal radial yardstick distance is 200km.When using 9 of co-occurrence matrix most
When good statistical parameter builds fixed strong model, the fixed strong error of RVM models is minimum.After the latitude of addition center, the fixed of RVM models misses by force
Difference reduction.RVM models have preferable high dimensional nonlinear disposal ability and intensity estimated capacity, and TC intensity effectively can be estimated
Meter.
It is as a reference point successively with every bit, based on fusion cloud image and the objective fixed strong models of RVM constructions TC.Using last
The fusion cloud image separately won, by the center of circle of TC centers in the range of away from centre distance 200km, from TC centers with radially every
50km outwards expands interception fusion cloud image for interval.The bright temperature gradient matrix of the fusion cloud image of interception is calculated, with every bit successively
It is as a reference point, angle of deviation battle array is calculated, the angle of deviation-gradient co-occurrence matrix is then constructed, the statistical parameter battle array of co-occurrence matrix is calculated
Minimum value, intermediate value and average.The present invention utilizes multiple statistical parameters (minimum value, the intermediate value and in co-occurrence matrix parametric array
Value) and the information structuring such as TC kernels and center latitude and the closely related characterization factor of TC intensity are combined, set up TC visitors using RVM
See fixed strong model.Test result indicates that, the fixed strong error of each yardstick is contrasted, optimal radial yardstick distance is 200km.Symbiosis square
It is surely strong that the average of battle array statistical parameter battle array is more suitable for TC.Determine strong model when 9 optimal statistical parameters using co-occurrence matrix are built
When, the fixed strong error of RVM models is minimum.After the latitude of addition center, the fixed strong error reduction of RVM models.RVM models have preferable
High dimensional nonlinear disposal ability and intensity estimated capacity, TC intensity can effectively be estimated.Contrast is used as ginseng using TC centers
Examination point and with the surely strong errors of the TC of every bit successively two schemes as a reference point, when as a reference point successively with every bit,
The error of the surely strong models of TC is smaller.Illustrate with every bit successively scheme as a reference point can extract more comprehensively, it is more rich, more
Accurate characteristic information, is favorably improved the objective fixed strong precision of TC.
Claims (5)
1. the tropical cyclone (TC) based on satellite cloud picture and Method Using Relevance Vector Machine (RVM) is objective to determine strong method.This method is Cloud
The application of figure data and machine learning algorithm in meteorological field surely strong TC.Comprise the following steps：
Step 1 is based on Laplacian-pyramid image blending algorithm to the infrared 1 passage cloud atlas and vapor channel in satellite cloud picture
Cloud atlas carries out fusion treatment, obtains a width fusion cloud image；
Step 2 by the center of circle of TC centers in away from the kilometer range of centre distance 200, from TC centers with radially it is every 50 kilometers
Interception fusion cloud image is outwards expanded for interval；
Step 3 calculates the bright temperature gradient matrix of the fusion cloud image of interception, and Zai Yi TC centers are as a reference point, calculates and obtains deviation
Angle battle array.Build in the angle of deviation-gradient co-occurrence matrix of fusion cloud image, co-occurrence matrix comprising 15 statistical parameters；
Step 4 is configured to characterize the best features factor of TC intensity, and tests optimal TC kernels chi based on the best features factor
Degree；
Step 5 is based on RVM under optimal kernel yardstick, using characterization factor and builds the objective fixed strong models of TC, and TC intensity is carried out
Objective estimation；
Step 6 is under optimal kernel yardstick, using characterization factor and combination TC centers latitude, builds TC based on RVM objective fixed strong
Model, objective estimation is carried out to TC intensity；
Step 7 calculates the bright temperature gradient matrix of the fusion cloud image of interception on the basis of step 2, then is used as reference using every bit
Point, calculates and obtains angle of deviation battle array.The fusion cloud image angle of deviation-gradient co-occurrence matrix parametric array is built, each parameter pair is then calculated
Answer minimum value, intermediate value and average；
The best features factor that step 8 is constructed based on step 4, construct relevant parameter can characterize the optimal spy of TC intensity
The factor is levied, optimal TC kernels yardstick is tested based on the best features factor；
Step 9 is based on RVM under optimal kernel yardstick, using characterization factor and builds the objective fixed strong models of TC, and TC intensity is carried out
Objective estimation；
Step 10, using characterization factor and combination TC centers latitude, builds TC objective fixed under optimal kernel yardstick based on RVM
Strong model, objective estimation is carried out to TC intensity.
2. the TC based on satellite cloud picture and RVM is objective as claimed in claim 1 determines strong method, it is characterised in that：For fusion
Satellite cloud picture, by the center of circle of TC centers in away from the kilometer range of centre distance 200, be from TC centers with radially every 50 kilometers
It is respectively 50km, 100km, 150km and 200km that interception fusion cloud image, i.e. kernel yardstick are outwards expanded in interval, studies different interior
The surely strong models of TC in the case of core yardstick.As a result show, though it is as a reference point with TC centers, or ginseng is used as using every bit successively
Examination point, when kernel yardstick is 200km, TC fixed strong error is minimum.Therefore, intercept and melt during using kernel yardstick by 200km
The data message of cloud atlas is closed as the objective fixed strong modelings of next step TC.
3. the TC based on satellite cloud picture and RVM is objective as claimed in claim 1 determines strong method, it is characterised in that：For fusion
Satellite cloud picture, by calculating angle of deviation battle array and bright temperature gradient matrix, constructs the angle of deviation-gradient co-occurrence matrix., Hong Jiguang in 1984
Gray Level-Gradient Co-occurrence Matrix is proposed, 15 statistical parameters are had.Based on Gray Level-Gradient Co-occurrence Matrix principle, the present invention is proposed
The angle of deviation-gradient co-occurrence matrix.The element definition of the angle of deviation-gradient co-occurrence matrix is in normalized angle of deviation matrix and normalizing
It is common that there is the total pixel number that the angle of deviation is i and gradient is j in the gradient image of change.To the following angle of deviation-gradient co-occurrence matrix
It is normalized：
4. the TC based on satellite cloud picture and RVM is objective as claimed in claim 1 determines strong method, it is characterised in that：It is configured to
The best features factor of TC intensity is characterized, due to having 15 statistical parameters in the angle of deviation-gradient co-occurrence matrix, is missed by modeling
Difference test, from small to large sequence is followed successively by：T6、T15、T1、T11、T14、T3、T13、T10、T5、T8、T2、T7、T4、T12、T9.
The present invention is studied constructing the multiple features factor related to TC intensity, since the minimum parameter T6 of error, is gradually increased
Characterization factor dimension, increases to 15 dimensions from 1 dimension, is utilized respectively RVM and sets up the surely strong models of 15 TC, tested by modeling error,
When characterization factor dimension reaches 9 dimension, mean absolute error and average relative error are all minimum, respectively T6, T15, T1, T11,
T14, T3, T13, T10, T5 are fixed strong when illustrating 9 statistical parameters as the multiple characterization factor closely related with TC intensity
The error of model is minimum.
In the case where kernel yardstick is 200km, with TC centers it is as a reference point when, on the basis of 9 characterization factors, add TC centers
Latitude, builds the surely strong models of TC using 10 characterization factors, is drawn through experiment：Compared to the fixed strong mould using 9 characterization factors
Type, in using 9 parameter (T6, T15, T1, T11, T14, T3, T13, T10, T5)+TC in the angle of deviation-gradient co-occurrence matrix
During heart latitude, TC is surely strong, and error is smaller.Draw a conclusion：When kernel yardstick be 200km when, using 10 characterization factors (T6, T15,
T1, T11, T14, T3, T13, T10, T5, TC center latitude) the surely strong models of TC are built, fixed strong error is smaller.
5. the TC based on satellite cloud picture and RVM is objective as claimed in claim 1 determines strong method, it is characterised in that：When with each
When point is as a reference point successively, the fusion cloud image angle of deviation-gradient co-occurrence matrix parametric array is built, each parameter correspondence is then calculated
Minimum value, intermediate value and average, fixed strong mould is built using statistical parameter battle array T6 minimum value, intermediate value, average with center wind speed respectively
Type, modeling tool is RVM.Test result indicates that：When use the average of co-occurrence matrix parametric array as modeling characterization factor when,
Mean absolute error surely strong TC and average relative error are all minimum, it can be seen that, the average of co-occurrence matrix parametric array is more suitable
For building the objective fixed strong models of TC.
Kernel yardstick be 200km under, with every bit it is as a reference point successively when, in above-mentioned 9 characterization factors, (each factor is each
From average) on the basis of, add TC centers latitude, build the surely strong models of TC using 10 characterization factors, drawn through experiment：Phase
Billy is with the fixed strong model constructed by 9 characterization factors (each respective average of the factor), when the utilization angle of deviation-gradient symbiosis square
During the respective average+TC centers latitude of 9 parameters (T6, T15, T1, T11, T14, T3, T13, T10, T5) in battle array, TC is surely strong
Error is minimum.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710406752.1A CN107230197B (en) | 2017-05-27 | 2017-05-27 | Tropical cyclone objective strength determination method based on satellite cloud image and RVM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710406752.1A CN107230197B (en) | 2017-05-27 | 2017-05-27 | Tropical cyclone objective strength determination method based on satellite cloud image and RVM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107230197A true CN107230197A (en) | 2017-10-03 |
| CN107230197B CN107230197B (en) | 2023-05-12 |
Family
ID=59934635
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710406752.1A Active CN107230197B (en) | 2017-05-27 | 2017-05-27 | Tropical cyclone objective strength determination method based on satellite cloud image and RVM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107230197B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109325960A (en) * | 2018-11-20 | 2019-02-12 | 南京信息工程大学 | A kind of infrared cloud image cyclone analysis method and analysis system |
| CN110059423A (en) * | 2019-04-23 | 2019-07-26 | 中国人民解放军国防科技大学 | Tropical cyclone objective strength determination method based on multi-factor generalized linear model |
| CN110119494A (en) * | 2018-12-31 | 2019-08-13 | 三亚中科遥感研究所 | It is a kind of based on main passive microwave remote sensing observation tropical cyclone determine strong method automatically |
| CN111127365A (en) * | 2019-12-26 | 2020-05-08 | 重庆矢崎仪表有限公司 | HUD distortion correction method based on cubic spline curve fitting |
| CN111695473A (en) * | 2020-06-03 | 2020-09-22 | 中国人民解放军国防科技大学 | Tropical cyclone strength objective monitoring method based on long-time and short-time memory network model |
| CN111862005A (en) * | 2020-07-01 | 2020-10-30 | 自然资源部第二海洋研究所 | Method and system for accurately positioning tropical cyclone center by using synthetic radar image |
| CN115082439A (en) * | 2022-07-22 | 2022-09-20 | 浙江大学 | Tropical cyclone strength determining method, medium and equipment fused with satellite cloud picture space-time information |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6879944B1 (en) * | 2000-03-07 | 2005-04-12 | Microsoft Corporation | Variational relevance vector machine |
| CN102938075A (en) * | 2012-11-29 | 2013-02-20 | 浙江师范大学 | RVM (relevant vector machine) method for maximum wind radius and typhoon eye dimension modeling |
| CN103839243A (en) * | 2014-02-19 | 2014-06-04 | 浙江师范大学 | Multi-channel satellite cloud picture fusion method based on Shearlet conversion |
-
2017
- 2017-05-27 CN CN201710406752.1A patent/CN107230197B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6879944B1 (en) * | 2000-03-07 | 2005-04-12 | Microsoft Corporation | Variational relevance vector machine |
| CN102938075A (en) * | 2012-11-29 | 2013-02-20 | 浙江师范大学 | RVM (relevant vector machine) method for maximum wind radius and typhoon eye dimension modeling |
| CN103839243A (en) * | 2014-02-19 | 2014-06-04 | 浙江师范大学 | Multi-channel satellite cloud picture fusion method based on Shearlet conversion |
Non-Patent Citations (2)
| Title |
|---|
| 王瑾,江吉喜: "热带气旋强度的卫星探测客观估计方法研究" * |
| 鲁小琴;雷小途;余晖;赵兵科;: "基于卫星资料进行热带气旋强度客观估算" * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109325960B (en) * | 2018-11-20 | 2021-07-09 | 南京信息工程大学 | Infrared cloud chart cyclone analysis method and analysis system |
| CN109325960A (en) * | 2018-11-20 | 2019-02-12 | 南京信息工程大学 | A kind of infrared cloud image cyclone analysis method and analysis system |
| CN110119494B (en) * | 2018-12-31 | 2023-03-31 | 三亚中科遥感研究所 | Tropical cyclone automatic strength determining method based on active and passive microwave remote sensing observation |
| CN110119494A (en) * | 2018-12-31 | 2019-08-13 | 三亚中科遥感研究所 | It is a kind of based on main passive microwave remote sensing observation tropical cyclone determine strong method automatically |
| CN110059423B (en) * | 2019-04-23 | 2023-01-31 | 中国人民解放军国防科技大学 | Tropical cyclone objective strength determining method based on multi-factor generalized linear model |
| CN110059423A (en) * | 2019-04-23 | 2019-07-26 | 中国人民解放军国防科技大学 | Tropical cyclone objective strength determination method based on multi-factor generalized linear model |
| CN111127365A (en) * | 2019-12-26 | 2020-05-08 | 重庆矢崎仪表有限公司 | HUD distortion correction method based on cubic spline curve fitting |
| CN111127365B (en) * | 2019-12-26 | 2023-08-29 | 重庆矢崎仪表有限公司 | HUD distortion correction method based on cubic spline curve fitting |
| CN111695473A (en) * | 2020-06-03 | 2020-09-22 | 中国人民解放军国防科技大学 | Tropical cyclone strength objective monitoring method based on long-time and short-time memory network model |
| CN111695473B (en) * | 2020-06-03 | 2023-12-19 | 中国人民解放军国防科技大学 | Tropical cyclone strength objective monitoring method based on long-short-term memory network model |
| CN111862005A (en) * | 2020-07-01 | 2020-10-30 | 自然资源部第二海洋研究所 | Method and system for accurately positioning tropical cyclone center by using synthetic radar image |
| CN111862005B (en) * | 2020-07-01 | 2023-11-17 | 自然资源部第二海洋研究所 | Method and system for precisely positioning tropical cyclone center by utilizing synthetic radar image |
| CN115082439A (en) * | 2022-07-22 | 2022-09-20 | 浙江大学 | Tropical cyclone strength determining method, medium and equipment fused with satellite cloud picture space-time information |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107230197B (en) | 2023-05-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107230197A (en) | Tropical cyclone based on satellite cloud picture and RVM is objective to determine strong method | |
| WO2021135630A1 (en) | Rolling bearing fault diagnosis method based on grcmse and manifold learning | |
| CN104751478B (en) | Object-oriented building change detection method based on multi-feature fusion | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN106295613A (en) | A kind of unmanned plane target localization method and system | |
| CN109871823B (en) | Satellite image ship detection method combining rotating frame and context information | |
| CN103226826B (en) | Based on the method for detecting change of remote sensing image of local entropy visual attention model | |
| CN108447057A (en) | SAR image change detection based on conspicuousness and depth convolutional network | |
| CN104197933A (en) | Method for enhancing and extracting star sliding in telescope field of view | |
| CN108171119B (en) | SAR image change detection method based on residual error network | |
| CN103106658A (en) | Island or reef coastline rapid obtaining method | |
| CN109631912A (en) | A kind of deep space spherical object passive ranging method | |
| CN111695473A (en) | Tropical cyclone strength objective monitoring method based on long-time and short-time memory network model | |
| CN115601661A (en) | Building change detection method for urban dynamic monitoring | |
| CN108959182A (en) | The small feature loss gravitational field modeling method returned based on Gaussian process | |
| CN114565594A (en) | Image anomaly detection method based on soft mask contrast loss | |
| Shang et al. | Spatiotemporal reflectance fusion using a generative adversarial network | |
| Feng et al. | A novel semi-supervised long-tailed learning framework with spatial neighborhood information for hyperspectral image classification | |
| CN113271539B (en) | Indoor target positioning method based on improved CNN model | |
| CN112818777B (en) | Remote sensing image target detection method based on dense connection and feature enhancement | |
| CN109738852A (en) | The distributed source two-dimensional space Power estimation method rebuild based on low-rank matrix | |
| CN113128446A (en) | Human body posture estimation method based on belief map enhanced network | |
| Li et al. | Learning to holistically detect bridges from large-size vhr remote sensing imagery | |
| CN113673011A (en) | Method for intelligently identifying tunnel invasion boundary in operation period based on point cloud data | |
| LU503778B1 (en) | The method for wind speed downscaling of wind farms based on super-resolution reconstruction technology |
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 |