CN108280473A - A kind of main deciduous species remote sensing recognition method in Mount Taishan based on sensitivity spectrum index and SVM - Google Patents
A kind of main deciduous species remote sensing recognition method in Mount Taishan based on sensitivity spectrum index and SVM Download PDFInfo
- Publication number
- CN108280473A CN108280473A CN201810053390.7A CN201810053390A CN108280473A CN 108280473 A CN108280473 A CN 108280473A CN 201810053390 A CN201810053390 A CN 201810053390A CN 108280473 A CN108280473 A CN 108280473A
- Authority
- CN
- China
- Prior art keywords
- svm
- sensitivity spectrum
- spectrum index
- remote sensing
- index
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a kind of main deciduous species remote sensing recognition methods in Mount Taishan based on sensitivity spectrum index and SVM, and this approach includes the following steps：The spectral reflectivity for carrying sampling area pixel respectively builds spectral index using mathematical algorithms, and itself and seeds classification is carried out correlation analysis, and first, each phase filters out 10 sensitivity spectrum indexes, builds 10 variable SVM models；Secondly, 10 sensitivity spectrum index constructions, 10 single argument SVM models are utilized respectively, select the highest 3 groups of variables of wherein precision as best sensitivity spectrum index；Then, 3 variable SVM models are built using best sensitivity spectrum index as conditional attribute；Finally, 3 variable of comparative analysis and 10 variable SVM model seeds recognition effects, determine best identified phase.The present invention can provide technical support for accurately identifying for Mount Taishan seeds with Management offorestry.
Description
Technical field
The present invention relates to a kind of wood recognition methods, specifically, being related to a kind of Thailand based on sensitivity spectrum index and SVM
The main deciduous species remote sensing recognition method in mountain.
Background technology
Mount Taishan is the important tourist attraction in China and typical Temperate Forest Ecosystems area, due to a varied topography, traffic
The reasons such as inconvenience, carrying out Mount Taishan wood recognition using the conventional methods such as field investigation or large aerial photos interpretation has certain difficulty
Degree, and large spatial scale classification is difficult to realize in the short time.Multispectral remote sensing has the characteristics that broad perspectives, period are short, repeatable,
Objects recognition can be carried out by characteristics such as image spectrum, textures, it may extensively using technology is provided for Mount Taishan wood recognition
(Zhou Hongze etc., 2000；Lehman et al., 2014), time and photo choice is to improve drawing essence according to the phenological period characteristic of vegetation
Spend the basis of (Zhang Huanxue etc., 2015).
The selection of Optimum temoral can strengthen the correlation of target seeds and its spectral signature, reduction other information interference (thousand
Bosom is satisfied, and 1998).There is scholar to be applied to the selection of Optimum temoral agriculturally, e.g., is based on HJ-1 satellite images, it is small by building
The multi-temporal NDVI curve data of wave conversion filtering method and the method for moving average, the multi-temporal NDVI curve after foundation is smooth is to rice
(Oryza sativa), corn (Zea mays L), the yield by estimation of three grande culture object of soybean (Glycine max) carry out Optimum temoral choosing
It selects (Irving great etc., 2010).Some scholars e.g. measure time and photo choice small emerging applied to forestry research by field spectroradiometer
Pacify 9, ridge chief species canopy spectra data, the season in visible light and near infrared band is planted according to canopy spectra Changeement tree
Phase change feature and difference, the results showed that, regular variation is presented because of the change in season for deciduous species spectral signature, and normal
Change unobvious greenery kind spectral signature year, more aspect data, which carry out classification, can obtain best effects (Xu Guangcai etc., 2013).
In terms of remote sensing image classification, when comprehensive NDVI time serieses, Optimum temoral spectral signature and textural characteristics are to Landsat 8
Sequence remote sensing images cotton (Gossypium spp) is classified, and overall classification accuracy carries out shadow 90% or more, using rough set
As feature selecting can effectively improve nicety of grading (Wang Wen waits quietly, 2017).
In seeds remote sensing recognition method, maximum likelihood estimate (Martin et al., 1998), neural network (field
Wait quietly, 2017), support vector cassification method (wait quietly, 2017) etc. achieve preferable recognition result (Li Deren etc.,
2013).Wherein, spectrum and data texturing based on high-resolution remote sensing image, with maximum likelihood method, mahalanobis distance, nerve
Network, support vector machine classification method, to South Sinkiang basin walnut (Juglans regia), jujube (Ziziphus jujuba
Mill), bergamot pear (Pyrus spp) and apple (Malus pumila) 4 planting fruit-trees carry out remote sensing recognition, and precision is respectively
58.32%, 58.70%, 68.70% and 69.71%, support vector machine method nicety of grading highest (Yue Jun etc., 2015).And
Based on support vector machine classifier, using HSI2 grades of product data of HJ-1A stars to the forest wolf tree of Daxing'an Mountainrange Tahe Region
Kind of group is identified research, after initial data and differential transform processing the seeds of data classify overall accuracy reach 80% with
On, Kappa coefficients are in 0.71 or more (Wang Lu etc., 2015), but by willow (Salix babylonica), poplar during this
The similar seeds of the curves of spectrum such as tree (PopulusL), aspen (Pobulus davidiana) are classified as one kind, and it is accurate that seeds are not implemented
Identification；Based on support vector machine classifier, using Landsat-8 images spectrum, texture information to forestry bureau of Wangqing day
Right Forest is classified, the results showed that, best using radial kernel function svm classifier precision, overall classification accuracy is reachable
89.58% (Li Mengying etc., 2017), but only realize the classification of broad-leaf forest, coniferous forest, theropencedrymion.It is automatic based on cellular
Machine and BP neural network algorithm carry out the classification of Landsat-TM remote sensing image Forest Types and compare, and are based on cellular automata classification side
The overall classification accuracy of method is that 88.71%, Kappa coefficients are 0.8291；The overall classification accuracy of BP neural network algorithm is
86.67%, Kappa coefficient are 0.7984 (field is waited quietly, 2017), also only realize coniferous forest, broad-leaf forest and theropencedrymion
Information identifies, does not carry out specific seeds classification.Studies have shown that the remote sensing wood recognition effect of support vector machines is preferable, but for
Accurately identifying for seeds is also fewer, and the spectral information in existing multispectral remote sensing wood recognition directly uses band value,
Spectral index can expand the fine difference (Wang Ling, 2012) between spectrum, improve wood recognition precision.
Invention content
The purpose of the present invention is to provide a kind of main deciduous species remote sensing in Mount Taishan based on sensitivity spectrum index and SVM to know
Other method.This method carries out time and photo choice using spectral index structure SVM, obtains better recognition effect.Mountain is dispersed with big face
Broad-leaved deciduous forest of the product based on oak class, Chinese scholartree, is the Typical Representative of Temperate Forest Ecosystems, wherein with Quercus acutissima, locust tree point
Cloth area is most wide, and the phenological period is similar.Therefore, the present invention is based on ZY-1 02C and ZY-3 remote sensing images by taking Quercus acutissima, locust tree as an example
Spectroscopic data is extracted, sensitivity spectrum index is established, is identified using two seeds of SVM algorithm pair, determine the best knowledge of two seeds
Other phase, the remote sensing recognition to study the subsequently other seeds in area provide basis and reference frame.
Its specific technical solution is：
A kind of main deciduous species remote sensing recognition method in Mount Taishan based on sensitivity spectrum index and SVM, includes the following steps：
The spectral reflectivity for carrying sampling area pixel respectively builds spectral index using mathematical algorithms, and by itself and tree
Kind classification carries out correlation analysis, and first, each phase filters out 10 sensitivity spectrum indexes, builds 10 variable SVM models；Its
It is secondary, 10 sensitivity spectrum index constructions, 10 single argument SVM models are utilized respectively, the highest 3 groups of variables of wherein precision is selected and makees
For best sensitivity spectrum index；Then, 3 variable SVM models are built using best sensitivity spectrum index as conditional attribute；Finally,
3 variable of comparative analysis and 10 variable SVM model seeds recognition effects, determine best identified phase.
Quercus acutissima, locust tree sample area spectral information (pixel value) are extracted according to shp in step 1, ERDAS
(1) the shp files with required attribute field are imported by import, is exported as arcinfo formats；
(2) arcinfo formats are switched to by img formats by vector-vector to raster；
(3) transformed img formatted files and the image of pixel value to be extracted are subjected to band combination, export ASCII texts
Part deletes 0 value.(erdas band combination methods：Modeler-model maker-CONDITIONAL functions)
Step 2, sensitivity spectrum index construction and screening：Correlation analysis determines sensitivity spectrum index, to build SVM10；Than
Compared with single argument SVM model accuracies, best sensitivity spectrum index is determined, to build SVM3。
Step 3 determines sensitive band：Spectral Characteristics Analysis is carried out to each wave band of each phase, using MATLAB by each wave band
Reflectance value carries out correlation analysis with seeds type, show that the most sensitive wave band of each phase is shown as：12 days the 3rd May, 4 waves
Section；September the 4th wave band on the 29th；December 7 the 4th wave band.To build sensitive band SVM.
Step 4 builds SVM classifier using the tool boxes libSVM of MATLAB.
Further, step 4 is specially：
(1) tool boxes libSVM are downloaded；
(2) setting MATLAB search work catalogues are libsvm-3.12；
(3) c/c++ compilers are selected, and inputs " make " in MATLAB and is compiled the matlab of libsvm；
(4) data set of MATLAB versions is loaded in matlab command windows input " load heart_scale "；
(5) it models and predicts by following two function command incomes, obtain SVM modeling accuracies, precision of prediction, prediction class
Not：
Model=svmtrain (heart_scale_label, heart_scale_inst),
[predict_label, accuracy]=svmpredict (heart_scale_label, heart_scale_
inst,model)。
Further, SVM classifier is built using radial kernel function.
Further, preferred by parameter, it obtains for Quercus acutissima, the highest SVM parameters of locust tree nicety of grading：Degree=3,
Gamma=0.5, Coef0=0.001, Epsilon=0.001, C=1, Nu=0.5, Shrinking=1, p=1.
Compared with prior art, beneficial effects of the present invention：
Relative to sensitive band, wood recognition precision can be effectively improved using sensitivity spectrum index construction SVM, locust tree is known
The raising of other precision is particularly evident.3 variable SVM entirety recognition effects are better than 10 variable SVM.May 12 (spring) is Mount Taishan Quercus acutissima
With the Optimum temoral of locust tree identification.The present invention can provide technical support for accurately identifying for Mount Taishan seeds with Management offorestry.
Description of the drawings
Fig. 1 is that sample distinguishes Butut；
Fig. 2 is that the present invention is based on the flows of the main deciduous species remote sensing recognition method in the Mount Taishan of sensitivity spectrum index and SVM
Figure.
Specific implementation mode
Technical scheme of the present invention is described in more detail with specific embodiment below in conjunction with the accompanying drawings.
1 materials and methods
1.1 research area's overviews
Mount Taishan is located in middle Shandong Province, and geographical location is 116 ° 50 '~117 ° 12 ' of east longitude, 36 ° 11 '~36 ° 31 ' of north latitude,
Height above sea level 1545m.Mount Taishan plant resources are abundant, and vegetation coverage is up to 90% or more, the nearly 10000hm of forest land area2, forest cover
For rate up to 81.5%, Forest Types include coniferous forest, broad-leaved deciduous forest, theropencedrymion etc., abundant vegetation resources make Mount Taishan at
For the huge ecological protective screen and important germplasm resource bank.
1.2 data
With resource No.1 02C (ZY-1 02C) on Mays 12nd, 2014, resource three (ZY-3) September in 2014 29 days and
December in 2014, the multi-spectrum remote sensing image of three phases on the 7th was data source.
Wherein, ZY-1 02C images have 3 wave bands, spatial resolution 10m；ZY-3 images have 4 wave bands, spatial discrimination
Rate 5.8m, specific spectral range are shown in Table 1.Since ZY-3 satellites are without in May, 2014 data, 3 wave bands and ZY- of ZY-1 02C
2~4 wave band spectral ranges in 3 are identical, are comparable.
1 image spectral range of table
Table 1 Spectral range of image
Note：For convenience of control, it is by each wave band number volumes of ZY-1 02C：2,3,4 wave band
Note:,the ZY-1 02C bands are numbered as 2,3,4bands to facilitate the
comparison.
Due to Mount Taishan vegetation coverage height, summer annual growth is vigorous, and SPECTRAL DIVERSITY is small between seeds, not easy to identify, because
The germination period May 12 (spring) of this selection deciduous broadleaf tree, 29 (autumn) of leaf fall period September and without (winter) three 7 days December leaf phase
A phase is identified.
With sampling point, uniform, seeds are typically principle, choose 119, Quercus acutissima sample area, 88, locust tree sample area, distribution such as Fig. 1.Cause
Each sample area includes multiple pixels, and there is also differences between pixel, and to take into account difference, selections pixel is experimental considerations unit.Quercus acutissima sample
Area includes pixel 2550 altogether, and locust tree sample area includes pixel 1635.According to the difference of spectral index, with equidistant sampling method,
Each seeds choose 2/3 pixel as SVM modeling samples, wherein Quercus acutissima 1700, locust tree 1090；1/3 pixel is as verification sample
Sheet, wherein Quercus acutissima 850, locust tree 545.
1.3 research method
Remote sensing image has carried out the pretreatments such as atmospheric correction, geometric correction, geometric accurate correction using preceding, wherein to reduce
Influence of the Mount Taishan complicated landform to data goes back emphasis and has carried out terrain radiant correction.As shown in Figure 2.
1.3.1 sensitivity spectrum index construction and screening
The spectral reflectivity for carrying sampling area pixel respectively, using mathematical algorithms structure spectral index (such as table 2), and will
It carries out correlation analysis with seeds classification, and first, each phase filters out 10 sensitivity spectrum indexes, builds 10 variable SVM moulds
Type；Secondly, 10 sensitivity spectrum index constructions, 10 single argument SVM models are utilized respectively, the highest 3 groups of changes of wherein precision are selected
Amount is used as best sensitivity spectrum index；Then, 3 variable SVM models are built using best sensitivity spectrum index as conditional attribute；
Finally, 3 variable of comparative analysis and 10 variable SVM model seeds recognition effects, determine best identified phase (Fig. 2).
2 spectral index construction method of table
Table 2 Constuction methods of spectral indices
Note：Ri,Rj(i, j=1,2,3,4)：Each wave band reflectivity
Note:Ri,Rj(i, j=1,2,3,4):Reflectances of the bands.
1.3.2 algorithm of support vector machine
Algorithm of support vector machine is a kind of mode identification method of the foundation in Statistical Learning Theory, earliest by Vapnik in
It proposes within 1995, can be used for data classification and regression analysis, be mainly used in area of pattern recognition.The key of support vector machines
It is kernel function, i.e., sample data is mapped to by kernel function by high-dimensional feature space, linear inner product fortune by original feature space
Non-linearization is calculated, searches out optimal hyperlane in feature space so that class interval maximization, sample data is accurately divided
Class (Cortes C et al., 1995).
Svm classifier often has with kernel function at present:Linear kernel function (Linear Kernel), Polynomial kernel function
(Polynomial Kernel), radial kernel function (RBF：Radial Basis Function), Sigmoid kernel functions
(Sigmoid Kernel) (Huang He etc., 2016), studies have shown that using the SVM classifier classification results of radial Kernel
Preferably (Luo Jiancheng etc., 2002), so the present invention selects radial kernel function to build SVM classifier.
2 results and analysis
2.1 sensitivity spectrum index analysis
It is respectively using each phase spectral index quantity constructed by 2 method of table：May 12 90；September 29 days 166；12
The moon 7 166.By correlation analysis, each phase sensitivity spectrum index such as table 3 is filtered out.
3 sensitivity spectrum index of table
Table 3 Sensitive spectral indices
Note：Si(i=2,3,4), Ai(i=1,2,3,4), Wi(i=1,2,3,4) respectively represent on May 12nd, 2014,
September in 2014 29 days, each wave band reflectivity of image on December 7th, 2014
Note:Si(i=2,3,4), Ai(i=1,2,3,4), Wi(i=1,2,3,4) represent the images
reflectance of each band in May 12,2014,September 29,2014 and December 7,2014
respectively.
May 12 secondly sensitivity spectrum index with seeds type correlation highest is December 7 0.8007 or more,
Correlation is up to 0.7506, and minimum 0.6913, September sensitivity spectrum index on the 29th and seeds type correlation are relatively low, highest
Correlation is only 0.5244 (table 3).The most sensitive wave band of each phase is shown as：12 days the 3rd May, 4 wave bands, i.e. S3、 S4；September 29
Day the 4th wave band, i.e. A4；December 7 the 4th wave band, i.e. W4。
By comparing single argument SVM model accuracies, the best sensitivity spectrum index of each phase is respectively：May 12 Quercus acutissima
For X2、X3、X6, locust tree X1、X9、X10；September Quercus acutissima on the 29th is X3、X4、X6, locust tree X1、X2、X8；December 7, Quercus acutissima was X1、
X6、X8, locust tree X3、X4、X5(table 3).
2.2 accuracy of identification are analyzed
To ensure that each phase recognition result is comparable, makes SVM construction methods in same level, use radial kernel letter
Under the premise of number ensures that recognition effect is optimal, single argument is consistent with the selection of each multivariable SVM model parameters, such as table 4.
4 supporting vector machine model parameter of table
Table 4 The model parameter of support vector machines
Each phase accuracy of identification of 5 Quercus acutissima of table, locust tree
Table 5 Different temporal recognition accuracy of Quercus acutissima
and Robinia pseudoacacia
Note：SVM10、SVM3It represents using 10 sensitivity spectrum indexes, 3 best sensitivity spectrum indexes as conditional attribute
SVM models
Note:SVM10、SVM3are SVM models with ten sensitive spectral indices、
three best sensitive spectral indices as conditional attributes
To have more comparativity, SVM is built using sensitive band as conditional attribute, is conditional attribute with sensitivity spectrum index
SVM carries out model accuracy comparison, comparing result such as table 5.Following analysis is with sensitivity spectrum index SVM3Or SVM10Best identified
Subject to precision.As seen from Table 5, it is slightly above sensitive band using the accuracy of identification of sensitivity spectrum exponent pair Quercus acutissima, wherein May
The 3rd, 4 wave band (S of ratio on the 12nd3、S4) 3.06%, 0.47% is averagely improved respectively；September the 4th wave band (A of ratio on the 29th4) averagely carry
It is high by 1.88%；December 7 the 4th wave band (W of ratio4) averagely improve 2.12%.Utilize the identification of sensitivity spectrum exponent pair locust tree
Precision is apparently higher than sensitive band, wherein May 12 the 3rd, 4 wave band (S of ratio3、S4) averagely improve 5.32% respectively,
3.67%；September the 4th wave band (A of ratio on the 29th4) averagely improve 15.90%；December 7 the 4th wave band (W of ratio4) averagely improve
15.52%.May 12 the 3rd wave band (S3) overall recognition accuracy is relatively low with the 4th wave band (S4), further illustrate May 12 most
Sensitive band is the 4th wave band (S4), it is consistent with the performance of other two phases sensitive bands.For the totality of Quercus acutissima, two seeds of locust tree
Accuracy of identification, sensitivity spectrum index are higher than sensitive band, wherein the 3rd, 4 wave band (S of ratio on May 123、S4) averagely improve
3.95%, 1.83%；September the 4th wave band (A of ratio on the 29th4) averagely improve 8.24%；December 7 the 4th wave band (W of ratio4) averagely carry
It is high by 14.82%.Illustrate that sensitivity spectrum index has more advantage in seeds remote sensing recognition than sensitive band, identification essence can be improved
Degree.In addition, S on May 122, September A on the 29th1、A2、A3With W on December 73Structure SVM overall recognition accuracy be respectively：
57.63%, 60.12%, 54.28%, 60.86%, 58.99%, with each wave band averaged spectrum reflectivity, each wave band and seeds
Type correlation size (table 4) shows unanimously, i.e., averaged spectrum reflectivity is bigger, correlation is bigger, and wood recognition precision is got over
It is high.
Relative to Quercus acutissima, the sensitivity spectrum index of locust tree becomes apparent from than the raising of sensitive band SVM precision, and reason may be fiber crops
Oak phenology feature is apparent, stronger with spectral signature correlation, and sensitive band can be realized more high-precision as seeds characteristic information
Wood recognition is spent, therefore relative to sensitive band, sensitivity spectrum index SVM identification Quercus acutissima precision improves unobvious.Sensitivity spectrum
Index construction SVM improves locust tree etc. with the lower seeds precision of spectral signature correlation particularly evident.
In wood recognition using sensitivity spectrum index, under different phases and SVM building modes, Quercus acutissima modeling accuracy is high
In 80%, locust tree September modeling accuracy on the 29th is relatively low, and in addition two phases are above 70%, and reason may be, and be by the end of September late summer
Phase, Quercus acutissima, Growth of Blaek Locust are vigorous, and vegetation coverage is high, SPECTRAL DIVERSITY unobvious between vegetation.December 7 SVM3Overall identification
Precision is 84.86%, is higher than SVM10(77.6%), other two phase SVM3Overall recognition accuracy and SVM10It is not much different, and
SVM10Operand is big, SVM3It is less to be related to variable, calculates simple.To have more comparativity, the present invention is with SVM3Recognition result is made most
Whole comparative analysis simultaneously carries out Mount Taishan Quercus acutissima, locust tree spatial distribution inverting.It can be seen that 3 phase SVM3For Quercus acutissima, locust tree
Overall recognition accuracy is respectively：On May 12 89.24%；September 29 days 66.81%；On December 7 84.86%.Wherein, May 12 days
Precision highest, and the Quercus acutissima accuracy of identification of 3 phase is above locust tree.
3 conclusions and discussion
The present invention is based on the Multi-spectral Remote Sensing Data of ZY-1 tri- phases of 02C and ZY-3, according to seeds phenological period characteristic,
By taking Quercus acutissima, locust tree as an example, using sensitivity spectrum index construction SVM models, the main deciduous broadleaf tree identification in Mount Taishan has been inquired into
Optimum temoral realizes seeds remote sensing recognition.
The present invention is based on the Multi-spectral Remote Sensing Data of ZY-1 tri- phases of 02C and ZY-3, according to seeds phenological period characteristic,
By taking Quercus acutissima, locust tree as an example, using sensitivity spectrum index construction SVM models, the main deciduous broadleaf tree identification in Mount Taishan has been inquired into
Time and photo choice realizes seeds remote sensing recognition.
1) Spectral Characteristics Analysis is carried out by standard of averaged spectrum reflectance value：Compare when each, May 12 highest；Respectively
Wave band compares, and 3 phase peak shows as the 4th wave band；Two seeds compare, and Quercus acutissima is totally higher than locust tree.The sensitivity of each phase
Wave band is：12 days the 3rd May, 4 wave bands；September the 4th wave band on the 29th；December 7 the 4th wave band, wherein May 12, the 4th wave band was
Most sensitive wave band.Each phase sensitive band performance is consistent, concentrates on 3,4 wave bands.
2) the SVM wood recognition precision built as conditional attribute using sensitivity spectrum index is apparently higher than sensitive band.
More other two phases, May 12 sensitive band, sensitivity spectrum index and the equal highest of seeds type correlation, wood recognition precision
Highest.Therefore, May 12 was the Optimum temoral identified to the main deciduous broadleaf tree in Mount Taishan in 3 phase, and accuracy of identification is
89.24%.
3) ZY-1 02C spatial resolutions are less than ZY-3, and 12 days Mays in 2014 of ZY-1 02C are for Quercus acutissima and locust tree
Accuracy of identification be above ZY-3 (table 5), it is in 3 phase to the Optimum temorals of two wood recognitions to more fully illustrate May 12.
May 12 to the best results of Mount Taishan Quercus acutissima, locust tree identification, traced it to its cause, it may be possible to which May is Deciduous And Broadleaf Trees
Kind germination peak period, Quercus acutissima, locust tree blade differ greatly, and high all more much bigger than locust tree, the SPECTRAL DIVERSITY of Quercus acutissima trunk, tree crown, tree
Obviously, Quercus acutissima is made to be more easily identified, precision is higher than locust tree.Wood recognition type of the present invention only relates to Quercus acutissima, locust tree, needs rear
More Mount Taishan deciduous broadleaf trees are added in continuous research.
The foregoing is only a preferred embodiment of the present invention, protection scope of the present invention is without being limited thereto, it is any ripe
Those skilled in the art are known in the technical scope of present disclosure, the letter for the technical solution that can be become apparent to
Altered or equivalence replacement are each fallen in protection scope of the present invention.
Claims (4)
1. a kind of main deciduous species remote sensing recognition method in Mount Taishan based on sensitivity spectrum index and SVM, which is characterized in that including
Following steps：
Quercus acutissima, locust tree sample area spectral information are extracted according to shp in step 1, ERDAS
(1) the shp files with required attribute field are imported by import, is exported as arcinfo formats；
(2) arcinfo formats are switched to by img formats by vector-vector to raster；
(3) transformed img formatted files and the image of pixel value to be extracted are subjected to band combination, export ascii text file, deletes
Except 0 value；
Step 2, sensitivity spectrum index construction and screening：Correlation analysis determines sensitivity spectrum index, to build SVM10；Compare list
Variable SVM model accuracies determine best sensitivity spectrum index, to build SVM3；
Step 3 determines sensitive band：Spectral Characteristics Analysis is carried out to each wave band of each phase, is reflected each wave band using MATLAB
Rate value carries out correlation analysis with seeds type, show that the most sensitive wave band of each phase is shown as：12 days the 3rd May, 4 wave bands；9
Month the 4th wave band on the 29th；December 7 the 4th wave band；To build sensitive band SVM；
Step 4 builds SVM classifier using the tool boxes libSVM of MATLAB.
2. the main deciduous species remote sensing recognition method in the Mount Taishan according to claim 1 based on sensitivity spectrum index and SVM,
It is characterized in that, step 4 is specially：
(1) tool boxes libSVM are downloaded；
(2) setting MATLAB search work catalogues are libsvm-3.12；
(3) c/c++ compilers are selected, and inputs " make " in MATLAB and is compiled the matlab of libsvm；
(4) data set of MATLAB versions is loaded in matlab command windows input " load heart_scale "；
(5) it models and predicts by following two function command incomes, obtain SVM modeling accuracies, precision of prediction, prediction classification：
Model=svmtrain (heart_scale_label, heart_scale_inst),
[predict_label, accuracy]=svmpredict (heart_scale_label, heart_scale_inst,
model)。
3. the main deciduous species remote sensing recognition method in the Mount Taishan according to claim 1 based on sensitivity spectrum index and SVM,
It is characterized in that, building SVM classifier using radial kernel function.
4. the main deciduous species remote sensing recognition method in the Mount Taishan according to claim 1 based on sensitivity spectrum index and SVM,
It is characterized in that, it is preferred by parameter, it obtains for Quercus acutissima, the highest SVM parameters of locust tree nicety of grading：Degree=3,
Gamma=0.5, Coef0=0.001, Epsilon=0.001, C=1, Nu=0.5, Shrinking=1, p=1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810053390.7A CN108280473A (en) | 2018-01-19 | 2018-01-19 | A kind of main deciduous species remote sensing recognition method in Mount Taishan based on sensitivity spectrum index and SVM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810053390.7A CN108280473A (en) | 2018-01-19 | 2018-01-19 | A kind of main deciduous species remote sensing recognition method in Mount Taishan based on sensitivity spectrum index and SVM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108280473A true CN108280473A (en) | 2018-07-13 |
Family
ID=62804141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810053390.7A Pending CN108280473A (en) | 2018-01-19 | 2018-01-19 | A kind of main deciduous species remote sensing recognition method in Mount Taishan based on sensitivity spectrum index and SVM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108280473A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102034A (en) * | 2018-09-04 | 2018-12-28 | 刘勇峰 | Text plays the accurate matching method and computer storage medium of walnut |
| CN111488822A (en) * | 2020-04-09 | 2020-08-04 | 北华航天工业学院 | Tree species information identification method based on full spectrum segment correlation analysis algorithm |
| CN113111763A (en) * | 2021-04-08 | 2021-07-13 | 洛阳师范学院 | Method and device for establishing spectral volume index and method for identifying tree species |
-
2018
- 2018-01-19 CN CN201810053390.7A patent/CN108280473A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102034A (en) * | 2018-09-04 | 2018-12-28 | 刘勇峰 | Text plays the accurate matching method and computer storage medium of walnut |
| CN109102034B (en) * | 2018-09-04 | 2021-12-21 | 刘勇峰 | Precise pairing method of Chinese and playing walnuts and computer storage medium |
| CN111488822A (en) * | 2020-04-09 | 2020-08-04 | 北华航天工业学院 | Tree species information identification method based on full spectrum segment correlation analysis algorithm |
| CN113111763A (en) * | 2021-04-08 | 2021-07-13 | 洛阳师范学院 | Method and device for establishing spectral volume index and method for identifying tree species |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Karlson et al. | Assessing the potential of multi-seasonal WorldView-2 imagery for mapping West African agroforestry tree species | |
| Dian et al. | Spectral and texture features combined for forest tree species classification with airborne hyperspectral imagery | |
| Atzberger et al. | Spatially constrained inversion of radiative transfer models for improved LAI mapping from future Sentinel-2 imagery | |
| Ferreira et al. | Retrieving structural and chemical properties of individual tree crowns in a highly diverse tropical forest with 3D radiative transfer modeling and imaging spectroscopy | |
| Liu et al. | Comparison of tasselled cap transformations based on the selective bands of Landsat 8 OLI TOA reflectance images | |
| George et al. | Forest tree species discrimination in western Himalaya using EO-1 Hyperion | |
| Sun et al. | Monitoring maize lodging grades via unmanned aerial vehicle multispectral image | |
| Wang et al. | Spatio-temporal spectral unmixing of time-series images | |
| Man et al. | Light detection and ranging and hyperspectral data for estimation of forest biomass: a review | |
| US9639755B2 (en) | Automated compound structure characterization in overhead imagery | |
| Zhang et al. | Extraction of tree crowns damaged by Dendrolimus tabulaeformis Tsai et Liu via spectral-spatial classification using UAV-based hyperspectral images | |
| CN110645961B (en) | Forest resource dynamic change detection method based on remote sensing and NDVI | |
| Marcinkowska-Ochtyra et al. | Subalpine and alpine vegetation classification based on hyperspectral APEX and simulated EnMAP images | |
| CN106960221A (en) | A kind of hyperspectral image classification method merged based on spectral signature and space characteristics and system | |
| CN108280473A (en) | A kind of main deciduous species remote sensing recognition method in Mount Taishan based on sensitivity spectrum index and SVM | |
| Dhumal et al. | Classification of Crops from remotely sensed Images: AnOverview | |
| Toniol et al. | Potential of hyperspectral metrics and classifiers for mapping Brazilian savannas in the rainy and dry seasons | |
| CN116310840A (en) | Winter wheat remote sensing identification method integrating multiple key weather period spectral features | |
| Li et al. | Blind nonlinear hyperspectral unmixing based on constrained kernel nonnegative matrix factorization | |
| Quan et al. | Tree species classification in a typical natural secondary forest using UAV-borne LiDAR and hyperspectral data | |
| CN105160623A (en) | Unsupervised hyperspectral data dimension reduction method based on block low-rank tensor model | |
| CN116543316B (en) | Method for identifying turf in paddy field by utilizing multi-time-phase high-resolution satellite image | |
| Cheng et al. | Two-stream isolation forest based on deep features for hyperspectral anomaly detection | |
| CN107831168A (en) | The method that remote sensing technology measures paddy field shelter-forest protection effect | |
| CN105224961A (en) | A kind of diffuse reflectance infrared spectroscopy of high resolution extracts and matching process |
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 |