CN110346312A - Winter wheat fringe head blight recognition methods based on Fei Shi linear discriminant and support vector machines technology - Google Patents
Winter wheat fringe head blight recognition methods based on Fei Shi linear discriminant and support vector machines technology Download PDFInfo
- Publication number
- CN110346312A CN110346312A CN201910652829.2A CN201910652829A CN110346312A CN 110346312 A CN110346312 A CN 110346312A CN 201910652829 A CN201910652829 A CN 201910652829A CN 110346312 A CN110346312 A CN 110346312A
- Authority
- CN
- China
- Prior art keywords
- feature
- sample
- support vector
- data
- linear discriminant
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/55—Specular reflectivity
Landscapes
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The present invention relates to the winter wheat fringe head blight recognition methods based on Fei Shi linear discriminant and support vector machines technology, and solving remote sensing fields compared with prior art there is no the defect that wheat scab research is carried out with Dan Suiwei carrier.The present invention is the following steps are included: data acquisition；Field hyperspectrum data prediction；The selection of Modelling feature；Fei Shi linear discriminant combines the foundation of model with support vector machines；Obtain wheatear EO-1 hyperion recognition result.The present invention realizes the head blight identification for being directed to single fringe wheat.
Description
Technical field
The present invention relates to winter wheat fringe head blight identification technology fields, are specifically based on Fei Shi linear discriminant and support
The winter wheat fringe head blight recognition methods of vector machine technology.
Background technique
At present high spectrum resolution remote sensing technique wheat diseases monitoring field research than wide, to the distant of wheat diseases
Sense monitoring and prediction aspect, research are concentrated mainly on stripe rust of wheat, the remote sensing monitoring and Study on Forecasting Method of wheat powdery mildew
On.However, the stripe rust of wheat that compares, the research that the research of powdery mildew infects wheat Fusariumsp currently with hyperspectral technique
Less, main research concentrates wheat seed nosomycosis classification in laboratory conditions upper or is only to wheat fringe portion Fusariumsp
It is identified, is not yet classified differentiation to the degree of its state of an illness.
It is found by literature survey, under the conditions of winter wheat list fringe, head blight spectral response characteristics not yet grinding by system
Study carefully conclusion.Since the observation condition of ground spectrum is controllable, Instrument Development is mature, more noise and interference can be avoided, passed through
Data are compared and meticulously analyzes and handles, some mechanism Journal of Sex Research can be carried out to crop disease remote sensing monitoring.Meanwhile this
Class research is also aviation/space remote sensing scale disease monitoring basis.
Non-imaged hyperspectral technique has the characteristics that spectral resolution is high, wave band number is more, informative, and near-earth is adopted
The non-imaged high-spectral data of collection is influenced by atmosphere and external environment, and small, signal-to-noise ratio is high, closer to the real spectrum of atural object.Cause
This under the conditions of single fringe, studies the disease spectral response characteristics of wheat scab using non-imaged spectral technique, small to the subsequent winter
Wheat head blight Study of recognition is of great significance.
Summary of the invention
It there is no the purpose of the present invention is to solve remote sensing fields in the prior art and gibberella saubinetii carried out with Dan Suiwei carrier
The defect of disease research, provides a kind of winter wheat fringe head blight recognition methods based on Fei Shi linear discriminant and support vector machines technology
To solve the above problems.
To achieve the goals above, technical scheme is as follows:
A kind of winter wheat fringe head blight recognition methods based on Fei Shi linear discriminant and support vector machines technology, including it is following
Step:
Data acquisition: the measured winter wheat list fringe spectroscopic data of spectrometer is obtained, wheat single ear spectroscopic data includes every plant
The high-spectral data in fringe front, side and upright three angles；
Field hyperspectrum data prediction: winter wheat list fringe spectroscopic data is standardized, and carries out one respectively
Rank differential calculation, continuum removal calculate and common vegetation index calculates, and obtains corresponding differential characteristics, continuum removal spy
Sign and vegetation index feature；
The selection of Modelling feature: it screens tight with head blight severity correlation height and extremely significant spectral signature and three kinds
The significant feature of weight degree class inherited；
Fei Shi linear discriminant combines the foundation of model with support vector machines: passing through ground high spectrometry and severity
The data building Fei Shi linear discriminant of investigation combines model with support vector machines；
It obtains wheatear EO-1 hyperion recognition result: using the optimal characteristics collection of field hyperspectrum data as input variable, dividing
Not Shu Ru Fei Shi linear discriminant with support vector machines combine model, obtain the recognition result of winter wheat fringe head blight.
The field hyperspectrum data prediction the following steps are included:
High-spectral data standardization:
The healthy sample spectrum data of a-quadrant are chosen as master sample, and with disease sample spectrum data and health
Sample spectrum data use the mean value of the healthy sample spectrum data of a-quadrant divided by B area and C area health respectively as benchmark
The mean value of sample spectrum data obtains two ratio curves, this two ratio curves are used to reflect the difference of two groups of spectral measurements
Property, the ratio R atio under i wavelengthiCalculation formula are as follows:
Wherein i indicates that wavelength, ∑ refer to reflectivity the sum of of institute's unsoundness sample at wavelength i in region,
Ref indicates reflectivity, and A, B and C respectively represent a-quadrant, B area and the region C；
The original spectral data for each sample that B area and the region C are obtained and the multiplication of corresponding ratio curve to get
Spectroscopic data after to standardization, the standardized calculation formula under i wavelength are as follows:
Wherein,Indicate the reflectivity of B area or C area sample spectrum at wavelength i,Indicate mark
Reflectivity after standardization；
The first derivative spectra data are obtained, difference method approximate calculation, calculation formula are used to the spectrum after standardization
It is as follows:
ρ'(λi)=[ρ (λi+1)-ρ(λi-1)]/(2 Δ λ),
Wherein, λiFor each band wavelength, ρ ' (λi) it is wavelength XiFirst differential, Δ λ are wavelength Xi-1To λiInterval, then
Acquire first differential feature；
Obtain continuum and remove feature: the 550nm-750nm wave band of chosen spectrum carries out continuum removal and obtains light relatively
Reflectivity is composed, the wave band position for finding out wave band depth respectively on this basis, absorbing peak area, waveband width and Absorption Characteristics
It sets；
It obtains vegetation index: the vegetation index of several common plant stress researchs is rule of thumb selected, according to each plant
Calculation formula is answered to obtain corresponding index characteristic by exponent pair.
The selection of the Modelling feature the following steps are included:
First differential feature, continuum removal three kinds of spectral signatures of feature and vegetation index that data prediction is obtained
Correlation analysis is carried out with the severity of winter wheat head blight respectively, related coefficient and P-value value is obtained, utilizes threshold method
It filters out and head blight high correlation R2>0.49 and extremely significant related P-value<0.001 spectral signature；
Respectively to health, slight, severe three classes sample combination of two, filtered out using the independent sample T method of inspection three
The feature of significant difference is presented in a category combinations；
Two Spectral Properties collection of two above conditional filtering is taken into intersection, is obtained sensitive to head blight and different in three classes
The optimal Spectral Properties collection that there were significant differences between disease severity class；
It repeats the above steps to obtain respectively from front, side and upright three angles and is suitble to the optimal of its head blight identification
Spectral Properties collection.
The Fei Shi linear discriminant foundation of model is combined with support vector machines the following steps are included:
She Dingfeishi linear discriminant model: the feature of higher-dimension is subjected to dimensionality reduction using Fei Shi linear discriminant model, by data
The one-dimensional space is projected to, finding one makes within-cluster variance minimum, the maximum best projection direction of inter _ class relationship；
Setting supporting vector machine model: transforming to higher dimensional space for the input space by nonlinear transformation, then new empty
Between in seek optimal classification surface, so that the linear problem that the nonlinear problem in former space is converted in new space be solved；
She Dingfeishi linear discriminant combines model with support vector machines: being carried out data using Fei Shi linear discriminant method
Dimensionality reduction, input variable of the obtained one-dimensional characteristic feature vector as supporting vector machine model；
Winter wheat fringe head blight sample data is divided into health, slight and severe three classes, wherein randomly selecting sample data
As training sample, training sample combines model with support vector machines as input, Xun Lianfeishi linear discriminant.
The acquisition wheatear EO-1 hyperion recognition result the following steps are included:
First differential feature that fringe front, side and upright three angles are screened respectively, continuum removal are special
Sign and vegetation index combine the optimal Spectral Properties collection to form single fringe scale different angle head blight；
Input variable of the Spectral Properties collection that each angle is obtained as model after training, the Fei Shi after training linearly sentence
Supporting vector machine model after other model and training exports serious to head blight from front, side and upright three angles respectively
Spend the result of identification.
The She Dingfeishi linear discriminant model the following steps are included:
N number of single fringe high-spectral data that original is obtained is defined as set X, includes N number of higher-dimension sample x in X1、x2、...、xn,
All samples are divided into health, slight and 3 class of severe according to head blight coincident with severity degree of condition, wherein N1It is a to belong to W1The sample of class
It is denoted as subset X1, N2It is a to belong to W2The sample of class is denoted as subset X2, N3It is a to belong to W3The sample of class is denoted as subset X3, Different categories of samples
Mean vector mi:
Dispersion matrix between sample classes SiAre as follows:
Total within class scatter matrix SwAre as follows:
Sw=S1+S2+S3；
Matrix between samples SbAre as follows:
Wherein m is the mean vector of all samples
One-dimensional space Y in the projected, all kinds of mean values
Within-class scatterWith total within-cluster variance
After projection, best projection direction is found, keeps Different categories of samples as separated as possible, it would be desirable that the difference of all kinds of mean values is bigger
Better, within-cluster variance is the smaller the better, defines fisher criterion function:
Denominator is enabled to be equal to non-zero constant, i.e. ωTSwω=C ≠ 0, defining Lagrange function is
L (ω, λ)=ωTSbω-λ(ωTSwω-c)
λ is that Lagrange multiplies word in formula, and Fisher criterion function is sought partial derivative to ω and partial derivative is enabled to be equal to zero, is obtained
It arrives:
Sbω*=λ Swω*
Wherein ω*It is the solution that Fisher criterion function takes maximum, that is, the space higher-dimension X is to the optimal of the one-dimensional space Y
Projection pattern.
The setting supporting vector machine model the following steps are included:
The support vector machines decision function of former feature space is
Wherein αi, i=1 ..., n are the solutions of following double optimization problem；
0≤αi≤ C, i=1 ..., n
Nonlinear transformation is carried out to former feature x, note new feature isThe support then constructed in new feature space to
Amount machine decision function is
Corresponding optimization problem becomes
Two sample inner product (x in former feature spacei.xj) become new feature spaceNew feature is empty
Between inner product be also former feature function, be denoted as kernel function
The support vector machines decision function in new space is write as
Wherein, factor alpha is the optimal solution of optimization problem；
0≤αi≤ C, i=1 ..., n
One suitable kernel function of selection can construct nonlinear support vector machines, select Radial basis kernel function herein
To construct nonlinear support vector machines.
Beneficial effect
Winter wheat fringe head blight recognition methods based on Fei Shi linear discriminant and support vector machines technology of the invention, and it is existing
Having technology to compare realizes the head blight for being directed to single fringe wheat identification.
The present invention is positive from single fringe wheat respectively by non-imaged hyperspectral technique, three angles in side and vertical surface,
The response characteristic of single fringe wheat Fusariumsp disease is analyzed using high-spectral data operation and transform method, and distinct methods are obtained
Spectral signature and the correlativity of severity of head blight evaluated；It is examined and is extracted using correlation analysis and independent sample T
The significant Spectral Properties collection of sensitive to wheat scab and different severity class inheriteds；With SVM and FLDA method and the two
The method combined establishes one respectively being capable of generation severity to winter wheat head blight under field, complicated farm environment
The valid model classified.
Detailed description of the invention
Fig. 1 is method precedence diagram of the invention；
Fig. 2 a is Fei Shi linear discriminant (FLDA), support vector machines (SVM) and Fei Shi linear discriminant and support vector machines phase
The overall accuracy comparison diagram that binding model (FLDA+SVM) is identified；
Fig. 2 b is Fei Shi linear discriminant (FLDA), support vector machines (SVM) and Fei Shi linear discriminant and support vector machines phase
The Kappa index contrast figure that binding model (FLDA+SVM) is identified.
Specific embodiment
The effect of to make to structure feature of the invention and being reached, has a better understanding and awareness, to preferable
Examples and drawings cooperation detailed description, is described as follows:
As shown in Figure 1, the winter wheat fringe of the present invention based on Fei Shi linear discriminant and support vector machines technology is red mould
Sick recognition methods, comprising the following steps:
The first step, data acquisition.Obtain the measured winter wheat list fringe spectroscopic data of spectrometer, wheat single ear spectroscopic data packet
Include the high-spectral data in every plant of fringe front, side and upright three angles.
Vertical measurement is conventional remote sensing technique, but acquisition of information is not complete, and horizontal measurement then can be more anti-than more fully
Disease infection situation is reflected, in order to which preferably lesion caused by head blight is reflected in spectral information, to improve the inspection of head blight
Precision is surveyed, therefore uses the spectra collection method of different angle to assess its practical superiority and inferiority in head blight detection.
Wheat scab spectroscopic data is carried out using AS FieledSpec Pro FR (350~2500nm) type spectrometer
Acquisition.The spectral resolution of spectrometer is 3nm within the scope of 350~1000nm, is 10nm, light within the scope of 1000~2500nm
Spectral sampling interval 1nm.Three angles of every plant of wheatear point carry out spectral measurement, respectively front, side, vertical surface.When observation
Spectrometer probe vertical is downward, highly remain 1.3m from the ground, probe field angle is 25 degree.
Second step, field hyperspectrum data prediction.Winter wheat list fringe spectroscopic data is standardized, and respectively
It carries out first differential calculating, continuum removal calculating and common vegetation index to calculate, obtains corresponding differential characteristics, continuum
Remove feature and vegetation index feature.
Spectrum standardization processing is conducive to improve wheat scab spectral matching factor and distinguishes the reliability of result.The processing
It uses the area A health sample reflectance as standard reflectivity in the process, the reflectivity of the sample in the area B, the area C is adjusted to the area A sample
Reflectivity same level on, this largely eliminates 3 sets of data in time, kind and the difference in terms of breeding time
Influence to spectral discrimination.Due to having used identical ratio curve to carry out each spectroscopic data in the area B and the area C data
Adjustment, therefore the relative spectral difference between healthy sample and different severity extent samples will not be changed.It is subsequent for spectroscopic data
And the analysis of spectral signature is based on the spectroscopic data after standardization and carries out.The specific steps of which are as follows:
(1) high-spectral data standardizes:
The healthy sample spectrum data of a-quadrant are chosen as master sample, and with disease sample spectrum data and health
Sample spectrum data use the mean value of the healthy sample spectrum data of a-quadrant divided by B area and C area health respectively as benchmark
The mean value of sample spectrum data obtains two ratio curves, this two ratio curves are used to reflect the difference of two groups of spectral measurements
Property, the ratio R atio under i wavelengthiCalculation formula are as follows:
Wherein i indicates that wavelength, ∑ refer to reflectivity the sum of of institute's unsoundness sample at wavelength i in region,
Ref indicates reflectivity, and A, B and C respectively represent a-quadrant, B area and the region C；
The original spectral data for each sample that B area and the region C are obtained and the multiplication of corresponding ratio curve to get
Spectroscopic data after to standardization, the standardized calculation formula under i wavelength are as follows:
Wherein,Indicate the reflectivity of B area or C area sample spectrum at wavelength i,Indicate mark
Reflectivity after standardization.
(2) the first derivative spectra data are obtained, difference method approximate calculation is used to the spectrum after standardization, are calculated public
Formula is as follows:
ρ'(λi)=[ρ (λi+1)-ρ(λi-1)]/(2 Δ λ),
Wherein, λiFor each band wavelength, ρ ' (λi) it is wavelength XiFirst differential, Δ λ are wavelength Xi-1To λiInterval, then
Acquire first differential feature；
Obtain continuum and remove feature: the 550nm-750nm wave band of chosen spectrum carries out continuum removal and obtains light relatively
Reflectivity is composed, the wave band position for finding out wave band depth respectively on this basis, absorbing peak area, waveband width and Absorption Characteristics
It sets；
It obtains vegetation index: the vegetation index of several common plant stress researchs is rule of thumb selected, according to each plant
Calculation formula is answered to obtain corresponding index characteristic by exponent pair.
Third step, the selection of Modelling feature.Screening and head blight severity correlation height and extremely significant spectral signature with
And the significant feature of three kinds of severity class inheriteds.
Spectroscopy differential technology is a kind of basic hyperspectral analysis technology, and the difference gauge of different rank is carried out to reflectance spectrum
It calculates, different spectrum parameters can be obtained.First differential conversion reaction amplitude of variation of the vegetation reflectivity in each wave band, can be with
The influence of atmospheric effect, ambient noise is reduced, preferably reflects the substantive characteristics of plant.Spectroscopic data is carried out single order by the present invention
The spectrum obtained after differential is compared with original spectrum, the phase of the curve of spectrum obtained after first differential and head blight severity
Closing property significantly improves, therefore coefficient R is up to 0.8 or more, and the present invention uses transformed by the first derivative spectra
EO-1 hyperion parameter includes position parameter, area parameter, ratio parameter and normalizes index parameter, red to wheatear to attempt to extract
The spectral signature of mildew disease sensitivity.
Continuum removal method is the absorption paddy of the curve of spectrum to be normalized on the continuum line for absorbing paddy, and the purpose is to disappear
Except uncorrelated background information, enhanced spectrum data absorption characteristic information.It is main to absorb paddy characteristic parameter, including absorbing wavelength position
It sets, depth and width, slope, area, symmetry etc..
Vegetation index is basic, common information extraction technology in agricultural remote sensing study on monitoring.This test trial passes through
Some common vegetation indexs extract wheat scab defect information.Based on priori knowledge, with reference to all kinds of vegetation indexs pair
Applicable cases in plant pest study on monitoring have tentatively preselected 21 vegetation indexs based on high-spectral data building, and
Inquire into its applicability in terms of assessing head blight severity.
(1) three kinds of spectrum of feature and vegetation index are removed to first differential feature, the continuum that data prediction obtains
Feature carries out correlation analysis with the severity of winter wheat head blight respectively, obtains related coefficient and P-value value, utilizes threshold
Value method filters out and head blight high correlation R2>0.49 and extremely significant related P-value<0.001 spectral signature；
(2) health, slight, severe three classes sample combination of two are filtered out using the independent sample T method of inspection respectively
The feature of significant difference is presented in three category combinations；
(3) two Spectral Properties collection of two above conditional filtering is taken into intersection, acquisition is sensitive to head blight and in three classes
The optimal Spectral Properties collection that there were significant differences between different diseases severity class；
(4) it repeats the above steps to obtain respectively from front, side and upright three angles and is suitble to the identification of its head blight
Optimal Spectral Properties collection.
4th step, Fei Shi linear discriminant combine the foundation of model with support vector machines: by ground high spectrometry with
And the data building Fei Shi linear discriminant of severity investigation combines the foundation of model with support vector machines.
FLDA is that former high dimensional feature vector is carried out dimension-reduction treatment, has each of high dimensional feature vector by linear combination
Dimension, weeds out and does not have effective factor to classification, the information for being conducive to category classification is fully retained down, so that final is sorted
Journey is easier efficient progress, and at the same time reducing the dimension of feature vector, the calculating effectively reduced in assorting process is multiple
Miscellaneous degree.
The target of SVM algorithm is the optimal solution for seeking to learn precision and learning ability under finite sample information, from theory
On from the point of view of, globally optimal solution can be obtained, local extremum problem can be fallen into avoid what is usually occurred in such as neural network.By
Literature survey discovery, SVM model application value with higher in the identification of small range disease data classification, therefore attempting will
SVM is applied to the identification of the wheat scab of single fringe scale.
In conjunction with the advantages of both FLDA and SVM, first by original high dimensional feature using being obtained in FLDA method reduction process
Low-dimensional input variable of the feature vector as SVM model, attempt the accuracy of identification for improving model.The specific steps of which are as follows:
(1) She Dingfeishi linear discriminant model: the feature of higher-dimension is subjected to dimensionality reduction using Fei Shi linear discriminant model, will be counted
According to the one-dimensional space is projected to, finding one makes within-cluster variance minimum, the maximum best projection direction of inter _ class relationship.
She Dingfeishi linear discriminant model the following steps are included:
A1) the N number of single fringe high-spectral data for obtaining original is defined as set X, includes N number of higher-dimension sample x in X1、x2、...、
xn, all samples are divided into health, slight and 3 class of severe according to head blight coincident with severity degree of condition, wherein N1It is a to belong to W1The sample of class
Originally it is denoted as subset X1, N2It is a to belong to W2The sample of class is denoted as subset X2, N3It is a to belong to W3The sample of class is denoted as subset X3, Different categories of samples
Mean vector mi:
Dispersion matrix between sample classes SiAre as follows:
Total within class scatter matrix SwAre as follows:
Sw=S1+S2+S3；
Matrix between samples SbAre as follows:
Wherein m is the mean vector of all samples
One-dimensional space Y in the projected, all kinds of mean values
Within-class scatterWith total within-cluster variance
A2 after) projecting, best projection direction is found, keeps Different categories of samples as separated as possible, it would be desirable that the difference of all kinds of mean values
It being the bigger the better, within-cluster variance is the smaller the better, define fisher criterion function:
Denominator is enabled to be equal to non-zero constant, i.e. ωTSwω=C ≠ 0, defining Lagrange function is
L (ω, λ)=ωTSbω-λ(ωTSwω-c)
λ is that Lagrange multiplies word in formula, and Fisher criterion function is sought partial derivative to ω and partial derivative is enabled to be equal to zero, is obtained
It arrives:
Sbω*=λ Swω*
Wherein ω*It is the solution that Fisher criterion function takes maximum, that is, the space higher-dimension X is to the optimal of the one-dimensional space Y
Projection pattern.
(2) it sets supporting vector machine model: the input space being transformed to by higher dimensional space by nonlinear transformation, then new
Optimal classification surface is sought in space, so that the linear problem that the nonlinear problem in former space is converted in new space be asked
Solution.
Set supporting vector machine model the following steps are included:
B1) the support vector machines decision function of former feature space is
Wherein αi, i=1 ..., n are the solutions of following double optimization problem；
0≤αi≤ C, i=1 ..., n
Nonlinear transformation is carried out to former feature x, note new feature isThe support then constructed in new feature space to
Amount machine decision function is
Corresponding optimization problem becomes
0≤αi≤ C, i=1 ..., n
B2) two sample inner product (x in former feature spacei.xj) become new feature spaceIt is new special
The inner product in sign space is also the function of former feature, is denoted as kernel function
The support vector machines decision function in new space is write as
Wherein, factor alpha is the optimal solution of optimization problem；
0≤αi≤ C, i=1 ..., n
One suitable kernel function of selection can construct nonlinear support vector machines, select Radial basis kernel function herein
To construct nonlinear support vector machines.
(3) She Dingfeishi linear discriminant combines model with support vector machines: utilizing Fei Shi linear discriminant method by data
Carry out dimensionality reduction, input variable of the obtained one-dimensional characteristic feature vector as supporting vector machine model.
(4) winter wheat fringe head blight sample data is divided into health, slight and severe three classes, wherein randomly selecting sample number
According to as training sample, Fei Shi linear discriminant model and supporting vector machine model (Fei Shi is respectively trained as input in training sample
The model that linear discriminant is combined with support vector machines technology), Fei Shi linear discriminant and supporting vector after respectively obtaining training
The model that machine technology combines.
The support vector machines of above-mentioned foundation is to belong to binary classifier, to realize health according to invention demand, slightly and again
Three classes sample classification is spent, we realize the classification of three classes using three binary classifiers, that is, to every two class in three classes
A classifier is constructed, classification results of each classifier to sample are counted, last sample classification is obtained in some classification
To poll the sample is just at most determined as this kind.
5th step obtains wheatear EO-1 hyperion recognition result: using the optimal characteristics collection of field hyperspectrum data as input
Variable inputs Fei Shi linear discriminant model respectively, inputs Fei Shi linear discriminant respectively with support vector machines and combines model, obtains
The recognition result of winter wheat fringe head blight.
(1) first differential feature that fringe front, side and upright three angles screen respectively, continuum are removed
Feature and vegetation index combine the optimal Spectral Properties collection to form single fringe scale different angle head blight；
(2) input variable of the Spectral Properties collection obtained each angle as model after training, the Fei Shi line after training
Property differentiate combined with support vector machines technology model export respectively it is tight to head blight from positive, side and upright three angles
The result of severe identification.
Wherein, Fei Shi linear discriminant (FLDA) and the confusion matrix of support vector machines (SVM) and recognition result contrast table be such as
Shown in table 1.
1 confusion matrix of table and recognition result contrast table
Note: U=user person's precision, OA=overall accuracy, P=producer's precision
As shown in table 1, a front surface and a side surface angle wheat head, the disease severity recognition effect based on two kinds of different type models
Overall more satisfactory, overall accuracy minimum 77.1%, up to 85.7%；However, FLDA the and SVM model of erect head is sentenced
Other precision is totally slightly lower, and respectively 62.9% and 65.7%.As a whole, the overall accuracy of FLDA model is slightly below SVM mould
Type, support vector machines performance ratio FLDA are more stable.
Since the characteristic variable for participating in model construction is on the high side, excessive variable not only will increase the difficulty of calculating, can also make
The interpretation of model is deteriorated.Primitive character variable is carried out dimensionality reduction using FLDA method by the advantages of therefore combining two kinds of models,
Using the feature vector of obtained low-dimensional as the input variable of SVM model, the precision of model is effectively raised.FLDA, SVM with
And the overall accuracy of tri- kinds of models of FLDA&SVM and the recognition effect of three angles compare such as Fig. 2.As shown in Figure 2 a, FLDA
Model can be effectively improved in the overall recognition accuracy of different angle by combining with SVM, and side precision is better than front, upright angle
It is minimum to spend accuracy of identification, respectively 88.6%, 85.7% and 68.6%, as shown in Figure 2 b, Kappa coefficient trend and totality
Precision growth trend is consistent, respectively 0.808,0.770 and 0.506.
It is found, the research of head blight is concentrated mainly under laboratory condition under wheat single ear scale, and reach by investigation
Preferable recognition effect is arrived.Under the conditions of field, forefathers attempt neural network classification algorithm being applied to high spectrum image
Pixel to distinguish wheat fringe portion health and disease area, and is compared with RBF supporting vector machine model, the results showed that, it is deep
It is ideal to the recognition effect of fringe portion health and disease area to spend convolution recurrent neural networks model, precision reaches 84.6%, RBF branch
Holding vector machine model precision is only 70.6%, but the training time of neural network model is than the time required for support vector machines
It is longer.This research is under the field condition of field, using non-imaged hyperspectral technique, using method of the FLDA in conjunction with SVM to whole
Strain wheat head health slightly with three kinds of classifications of severe carries out classification differentiation, the results showed that, FLDA combines foundation with SVM algorithm
The accuracy of identification of model is higher when model ratio FLDA and SVM algorithm are used alone, and in side, fringe angle full accuracy reaches
88.6%.The forefathers that compare research, this research under the field condition of field, increase the identification scale of wheatear head blight with
Classification difficulty, and achieve ideal recognition effect.
The basic principles, main features and advantages of the present invention have been shown and described above.The technology of the industry
Personnel are it should be appreciated that the present invention is not limited to the above embodiments, and what is described in the above embodiment and the description is only the present invention
Principle, various changes and improvements may be made to the invention without departing from the spirit and scope of the present invention, these variation and
Improvement is both fallen in the range of claimed invention.The present invention claims protection scope by appended claims and its
Equivalent defines.
Claims (7)
1. a kind of winter wheat fringe head blight recognition methods based on Fei Shi linear discriminant and support vector machines technology, feature exist
In, comprising the following steps:
11) data acquisition: the measured winter wheat list fringe spectroscopic data of spectrometer is obtained, winter wheat list fringe spectroscopic data includes every plant
The high-spectral data in fringe front, side and upright three angles；
12) field hyperspectrum data prediction: winter wheat list fringe spectroscopic data is standardized, and carries out single order respectively
Differential calculation, continuum removal calculate and common vegetation index calculates, and obtains corresponding differential characteristics, continuum removal feature
And vegetation index feature；
13) it the selection of Modelling feature: screens tight with head blight severity correlation height and extremely significant spectral signature and three kinds
The significant feature of weight degree class inherited；
14) Fei Shi linear discriminant combines the foundation of model with support vector machines: passing through ground high spectrometry and severity
The data building Fei Shi linear discriminant of investigation combines model with support vector machines；
15) wheatear EO-1 hyperion recognition result is obtained: using the optimal characteristics collection of field hyperspectrum data as input variable, respectively
It inputs Fei Shi linear discriminant and combines model with support vector machines, obtain the recognition result of winter wheat fringe head blight.
2. according to claim 1 identified based on the winter wheat fringe head blight of Fei Shi linear discriminant and support vector machines technology
Method, which is characterized in that the field hyperspectrum data prediction the following steps are included:
21) high-spectral data standardizes:
The healthy sample spectrum data of a-quadrant are chosen as master sample, and with disease sample spectrum data and healthy sample
Spectroscopic data uses the mean value of the healthy sample spectrum data of a-quadrant divided by B area and C area health sample respectively as benchmark
The mean value of spectroscopic data obtains two ratio curves, this two ratio curves are used to reflect the otherness of two groups of spectral measurements, i
Ratio R atio under wavelengthiCalculation formula are as follows:
Wherein i indicates that wavelength, ∑ refer to that reflectivity the sum of of institute's unsoundness sample at wavelength i in region, Ref indicate reflectivity,
A, B and C respectively represents a-quadrant, B area and the region C；
The original spectral data for each sample that B area and the region C obtain is multiplied with corresponding ratio curve to get mark is arrived
Spectroscopic data after standardization, the standardized calculation formula under i wavelength are as follows:
Wherein,Indicate the reflectivity of B area or C area sample spectrum at wavelength i,Indicate standardization
Reflectivity afterwards；
22) the first derivative spectra data are obtained, difference method approximate calculation are used to the spectrum after standardization, calculation formula is such as
Under:
ρ'(λi)=[ρ (λi+1)-ρ(λi-1)]/(2 Δ λ),
Wherein, λiFor each band wavelength, ρ ' (λi) it is wavelength XiFirst differential, Δ λ are wavelength Xi-1To λiInterval, then acquire
First differential feature；
Obtain continuum and remove feature: it is anti-that the 550nm-750nm wave band of chosen spectrum carries out continuum removal acquisition relative spectral
Rate is penetrated, the band po sition for finding out wave band depth respectively on this basis, absorbing peak area, waveband width and Absorption Characteristics；
It obtains vegetation index: rule of thumb selecting the vegetation index of several common plant stress researchs, referred to according to each vegetation
The corresponding calculation formula of number obtains corresponding index characteristic.
3. according to claim 1 identified based on the winter wheat fringe head blight of Fei Shi linear discriminant and support vector machines technology
Method, which is characterized in that the selection of the Modelling feature the following steps are included:
31) three kinds of spectral signatures of feature and vegetation index are removed to first differential feature, the continuum that data prediction obtains
Correlation analysis is carried out with the severity of winter wheat head blight respectively, related coefficient and P-value value is obtained, utilizes threshold method
It filters out and head blight high correlation R2>0.49 and extremely significant related P-value<0.001 spectral signature；
32) health, slight, severe three classes sample combination of two are filtered out using the independent sample T method of inspection at three respectively
The feature of significant difference is presented in category combinations；
33) two Spectral Properties collection of two above conditional filtering is taken into intersection, obtained sensitive to head blight and different in three classes
The optimal Spectral Properties collection that there were significant differences between disease severity class；
34) it repeats the above steps to obtain respectively from front, side and upright three angles and is suitble to the optimal of its head blight identification
Spectral Properties collection.
4. according to claim 1 identified based on the winter wheat fringe head blight of Fei Shi linear discriminant and support vector machines technology
Method, which is characterized in that the Fei Shi linear discriminant foundation of model is combined with support vector machines the following steps are included:
41) She Dingfeishi linear discriminant model: the feature of higher-dimension is subjected to dimensionality reduction using Fei Shi linear discriminant model, data are thrown
For shadow to the one-dimensional space, finding one makes within-cluster variance minimum, the maximum best projection direction of inter _ class relationship；
42) it sets supporting vector machine model: the input space being transformed to by higher dimensional space by nonlinear transformation, then in new space
In seek optimal classification surface, so that the linear problem that the nonlinear problem in former space is converted in new space be solved；
43) She Dingfeishi linear discriminant combines model with support vector machines: being carried out data using Fei Shi linear discriminant method
Dimensionality reduction, input variable of the obtained one-dimensional characteristic feature vector as supporting vector machine model；
44) winter wheat fringe head blight sample data is divided into health, slight and severe three classes, wherein randomly selecting sample data work
For training sample, training sample combines model with support vector machines as input, Xun Lianfeishi linear discriminant.
5. according to claim 1 identified based on the winter wheat fringe head blight of Fei Shi linear discriminant and support vector machines technology
Method, which is characterized in that the acquisition wheatear EO-1 hyperion recognition result the following steps are included:
51) first differential feature, the continuum removal feature screened fringe front, side and upright three angles respectively
And vegetation index combines the optimal Spectral Properties collection to form single fringe scale different angle head blight；
52) input variable of the Spectral Properties collection obtained each angle as model after training, the Fei Shi after training linearly sentence
Supporting vector machine model after other model and training exports serious to head blight from front, side and upright three angles respectively
Spend the result of identification.
6. according to claim 4 identified based on the winter wheat fringe head blight of Fei Shi linear discriminant and support vector machines technology
Method, which is characterized in that the She Dingfeishi linear discriminant model the following steps are included:
61) the N number of single fringe high-spectral data for obtaining original is defined as set X, includes N number of higher-dimension sample x in X1、x2、...、xn,
All samples are divided into health, slight and 3 class of severe according to head blight coincident with severity degree of condition, wherein N1It is a to belong to W1The sample of class
It is denoted as subset X1, N2It is a to belong to W2The sample of class is denoted as subset X2, N3It is a to belong to W3The sample of class is denoted as subset X3, Different categories of samples
Mean vector mi:
Dispersion matrix between sample classes SiAre as follows:
Total within class scatter matrix SwAre as follows:
Sw=S1+S2+S3；
Matrix between samples SbAre as follows:
Wherein m is the mean vector of all samples
One-dimensional space Y in the projected, all kinds of mean values
Within-class scatterWith total within-cluster variance
62) after projecting, best projection direction is found, keeps Different categories of samples as separated as possible, it would be desirable that the difference of all kinds of mean values is bigger
Better, within-cluster variance is the smaller the better, defines fisher criterion function:
Denominator is enabled to be equal to non-zero constant, i.e. ωTSwω=C ≠ 0, defining Lagrange function is
L (ω, λ)=ωTSbω-λ(ωTSwω-c)
λ is that Lagrange multiplies word in formula, and Fisher criterion function is sought partial derivative to ω and partial derivative is enabled to be equal to zero, is obtained:
Sbω*=λ Swω*
Wherein ω*It is the solution that Fisher criterion function takes maximum, that is, the space higher-dimension X to the optimal projection side in the one-dimensional space Y
Formula.
7. according to claim 4 identified based on the winter wheat fringe head blight of Fei Shi linear discriminant and support vector machines technology
Method, which is characterized in that the setting supporting vector machine model the following steps are included:
71) the support vector machines decision function of former feature space is
Wherein αi, i=1 ..., n are the solutions of following double optimization problem；
Nonlinear transformation is carried out to former feature x, note new feature isThe support vector machines then constructed in new feature space
Decision function is
Corresponding optimization problem becomes
72) two sample inner product (x in former feature spacei.xj) become new feature spaceNew feature space
Inner product be also former feature function, be denoted as kernel function
The support vector machines decision function in new space is write as
Wherein, factor alpha is the optimal solution of optimization problem；
One suitable kernel function of selection can construct nonlinear support vector machines, select Radial basis kernel function to carry out structure herein
Build nonlinear support vector machines.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910652829.2A CN110346312B (en) | 2019-07-19 | 2019-07-19 | Winter wheat head gibberellic disease identification method based on Fisher linear discrimination and support vector machine technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910652829.2A CN110346312B (en) | 2019-07-19 | 2019-07-19 | Winter wheat head gibberellic disease identification method based on Fisher linear discrimination and support vector machine technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110346312A true CN110346312A (en) | 2019-10-18 |
| CN110346312B CN110346312B (en) | 2021-08-10 |
Family
ID=68179244
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910652829.2A Active CN110346312B (en) | 2019-07-19 | 2019-07-19 | Winter wheat head gibberellic disease identification method based on Fisher linear discrimination and support vector machine technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110346312B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929944A (en) * | 2019-11-28 | 2020-03-27 | 安徽大学 | Wheat scab disease severity prediction method based on hyperspectral image and spectral feature fusion technology |
| CN111767863A (en) * | 2020-07-01 | 2020-10-13 | 安徽大学 | Winter wheat scab remote sensing identification method based on near-earth hyperspectral technology |
| CN111832507A (en) * | 2020-07-20 | 2020-10-27 | 安徽大学 | Wheat head top spectrum information-based wheat scab remote sensing identification method |
| CN112287886A (en) * | 2020-11-19 | 2021-01-29 | 安徽农业大学 | Wheat plant nitrogen content estimation method based on hyperspectral image fusion map features |
| CN113155747A (en) * | 2021-01-15 | 2021-07-23 | 西安科技大学 | Method for constructing fractional order differential spectral index and monitoring wheat stripe rust thereof |
| CN113176214A (en) * | 2021-04-20 | 2021-07-27 | 安徽大学 | Wheat head gibberellic disease monitoring method by applying continuous wavelet analysis through non-imaging spectrum |
| CN114199800A (en) * | 2021-12-08 | 2022-03-18 | 华智生物技术有限公司 | Method, system, equipment and medium for identifying rice sheath blight |
| CN114332589A (en) * | 2021-12-31 | 2022-04-12 | 中国科学院紫金山天文台 | Method for accurately detecting surface water or hydroxyl of atmospheric celestial body |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103218625A (en) * | 2013-05-10 | 2013-07-24 | 陆嘉恒 | Automatic remote sensing image interpretation method based on cost-sensitive support vector machine |
| CN105424645A (en) * | 2015-10-29 | 2016-03-23 | 福州大学 | Method for fast identifying clinical pathogens based on principal component analysis and Fisher discriminance |
| CN107527326A (en) * | 2017-08-17 | 2017-12-29 | 安徽农业大学 | A kind of wheat scab diagnostic method based on high light spectrum image-forming |
| CN108985237A (en) * | 2018-07-20 | 2018-12-11 | 安徽农业大学 | A kind of detection method and its system of the wheat scab based on depth mixing |
| CN109117743A (en) * | 2018-07-20 | 2019-01-01 | 安徽农业大学 | A kind of detection method and its system of the wheat scab based on depth convolution |
| CN109187552A (en) * | 2018-08-30 | 2019-01-11 | 安徽农业大学 | A kind of gibberella saubinetii disease grade determination method based on cloud model |
| CN109308697A (en) * | 2018-09-18 | 2019-02-05 | 安徽工业大学 | A kind of leaf disease recognition method based on machine learning algorithm |
| CN109657653A (en) * | 2019-01-21 | 2019-04-19 | 安徽大学 | A kind of wheat seed head blight recognition methods based on Imaging Hyperspectral Data |
| CN110008905A (en) * | 2019-04-09 | 2019-07-12 | 安徽大学 | Regional scale stripe rust of wheat monitoring method based on the red side wave section of remote sensing image |
-
2019
- 2019-07-19 CN CN201910652829.2A patent/CN110346312B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103218625A (en) * | 2013-05-10 | 2013-07-24 | 陆嘉恒 | Automatic remote sensing image interpretation method based on cost-sensitive support vector machine |
| CN105424645A (en) * | 2015-10-29 | 2016-03-23 | 福州大学 | Method for fast identifying clinical pathogens based on principal component analysis and Fisher discriminance |
| CN107527326A (en) * | 2017-08-17 | 2017-12-29 | 安徽农业大学 | A kind of wheat scab diagnostic method based on high light spectrum image-forming |
| CN108985237A (en) * | 2018-07-20 | 2018-12-11 | 安徽农业大学 | A kind of detection method and its system of the wheat scab based on depth mixing |
| CN109117743A (en) * | 2018-07-20 | 2019-01-01 | 安徽农业大学 | A kind of detection method and its system of the wheat scab based on depth convolution |
| CN109187552A (en) * | 2018-08-30 | 2019-01-11 | 安徽农业大学 | A kind of gibberella saubinetii disease grade determination method based on cloud model |
| CN109308697A (en) * | 2018-09-18 | 2019-02-05 | 安徽工业大学 | A kind of leaf disease recognition method based on machine learning algorithm |
| CN109657653A (en) * | 2019-01-21 | 2019-04-19 | 安徽大学 | A kind of wheat seed head blight recognition methods based on Imaging Hyperspectral Data |
| CN110008905A (en) * | 2019-04-09 | 2019-07-12 | 安徽大学 | Regional scale stripe rust of wheat monitoring method based on the red side wave section of remote sensing image |
Non-Patent Citations (1)
| Title |
|---|
| 梁琨: "基于高光谱成像技术的小麦籽粒赤霉病识别", 《农业机械学报》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929944A (en) * | 2019-11-28 | 2020-03-27 | 安徽大学 | Wheat scab disease severity prediction method based on hyperspectral image and spectral feature fusion technology |
| CN111767863A (en) * | 2020-07-01 | 2020-10-13 | 安徽大学 | Winter wheat scab remote sensing identification method based on near-earth hyperspectral technology |
| CN111767863B (en) * | 2020-07-01 | 2023-10-31 | 安徽大学 | Remote sensing identification method for winter wheat scab based on near-earth hyperspectral technology |
| CN111832507A (en) * | 2020-07-20 | 2020-10-27 | 安徽大学 | Wheat head top spectrum information-based wheat scab remote sensing identification method |
| CN111832507B (en) * | 2020-07-20 | 2024-04-09 | 安徽大学 | Wheat scab remote sensing identification method based on wheat head spectral information |
| CN112287886A (en) * | 2020-11-19 | 2021-01-29 | 安徽农业大学 | Wheat plant nitrogen content estimation method based on hyperspectral image fusion map features |
| CN112287886B (en) * | 2020-11-19 | 2023-09-22 | 安徽农业大学 | Wheat plant nitrogen content estimation method based on hyperspectral image fusion map features |
| CN113155747A (en) * | 2021-01-15 | 2021-07-23 | 西安科技大学 | Method for constructing fractional order differential spectral index and monitoring wheat stripe rust thereof |
| CN113176214A (en) * | 2021-04-20 | 2021-07-27 | 安徽大学 | Wheat head gibberellic disease monitoring method by applying continuous wavelet analysis through non-imaging spectrum |
| CN114199800A (en) * | 2021-12-08 | 2022-03-18 | 华智生物技术有限公司 | Method, system, equipment and medium for identifying rice sheath blight |
| CN114199800B (en) * | 2021-12-08 | 2023-11-21 | 华智生物技术有限公司 | Rice sheath blight disease identification method, system, equipment and medium |
| CN114332589A (en) * | 2021-12-31 | 2022-04-12 | 中国科学院紫金山天文台 | Method for accurately detecting surface water or hydroxyl of atmospheric celestial body |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110346312B (en) | 2021-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110346312A (en) | Winter wheat fringe head blight recognition methods based on Fei Shi linear discriminant and support vector machines technology | |
| Pilling et al. | High-throughput quantum cascade laser (QCL) spectral histopathology: a practical approach towards clinical translation | |
| Lu et al. | Classification of Camellia (Theaceae) species using leaf architecture variations and pattern recognition techniques | |
| CN110717368A (en) | Qualitative classification method for textiles | |
| Huang et al. | Hyperspectral image analysis based on BoSW model for rice panicle blast grading | |
| CN109858477A (en) | The Raman spectrum analysis method of object is identified in complex environment with depth forest | |
| CN103235095A (en) | Water-injected meat detection method and device | |
| Feng et al. | A deep convolutional neural network-based wavelength selection method for spectral characteristics of rice blast disease | |
| CN111767863B (en) | Remote sensing identification method for winter wheat scab based on near-earth hyperspectral technology | |
| CN108108719A (en) | A kind of Weighted Kernel is sparse and cooperates with the Hyperspectral Image Classification method for representing coefficient | |
| CN112528559B (en) | Chlorophyll a concentration inversion method combining pre-classification and machine learning | |
| CN104616022B (en) | A kind of sorting technique of near infrared spectrum | |
| Cui et al. | Identification of maize seed varieties based on near infrared reflectance spectroscopy and chemometrics | |
| CN111523587A (en) | Woody plant species spectrum identification method based on machine learning | |
| CN110779875B (en) | Method for detecting moisture content of winter wheat ear based on hyperspectral technology | |
| Li et al. | A novel hyperspectral imaging and modeling method for the component identification of woven fabrics | |
| CN105869161B (en) | Hyperspectral image band selection method based on image quality evaluation | |
| Wang et al. | Prediction of soil organic matter using VNIR spectral parameters extracted from shape characteristics | |
| Agjee et al. | Remote sensing bio-control damage on aquatic invasive alien plant species | |
| CN109146003B (en) | Hyperspectral imaging-based classification identification method, device and system | |
| US20030087456A1 (en) | Within-sample variance classification of samples | |
| Michalopoulou et al. | RDX detection with THz spectroscopy | |
| Long et al. | Identification of spectral features in the longwave infrared (LWIR) spectra of leaves for the discrimination of tropical dry forest tree species | |
| Jinzhu et al. | Discrimination of tomato yellow leaf curl disease using hyperspectral imaging | |
| Lin et al. | Hyperspectral estimation of soil composition contents based on kernel principal component analysis and machine learning model |
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 |