Using Symmetry in Robust Model Fitting
Using symmetry in robust model ?tting
Hanzi Wang *,David Suter
Department of Electrical and Computer Systems Engineering,Monash University,Wellington Road,Clayton,Vic.3800,Australia
Received 27August 2002;received in revised form 15April 2003
Abstract
The pattern recognition and computer vision communities often employ robust methods for model ?tting.In particular,high breakdown-point methods such as least median of squares (LMedS)and least trimmed squares (LTS)have often been used in situations where the data are contaminated with outliers.However,though the breakdown point of these methods can be as high as 50%(they can be robust to up to 50%contamination),they can break down at unexpectedly lower percentages when the outliers are clustered.In this paper,we demonstrate the fragility of LMedS and LTS and analyze the reasons that cause the fragility of these methods in the situation when a large percentage of clustered outliers exist in the data.We adapt the concept of ‘‘symmetry distance’’to formulate an improved regression method,called the least trimmed symmetry distance (LTSD).Experimental results are presented to show that the LTSD performs better than LMedS and LTS under a large percentage of clustered outliers and large standard variance of inliers.
ó2003Elsevier B.V.All rights reserved.
Keywords:Robust regression;Symmetry distance;Clustered outliers;Breakdown point
1.Introduction
One major task of computer vision,pattern recognition,machine learning,and related areas:is to ?t a model to noisy data (with outliers)(Fischler and Bolles,1981;Rousseeuw,1984;Rousseeuw and Leroy,1987;Zhang,1997;Danuser and Stricker,1998;Stewart,1999).It is common to employ ‘‘regression analysis’’to undertake such
tasks (Rousseeuw and Leroy,1987).The most common form of regression analysis is the least squares (LS)method,which can achieve optimum results under Gaussian distributed noise.But this method is extremely sensitive to outliers (gross errors or samples belonging to another structure and distribution).The breakdown point of an es-timator may be roughly de?ned as the smallest percentage of outlier contamination that can cause the estimator to produce arbitrarily large values.Mathematically,let Z be any sample of n data points ex 1;y 1T;...;ex n ;y n T,Z ?f z 1;...;z n g and z i ?f x i ;y i g .For m 6n ,the ?nite-sample break-down point of a regression estimator T can be written as (Rousseeuw and Leroy,1987,pp.10):
*
Corresponding author.Tel.:+61-395-159471;fax:+61-399-053454.
E-mail addresses:hanzi.wang@https://www.wendangku.net/doc/ba7196474.html,.au (H.Wang),d.suter@https://www.wendangku.net/doc/ba7196474.html,.au (D.Suter).
0167-8655/$-see front matter ó2003Elsevier B.V.All rights reserved.
doi:10.1016/S0167-8655(03)00156-9
Pattern Recognition Letters 24(2003)
2953–2966
e?n eT;ZT?min
m
;sup
Z?2Z m
k T neZ?Tk
?1
e1:1T
Because one single outlier is su?cient to force the LS estimator to produce arbitrarily large value,the LS estimator has a breakdown point of0%.
Since data contamination is usually unavoid-able(due to faulty feature extraction,sensor noise and failure,segmentation errors,etc.),there has recently been a general recognition that algorithms should be robust(Haralick,1986).Robust regres-sion methods are a class of techniques that can tolerate gross errors(outliers).Some robust methods also have a high breakdown point.
Several robust estimators with apparently high breakdown points have been developed during the past three decades.The least median of squares (LMedS)estimator and the least trimmed squares (LTS)estimator(Rousseeuw,1984;Rousseeuw and Leroy,1987)are the two most popular methods with claimed high breakdown points,and are currently used most widely in computer vision ?eld(Roth and Levine,1990;Bab-Hadiashar and Suter,1998;Ming and Haralick,2000,etc.).They are based on the idea that the correct?t will cor-respond to the one with the least median of re-siduals(for LMedS),or the least sum of trimmed squared residuals(for LTS).The essence of the argument claiming a high breakdown point for the LMedS is that if the uncontaminated data are in the majority,then the median of the residuals should be una?ected by the outliers,and thus the median residual should be a reliable measure of the quality of the?t.Likewise,since the LTS method relies only on(the sum of squares of)the h smallest residuals,for some choice of the para-meter h,it is thought that this should be robust to contamination so long as h data points,at least, belong to the true?t.Though the robustness of these methods,compared with patently non-robust methods such as Least Squares,is now widely demonstrated and acknowledged,it is per-haps less widely recognized that LMedS and LTS can breakdown at surprisingly low contamination if those outliers are clustered.Due to the a?ects of clustered outliers,the correct?t may not corre-spond to the?t with the least median of squared residual(for LMedS)or the least trimmed squared residuals(for LTS).It is worth mentioning that this phenomenon is not limited to the LMedS,and LTS.It also happens to most other robust esti-mators such as random sample consensus––RANSAC(Fischler and Bolles,1981),residual consensus estimator––RESC(Yu et al.,1994), adaptive least k th estimator––ALKS(Lee et al., 1998),etc.The mechanism of the breakdown in these robust estimators is similar to that of the LMedS and LTS(see Wang and Suter,2003).In this paper,we restrict ourselves to the better known and more widely applied LMedS and LTS. Making clear the reasons that cause the break-down of the LMedS and LTS will help understand the fragility of other robust estimators to clustered outliers.
The key to salvaging the robustness of LMedS and LTS,even in the presence of clustered outliers, can be recognize that LMedS and LTS(and some other robust estimators such as M-estimators, ALKS,etc.)only depend upon a single statistical property of the data(the median of the squared residuals,or the sum of the smallest h residuals, respectively).If one expects some other property of the data,associated with the true?t,to hold, then one may seek to incorporate a measure of that property into the problem formulation.In this paper,we restrict ourselves to one such property––symmetry.
Symmetry is very common and important in our world.When we?t circles,ellipses,or any symmetric object,one of the most basic features in the model is symmetry.In our method,we intro-duce symmetry into the model?tting and thereby propose an improved method––the least trimmed symmetry distance(LTSD).The LTSD is in?u-enced not only by the sizes of the residuals of data points,but also by the symmetry of the data points and has applications where one is trying to?t a symmetric model(e.g.circle and ellipses).Experi-ments show that the LTSD method works better than LMedS and LTS under a large number of clustered outliers.
The main contributions of this paper are as follows:
1.We illustrate situations where LMedS and LTS
fail to correctly?t the data in the presence of
2954H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–2966
clustered outliers,and analyze the reasons that cause the breakdown of these two methods.
This provides an important cautionary note when employing these two robust estimators in situations where the outliers are clustered.
2.We introduce the concept of symmetry distance
(SD)into model?tting.The concept of SD in computer vision is not novel.However it is a novel concept in the?eld of model?tting.Based on Su et al.?s point SD(2001),we propose a novel symmetry distance and apply it to model ?tting.
3.We experimentally show that the proposed
method work better than LMedS and LTS under a large percentage of clustered outliers for both simulated and real data.
This paper is organized as follows:in Section2, several robust estimators are reviewed.The rea-sons why both LMedS and LTS fail to?t a model under a large percentage of clustered outliers are explored.In Section3a novel SD measure is given and our proposed method is developed in Section 4.Experiments demonstrating the utility of the approach(for circle?tting and ellipses?tting)are given in Section5.Finally,the contributions of this paper and future work are summarized in Section6.
2.Robust estimators
The classical linear model can be written in the form:
y i?x i1h1tááátx ip h pte iei?1;...;nTe2:1Twhere the error term,e i,is usually assumed to be normally distributed with mean zero and standard deviation.The aim of regression analysis is to es-timate h?eh1;...;h pTt from the dataex i1;...; x ip;y iT.
The ordinary LS method estimates^h by
min
^h X n
i?1
r2
i
e2:2T
where the residual of the i th datum:r i?y iàx i1^h1àáááàx ip^h p.
Although the LS method has low computational cost and high e?ciency when the data are Gaussian distributional,it is very sensitive to outliers.In order to reduce the in?uence of outliers,a number of robust estimators have been developed.Among these estimators,the maximum-likelihood-type estimators,i.e.the M-estimators,are best known (Huber,1973,1981).
2.1.M-estimators
The essence of the M-estimator is to replace the squared residuals r2
i
in(2.2)by another function of the residuals:
min
^h
X n
i?1
qer iTe2:3T
where qeáTis a symmetric,positive-de?nite func-tion with a unique minimum at zero.Di?erent choices of qer iTwill yield di?erent M estimators. Unfortunately,it has been proved that M-estima-tors have a breakdown point of at most1=ept1T, where p represents the dimension of parameter vector(Li,1985).This means that the breakdown point will diminish when the dimension of the parameter vector increases.
2.2.The repeated median method
Before the repeated median(RM)estimator,it was controversial whether it was possible to?nd a robust estimator with a high breakdown point.In 1982,Siegel proposed the RM estimator with a high breakdown point of50%(Siegel,1982).
The RM method can be summarized as fol-lowed:take any p observations(p is the dimension of the model),ex i1;y i1T;...;ex ip;y ipT.The parameter vector^h?e^h1;...;^h pTt can be calculated from this set of data points.The j th coordinate of this vector is denoted by h jei1;...;i pT.Then the RM estimator is de?ned as:
^h?med
i1
e...emed
i pà1
emed
i p
h jei1;...;i pTTT...Te2:4T
The RM estimator has succeeded in solving the problems with small p.But the time complexity of the RM estimator is Oen p log p nT,which is very high for multidimensional models and so the
H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–29662955
utility of the method is restricted to low dimen-sional models.Since this method is not widely used in the computer vision community,we do not consider it further in this paper.
2.3.The least median of squares method
Rousseeuw proposed the LMedS in1984.The LMedS method has excellent global robustness and claimed high breakdown point(also50%). Over the last two decades,LMedS has been growing in popularity.For example,Kumar and Hanson(1989)used the LMedS to solve the pose estimation problem;Roth and Levine(1990)em-ployed it for range image segmentation;Meer et al. (1990)applied it for image structure analysis in the piecewise polynomial?eld;Zhang(1997)used the least median squares in conic?tting;and Bab-Hadiashar and Suter(1998)employed it for optic ?ow calculation.
The LMedS method is based on the simple idea of replacing the sum in the least sum of squares formulation by a median.That is,the LMedS es-timate is given by
min
^h med
i
r2
i
e2:5T
A drawback of the LMedS method is that it esti-mates the parameters by solving this non-linear minimization problem.No explicit formula exists for the solution of this problem––the exact solu-tion can only be determined by a search in the space of all possible estimates.There are Oen pTp-tuples of data,and it takes Oen log nTtime to?nd the median of the residuals of the whole data for each p-tuple.Thus it costs Oen pt1log nTfor the LMedS method.In order to reduce the cost to a feasible value,a Monte Carlo type technique,as described next,is usually employed.
A p-tuple is‘‘clean’’if it consists of p good observations without contamination by outliers. One performs,m times,random selections of p-tuples,where one chooses m so that the probability (P)that at least one of the m p-tuples is‘‘clean’’is almost1.Let e be the fraction of outliers contained in the whole set of points.The probability P can be expressed as follows:
P?1àe1àe1àeTpTme2:6TThus one can determine m for given values of e,p and P by:
m?
loge1àPT
p
e2:7T
The relative e?ciency of the LMedS method is poor when Gaussian noise is present in the data. The LMedS has a low convergence rate of order nà1=3.Rousseeuw improved the LMedS method by further carrying out a weighted LS procedure after the initial LMedS.In this adaptation,a prelimi-nary scale estimate is found by:
S?1:4826e1t5=enàpTTM ie2:8Twhere M i is the median of the residuals returned by the LMedS procedure.
The weight function W i which will be assigned to the i th data point is usually given by:
W i?
1r2
i
6e2:5ST2
0r2
i
>e2:5ST2
e2:9T
The data points corresponding to W i?0are likely to be outliers.The data points having W i?1are inliers and will in?uence the weighted LS estimate:
min
^h
X n
i?1
W i r2
i
e2:10T
The LMedS method may be locally unstable when ?tting models to data.This means that an in?ni-tesimal change in the data can greatly alter the output(Thomas and Simon,1992).
2.4.The least trimmed squares method
The LTS method was introduced by Rousseeuw (1984),Rousseeuw and Leroy(1987)to improve the low e?ciency of LMedS.The LTS estimator can be expressed as
min
^h
X h
i?1
er2T
i:n
e2:11T
whereer2T
1:n
6ááá6er2T
n:n
are the ordered squared residuals,h is the trimming constant.
The method uses h data points(out of n)to estimate the parameters.The coverage value,h, may be set from n=2to n.The aim of LTS esti-mator is to?nd the h-subset with smallest LS
2956H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–2966
residuals and use the h-subset to estimate para-meters of models.The claimed LTS breakdown point isenàhT=n.When h is set to n=2,the LTS esti-mator has a claimed high breakdown value of50%.
The advantages of LTS over LMedS are:
?It is less sensitive to local e?ects than LMedS,
i.e.it has more local stability.
?LTS has better statistical e?ciency than LMedS (H€o ssjer,1994).It converges like nà1=2.
Due to these merits,LTS is preferred to LMedS (Rousseeuw and Aelst,1999;Ming and Haralick, 2000).
LTS is also implemented by using random sampling,because the number of all possible h-
subsetseC h
n Tgrows fast with n.There are two
commonly employed ways to generate a h-subset (Rousseeuw and Driessen,1999):
1.Directly generate a random h-subset from the
data points n.
2.Firstly generate a random p-subset.If the rank
of this p-subset is less than p,randomly add data points until the rank is equal to p.Next, use this subset to compute parameters^h jej?1;...;pTand residuals r iei?1;...;nT.Sort the residuals into j repe1Tj6ááá6j repehTj6ááá6 j repenTj,and h-subset is set to:H:?f pe1T;...;
pehTg.
Although the?rst way is easier than the second, the h-subset yielded by the?rst method may con-tain a lot of outliers.Indeed,the chance of gen-erating a‘‘clean’’h-subset by method(1)tends to zero with increasing n.In contrast,it is easier to ?nd a‘‘clean’’p-subset without outliers by method (2).Therefore,method(2)can generate more (good)initial subsets with size h than method(1).
Like LMedS,the e?ciency of LTS can be im-proved by adopting a weighted LS re?nement as the last stage.
2.5.Factors a?ecting the achievements of LMedS and LTS
The LMedS method and the LTS method are based on the idea that the correct?t is determined by a simple situation:the least median of the squared residuals(for LMedS),or by the least sum of trimmed squared residuals(for LTS);and that such a statistic is not in?uenced by the outliers.
Consider the contaminated distributions as follows(Hampel et al.,1986;Haralick,1986):
F?e1àeTF0te He2:12Twhere F0is an inlier distribution,and H is an out-lier distribution.
The equation above is also called the gross error model.When the standard variance of F0is small ((1)and that of H is large or H is uniform dis-tributed,the assumptions leading to the robustness of LMedS or LTS,are true.However,when F0is ‘‘scattered’’,i.e.the standard variance of F0is big, and H is clustered distributed with high density, the assumption is not always true.
Let us investigate an example.In Fig.1,100 good data(inliers)with noise distribution F0(bi-variate normal with unit variance)were generated by adding the noise to samples of a circle with radius10.0and center ate0:0;0:0T.Then80clus-tered outliers were added,possessing a spherical bivariate normal distribution with one unit stan-dard variance and meane20:0;6:0T.As Fig.1 shows,both LMedS and LTS failed to?t the circle:LMS returned the result with a radius equal to7.4239and the center was located ate13:1294;
H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–29662957
8:6289T.The results obtained by LTS were:the radius equalled19.3069and the center was at e1:1445;10:1470T.
It is important to point out that the failure is inherent,and not simply an artefact of our im-plementation.Let us check the median of the re-siduals(for LMedS)and the sum of trimmed squared residuals(for LTS)and we will under-stand why LMedS and LTS failed to?t to the circle.The median of residuals of the perfect?t is 5.7928.However the median of residuals of?nal result by the LMedS method is5.1479.
In fact,during the searching procedure,the LMedS estimator consistently minimises the me-dian of the residuals,starting with initial?ts that have a larger median residual than the true?t, but successively?nding?ts with lower median resi-duals––proceeding to even lower median residuals than that possessed by the true?t(shown in Fig.
2).
The reason that LTS failed is similar.LTS?nds the?t with smallest trimmed squared residuals. The value of the least sum of trimmed squared residuals obtained is32.6378.However,the same statistic for the‘‘true?t’’is58.6688.Clearly,LTS has‘‘correctly’’,by its criterion,obtained a‘‘bet-ter’’?t(but in fact,the wrong one).The problem is not with the implementation but with the criterion.
Now,let us consider another example showing that the results of the LMedS and the LTS are a?ected by the standard variance of the inliers.We generated a circle with radius10.0and center at e0:0;0:0T.In addition,clustered outliers were added to the circle with meane20:0;6:0Tand unit standard variance.In total,100data points were generated.At?rst,we assigned100data to the circle without any outliers.Then we repeatedly moved two points from the circle to the clustered outliers until50data were left in the circle.Thus, the percentage of outliers changed from0%to 50%.In addition,for each percentage of clustered outliers,we varied the standard variance of the inliers from0.4to1.6with a step size of0.3.Fig. 3(a)illustrates one example of the distribution of the data,with38%clustered outliers and the standard variance of inliers1.3.
From Fig.3(b)and(c),we can see that when the standard variance of inliers is no more than 1.0,LMedS can give the right results under high percentage of outliers(more than44%).However, when the standard variance of inliers is more than 1.0,LMedS does not give the right result even when the percentage of outliers is less then40%. From Fig.3(b),we can see when the standard variance of inliers is0.4,the LTS estimator can correctly give the results even under50%clustered outliers;while when the standard variance of in-liers is1.6,LTS does not give the right results even when only30%of the data are outliers.
From the discussion above,we now see several conditions under which LMedS and LTS failed to be robust.A crucial point is:these methods mea-sure only one single statistic:the least median of residuals or the least sum of trimmed squared of residuals,omitting other characteristics of the data.If we look at the failures,we can see that the results lost the most basic and common feature of the inliers with respect to the?tted circle––sym-metry.
Symmetry is considered a pre-attentive feature that enhances recognition and reconstruction of
2958H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–2966
shapes and objects(Attneave,1995).Symmetry exists in many man-made and natural objects.In the next section,we will introduce the concept of SD into robust regression methods and propose an improved method,called the LTSD,by which the better performance is acquired even when data include clustered outliers.
3.The symmetry distance
Symmetry exists almost at all places of the world.A square,a cube,a sphere,and a lot of geometric patterns show symmetry.Architecture usually displays symmetry.Symmetry is also an important parameter in physical and chemical processes and is an important criterion in medical diagnosis.Even we human beings show symmetry, (for instance,our faces and bodies are roughly symmetrical between right and left).One of the most basic features in the shapes of models we often?t/impose on our data, e.g.circles and ellipsis,is the symmetry of the model.Symmetric data should suggest symmetric models and data that is symmetrically distributed should be pre-ferred as the inlier data(as opposed to the out-liers).For decades,symmetry has widely been studied in computer vision community(Bigun, 1988;Marola,1989;Nalwa,1989;Kirby and Sirovich,1990;Reisfeld et al.,1992;Zabrodsky
H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–29662959
et al.,1995;Su and Chou,2001).More detailed de?nitions of symmetry can be found in(Zab-rodsky,1993).We demonstrate here that symme-try can also be used as a feature to enhance the performance of robust estimators when?tting models with symmetric structure.
3.1.The symmetry distance
The exact mathematical de?nition of symmetry (Miller,1972;Weyl,1952)is insu?cient to describe and quantify symmetry found both in the natural world and in the visual world.
Su and Chou(2001)proposed a SD measure based on the concept of‘‘point symmetry’’.Given n points x i;i?1;...;N and a reference vector C (e.g.the centroid of the data),the point SD be-tween a point x j and C is de?ned as follows:
d sex j;CT?min
i?1;...;N
and i?j kex jàCTtex iàCTk
j i
e3:1T
From(3.1)we can see that the point SD is non-negative de?nition.In essence,the measure tries to‘‘balance’’data points with others symmetric about the centroid––for example,x i?e2Càx jTexists in the data,d sex j;CT?0.
However,according to(3.1),one point could be used repeatedly as the‘‘balancing point’’with re-spect to the center.This does not seem to properly capture the notion of symmetry.
In order to avoid one point being used as a ‘‘symmetric point’’more than one time by other points,we re?ne the point SD between a point x j and C as follows:
D sex j;CT?min
i?1;...;N
and i?j
and i2R kex jàCTtex iàCTk
k x jàC ktk x iàC k
e3:2T
where R is a set of points that have been used as ‘‘symmetric point’’.
Based on the concept of‘‘point SD’’,we pro-pose a non-metric SD.Given a pattern x consisted of n points x1;...;x n and a reference vector C,the symmetry distance of the pattern x with respect to the reference vector C is:SD nex;CT?
1
n
X n
i?1
D sex i;CTe3:3T
When the SD of a pattern is equal to0.0,the pattern is perfect symmetric;when the SD of a pattern is very big,the pattern has little sym-metry.
4.The proposed method
We proposed a new method,which couples the LTS method with the symmetry distance measure. Besides residuals,we also choose SD as a criterion in the model?tting.For simplicity,we call the proposed method the LTSD(least trimmed SD). Mathematically,the LTSD estimate can be written as:
^h?arg min
h;C
SD hex;CTe4:1T
Only h data points with the smallest sorted resi-duals are used to calculate the SD.The estimated parameters correspond to the least SD.
The speci?c procedures of the proposed method is given as follows:
Step1.Set repeat times(RT)according to equa-tion(2.7).Initialise h with?entpt1T=
2 6h6n.If we want LTSD to have a
high breakdown point,say50%,we can
set h?entpt1T=2.
Step2.Randomly choose p-subsets,and extend to
a h-subset H1by the method(2)in Section
2.4.
https://www.wendangku.net/doc/ba7196474.html,pute^h1by LS method based on H1.
Compute symmetry distance SD1based
on^h1and H1using(3.3)in Section3and
using the centre of the?t(circle or ellipse)
as the reference vector C.Decrement RT
and if RT is smaller than0,go to step4,
otherwise,go to step2.We calculate the
parameters^h based on h-subset instead
of p-subset in order to improve the statis-
tical e?ciency.
Step4.Finally,output^h with the lowest SD.
2960H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–2966
5.Experimental results
In this section,we will show several examples using the proposed method to?t a model with symmetrical structures.Circle?tting and ellipse ?tting have been very popular topics in the com-puter vision?eld.One of the obvious characteris-tics of circles and ellipses is that they are symmetric. We?rst present an example of circle?tting,to provide insights into the proposed method.We then present a relatively more complicated exam-ple of ellipse?tting.The results are compared with those of the LMedS method and the LTS method.
5.1.Circle?tting
In Fig.4,about45%clustered outliers were
added to the original circle data.Since LMedS and LTS only rely on the residuals of the data points, their results were a?ected by the standard variance of the inliers and percentages of the clustered out-liers.Therefore,they failed to?t the circle under high percentage of clustered outliers(see Fig.1). However,because the LTSD method considers the symmetry of the object,this enables LTSD?nd the right model(see Fig.4):the true centre and radius of the circle are respectivelye0:0;0:0Tand 10.0;by the LTSD method,we obtained centre eà0:23;0:01Tand radius10.06.
Another example showing the advantages of the proposed method is given in Fig.5(a)(corre-sponding to Fig.3(a)).From Fig.5(a),we can see that when the outliers are clustered,the LMedS and LTS broke down under very low percentages of outliers,in this case,they both broke down under38%outliers!In comparison to the LMedS and LTS methods,the proposed method gives the
H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–29662961
most accurate results.The proposed method is a?ected less by the standard variance of the inliers and the percentages of the clustered outliers.Fig. 5(b)shows that the radius found by the LTSD method in circle?tting(true radius is10.0)chan-ged less under di?erent standard variance of the inliers and percentages of clustered outliers.In comparison to Fig.3(b)and(c),the?uctuation of the radius found by the LTSD method is smaller. Even when50%clustered outliers exist in the data and the standard variance of inliers is1.6,the re-sults did not(yet)break down.However,both the LMedS and the LTS broke down.
5.2.Ellipse?tting
Ellipses are one of most common and important primitive models in computer vision and pattern
recognition,and often occur in geometric shapes, man-made and natural scenes.Ellipse?tting is a very important task for many industrial applica-tions because it can reduce the data and bene?t the higher level processing(Fitzgibbon et al.,1999). Circles may be projected into ellipses under per-spective projection.Thus ellipses are frequently used in computer vision for model matching (Sampson,1982;Fitzgibbon et al.,1999;Robin, 1999,etc.).In this subsection,we apply the pro-posed robust method to ellipse?tting.
A general conic equation can be written as follows:
ax2tbxytcy2tdxteytf?0e5:1Tea;b;c;d;e;fTare the parameters needed to?nd from the given data.When b2<4ac,the equation above corresponds to ellipses.
The ellipse can also be represented by its more intuitive geometric parameters:ex cos hty sin hàx c cos hày c sin hT2
A2
t
eàx sin hty cos htx c sin hày c cos hT2
B2
?1
e5:2Twhereex c;y cTis the center of the ellipse,f A;B g are the major and minor axes,and h is the orientation of the ellipse.
The relation betweenea;b;c;d;e;fTandex c;y c; A;B;hTis(Robin,1999):
x c?
beà2cd
4acàb2
y c?
bdà2ae
4acàb2
f A;B g?2
???????????????????????????????????????????????????????
à2f
atc?f
???????????????????????????????
b2t
aàc
f
2
s
v u
u u
u t
h?
1
2
tanà1
b
aàc
8
>>>
>>>
>>>
>>>
><
>>>
>>>
>>>
>>>
>:
e5:3T
Table1
Comparison of the estimated parameters by the LTSD,LTS,and LMedS methods in ellipse?tting under40%clustered outliers
x c y c Major axis Minor axis h(°)
True value0.00.010.08.00.0
The LTSD method)0.125)0.1459.7607.810 6.355
The LTS method19.786 5.162 3.328 3.03534.129
The LMedS method9.560 5.20811.757 3.679)3.307
2962H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–2966
It is convenient to ?nd ea ;b ;c ;d ;e ;f T?rst by the given data and then convert to ex c ;y c ;A ;B ;h T.As illustrated in Fig.6and Table 1,200data were generated with 40%clustered outliers.The outliers were compacted within a region of radius 5and center at e20:0;5:0T.The ellipse had a stan-dard variance 0.8,major axis 10.0,minor axis
8.0,
Fig.7.Fitting a mouse pad:(a)a mouse pad with some ?owers;(b)the edge image by using Canny operator and (c)the results obtained by the LTSD,LTS and LMedS
methods.
Fig.8.Fitting the ellipse in a cup:(a)a real cup image;(b)the edge of the cup by applying Prewitt operator and (c)com-parative results obtained by the LTSD,LTS and LMedS methods.
H.Wang,D.Suter /Pattern Recognition Letters 24(2003)2953–29662963
centere0:0;0:0T,and orientation to horizon direc-tion is0.0°.The results of LTS and LMedS were seriously a?ected by the clustered outliers.How-ever,the LTSD method worked well.
Next,we will apply the LTSD method to real images.
5.3.Experiments with real images
The?rst example is to?t an ellipse in an image of a mouse pad,shown in Fig.7.The edge image was obtained by using Canny operator with threshold 0.07.In total,310data points were in the edge image (Fig.7(b)).The clustered outliers,due to the?ower, occupy50%of the data.Three methods(the LTSD, LTS and LMedS)were applied to detect the mouse pad edge.As shown in Fig.7(c),both LTSD and
LTS correctly found the edge of the mouse pad. However,LMedS fails to detect the edge of the mouse pad.This is because under the condition that the standard variance of inliers is small,the statis-tical e?ciency of LTS is better than LMedS.
Fig.8shows the use of the LTSD method to?t an ellipse to the rim of a cup.Fig.8(a)gives a real cup image.After applying the Prewitt operator, the edge of the cup is detected is shown in Fig. 8(b).We can see there is a high percentage(about 45%)of clustered outliers existing in the edge im-age,external to the rim of the cup(the ellipse we shall try to?t),mainly due to the?gure on the cup. However,the rim of the cup has a symmetric elliptical structure.Fig.8(c)shows that the LTSD method correctly?nds the ellipse in the opening of the cup,while both the LTS and the LMedS fail to correctly?t the ellipse.
5.4.Experiments for the data with uniform outliers
Finally,we investigated the characteristics of the LTSD under uniform outliers.We generated 200data points with40%uniform outliers(see Fig.
9).The ellipse had a standard variance0.5,major axis10.0,minor axis8.0,centere0:0;0:0T,and orientation to horizon direction h is0.0°.The uniform outliers were randomly distributed in a rectangle with left upper cornereà20:0;20:0Tand right lower cornere20:0;à20:0T.We repeated the performance100times and the averaged results were shown in Table2.We can see the LTSD method can also work well in uniform outliers. 6.Conclusion
The fragility of traditionally employed robust estimators:LMedS and LTS,in the presence of clustered outliers has been demonstrated in this paper(a similar story applies more widely––see Wang and Suter,2003).These robust estimators can break down at surprisingly lower percentage of outliers when the outliers are clustered.Thus this paper provides an important cautionary note
Table2
Comparison of the estimated parameters by the LTSD,LTS,and LMedS methods in ellipse?tting with40%randomly distributed outliers
x c y c Major axis Minor axis h(°)
True value0.00.010.08.00.0
The LTSD method)0.1070.12110.0058.024)1.119
The LTS method)0.0090.1209.8777.959)2.117
The LMedS method0.0070.0029.9878.062)0.981
2964H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–2966
to the computer vision community to carefully employ robust estimators when outliers are clus-tered.We also proposed a new method that in-corporates SD into model?tting.The comparative study shows that this method can achieve better performance than the LMedS method and the LTS method especially when large percentages of clus-tered outliers exist in the data and the standard variance of inliers is large.The price paid for the improvement in?tting models is an increase of the computational complexity due to the complicated de?nition of symmetry distance.It takes about Oen2Ttime to compute SD for each p-subset.The time complexity of the proposed method is also related the times that how many p-subsets are needed(according to equation(2.7)).The pro-posed method can be applied to other symmetric shapes?tting and other?elds.
Unfortunately,our method was especially de-signed for spatially symmetric data distributions. For inlier distributions that are not spatially symmetric(including structures that,though they may be symmetric,have large amounts of missing or occluded data so that the visible inliers are not symmetric),the LTSD is not a good choice. However,the LTSD does provide a feasible way to greatly improve achievements of conventional estimators––the LMedS and the LTS,especially, when the data contain inliers(with symmetry)with large variance and are contaminated by large percentage of clustered outliers.
Further work includes?nding a more e?cient and simple de?nition of SD by which the compu-tational complexity can be reduced moderately. We will also extend our horizon to other features of the model studied so that the new method can be used more widely(data with di?erent types of symmetry and reducing the limits required in that symmetry).
Acknowledgements
This work is supported by the Australia Re-search Council(ARC),under the grant A10017082. The authors would like to thank reviewers for their valuable comments.The authors also thank Alireza Bab-Hadiashar for his helpful suggestions.References
Attneave,F.,1995.Symmetry Information and Memory for Pattern.Am.J.Psychol.68,209–222.
Bab-Hadiashar, A.,Suter, D.,1998.Robust optical?ow https://www.wendangku.net/doc/ba7196474.html,put.Vision29(1),59–77.
Bigun,J.,1988.Recognition of local symmetries in gray value images by harmonic functions.In:Proc.Int?l Conf.Pattern Recognition,pp.345–347.
Danuser,G.,Stricker,M.,1998.Parametric model?tting: From inlier characterization to outlier detection.IEEE Trans.Pattern Anal.20(2),263–280.
Fischler,M.A.,Bolles,R.C.,1981.Random sample consensus:
A paradigm for model?tting with applications to image
analysis and automated https://www.wendangku.net/doc/ba7196474.html,mun.ACM24
(6),381–395.
Fitzgibbon,A.,Pilu,M.,Fisher,R.B.,1999.Direct least square ?tting of ellipses.IEEE Trans.Pattern Anal.21(5),476–480.
Hampel,F.R,Rousseeuw,P.J.,Ronchetti,E.,Stahel,W.A., 1986.Robust Statistics:The Approach Based on In?uence Functions.John Wiley&Sons,New York.
Haralick,R.M.,https://www.wendangku.net/doc/ba7196474.html,puter vision theory:The lack https://www.wendangku.net/doc/ba7196474.html,put.Vision Graph.36,372–386.
H€o ssjer,O.,1994.Rand-based estimates in the linear model with high breakdown point.J.Am.Stat.Assoc.89,149–158.
Huber,P.J.,1973.Robust regression:Asymptotics,conjectures and Monte Carlo.Ann.Stat.1,799–821.
Huber,P.J.,1981.Robust Statistics.Wiley,New York. Kirby,M.,Sirovich,L.,1990.Application of the Karhunen-Loeve procedure for the characterization of human faces.
IEEE Trans.Pattern Anal.12(1),103–108.
Kumar,R.,Hanson,A.R.,1989.Robust estimation of camera location and orientation from noisy data having outliers.In: Proc.Workshop on International3D Scences,Austin,TX, pp.52–60.
Lee,K.-M.,Meer,P.,et al.,1998.Robust adaptive segmenta-tion of range images.IEEE Trans.Pattern Anal.20(2), 200–205.
Li,G.,1985.Robust Regression,Exploring Data Tables, Trends and Shapes.Wiley,New York.
Marola,G.,1989.On the detection of the axes of symmetry of symmetric and almost symmetric planar Images.IEEE Trans.Pattern Anal.11(1),104–108.
Meer,P.,Mintz,D.,Rosenfeld,A.,1990.Least median of squares based robust analysis of image structure.In:Proc.
DARPA Image Understanding Workshop,Pittsburgh,PA, pp.231–254.
Miller,W.,1972.Symmetry Groups and Their Applications.
Academic Press,London.
Ming,Y.,Haralick,R.M.,2000.Optical?ow from a least-trimmed squares based adaptive approach.In:Proc.15th International Conference on Pattern Recognition,3,pp.
1052–1055.
Nalwa,V.S.,1989.Line-drawing Interpretation:Bilateral symmetry.IEEE Trans.Pattern Anal.11(10),1117–1120.
H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–29662965
Reisfeld,D.,Wolfson,H.,Yeshurun,Y.,1992.Robust detection of facial features by generalized symmetry.In:Proc.Int?l Conf.Pattern Recognition,Champaign,III.,pp.117–120. Robin,P.L.,1999.Further?ve-point?t ellipse?tting.Graph.
Model Image Proc.61,245–259.
Roth,G.,Levine,M.D.,1990.Segmentation of geometric signals using Robust?tting.In:Proc.10th.Conf.on Pattern Recgnition1,pp.826–831.
Rousseeuw,1984.Least median of squares regression.J Am.
Stat.Assoc.79,871–880.
Rousseeuw,P.J.,Leroy, A.,1987.Robust Regression and Outlier Detection.John Wiley&Sons,New York. Rousseeuw,P.J.,Aelst,S.V.,1999.Positive-Breakdown Robust Methods in Computer Vision.In:Computing Science and Statistics,vol.31.Interface Foundation of North America, Inc.451–460.
Rousseeuw,P.J.,Driessen,K.V.,https://www.wendangku.net/doc/ba7196474.html,puting LTS Regression for Large Data Sets,Technique Report,Univ.
of Antwerp.
Sampson,P.D.,1982.Fitting conic sections to very scattered data:An iterative re?nement of the bookstein algorithm.
Comput.Graph.Image Proc.18,97–108.
Siegel,A.F,1982.Robust regression using repeated medians.
Biometrika69,242–244.Stewart,C.V.,1999.Robust parameter estimation in computer vision.SIAM Rev.41(3),513–537.
Su,M.C.,Chou,C.H.,2001.A modi?ed version of the K-means algorithm with a distance based on cluster symmetry.
IEEE Trans.Pattern Anal.23(6),674–680.
Thomas,P.H.,Simon,J.S.,1992.A cautionary note on the method of least median squares.Am.Stat.46(2),79–83. Wang,H.,Suter,D.,2003.MDPE:A Very Robust Estimator for Model Fitting and Range Image Segmentation,Tech-nical Report(MECSE-2003-3),Monash University. Weyl,H.,1952.Symmetry.Princeton Univ.Press,Princeton, NJ.
Yu,X.,Bui,T.D.,et al.,1994.Robust estimation for range image segmentation and reconstruction.IEEE Trans.Pat-tern Anal.16(5),530–538.
Zabrodsky,H.,https://www.wendangku.net/doc/ba7196474.html,putational Aspects of Pattern Characterization––Continuous Symmetry,PhD thesis, Hebrew Univ.Jerusalem,Israel.
Zabrodsky,H.,Peleg,S.,Avnir, D.,1995.Symmetry as a continuous feature.IEEE Trans.Pattern Anal.17(12), 1154–1166.
Zhang,Z.,1997.Parameter estimation techniques:A tutorial with application to conic?tting.Image Vision Comput.15, 59–76.
2966H.Wang,D.Suter/Pattern Recognition Letters24(2003)2953–2966
等效电路模型参数在线辨识
第四章 等效电路模型参数在线辨识 通过第三章函数拟合的方法可以确定钒电池等效电路模型中的参数，但是在实际运行过程中模型参数随着工作环境温度、充放电循环次数、SOC 等因素发生变化，根据离线试验数据计算得到的参数值估算电池SOC 可能会造成较大的估计误差。因此，在实际运行时，应对钒电池等效电路模型参数进行在线辨识，做出实时修正，提高基于模型估算SOC 的精度。 4.1 基于遗忘因子的最小二乘算法 参数辨识是根据被测系统的输入输出来，通过一定的算法，获得让模型输出值尽量接近系统实际输出值的模型参数估计值。根据能否实时辨识系统的模型参数，可以将常用的参数辨识方法分为离线和在线两类，离线辨识只能在数据采集完成后进行，不能对系统模型实时地在线调整参数，对于具有非线性特性的电池系统往往不能得到满意的辨识结果；在线辨识方法一般能够根据实时采集到的数据对系统模型进行辨识，在线调整系统模型参数。常用的辨识方法有最小二乘法、极大似然估计法和Kalman 滤波法等。因最小二乘法原理简明、收敛较快、容易理解和掌握、方便编程实现等特点，在进行电池模型参数辨识时采用了效果较好的含遗忘因子的递推最小二乘法。 4.1.1 批处理最小二乘法简介 假设被辨识的系统模型： 12121212()()()1n n n n b z b z b z y z G z u z a z a z a z ------+++==++++L L （4-1） 其相应的差分方程为： 1 1 ()()()n n i i i i y k a y k i b u k i ===--+-∑∑（4-2） 若考虑被辨识系统或观测信息中含有噪声，则被辨识模型式（4-2）可改写为： 1 1 ()()()()n n i i i i z k a y k i b u k i v k ===--+-+∑∑（4-3） 式中， ()z k 为系统输出量的第k 次观测值；()y k 为系统输出量的第k 次真值，()y k i -为系统输出量的第k i -次真值；()u k 为系统的第k 个输入值，()u k i -为 系统的第k i -个输入值；()v k 为均值为0的随机噪声。
温室蔬菜常见病虫害及预防措施
龙源期刊网 https://www.wendangku.net/doc/ba7196474.html, 温室蔬菜常见病虫害及预防措施 作者：赫素真 来源：《现代园艺·园林版》2017年第06期 摘要：主要针对温室蔬菜常见病虫害及预防措施进行研究。 关键词：温室蔬菜;病虫害;预防措施 1常见的病害与化学防治 1.1灰霉病 灰霉病是一种比较常见的病害，主要危害的蔬菜是辣椒、茄子类，这种病害是真菌感染引起，且不同的危害部位有不同的表现现状，像蔬菜叶子会发生颜色的变化，会由绿变灰白，且腐烂，果实在初始状态，直接腐烂，腐烂处会出现灰霉，最后果实脱离组织掉落。温室中的潮湿环境极易滋生真菌，所以要控制好温室的含水量。此外，在发病时，要及时对其进行清理，避免此真菌对其他的正常蔬菜产生影响。在进行化学防治时，可以用成分为嘧霉胺，含量不少于70%，剂型为水分散粒剂，用量为9（g/ha）的绿亨1000～1500倍液体进行有效预防。 1.2根腐病 根腐病主要作用于蔬菜的根部，导致根部直接腐烂，且死亡。根腐烂的根本原因还是温室里的水分太多了，又得不到充足的太阳光照射，或者根部生长的土壤环境中土结块或者土中的含水量过多，促使真菌滋生的条件。所以对其进行预防，使其生长的环境和周围的环境保持水分适量。在进行化学防治时，可以采用绿亨1号5000～6000倍液防治。 1.3霜霉病 霜霉病发病时期不确定，主要作用于蔬菜叶，主要表现现状是在叶背呈现霜霉，在正面呈现斑点，同时叶的正反面颜色逐渐褪为黄色。温室环境温度过高、空气中的水分含量较大，以及蔬菜之间的间距没有控制在合理的范围内，都会为滋生相关的病菌提供条件。所以对其进行预防时，针对病菌生存的环境，逐一改变。在进行化学治疗时，可以采用含量为72%、有效成分为霜脲、锰锌的可湿性粉剂防治。 1.4疫病 温室大棚的特点就是温度高、湿度大、严密性强，这样有助于相关蔬菜疫病的产生，主要受影响的蔬菜有茄子、辣椒等。在进行预防时，就要对大棚进行定时的通风换气，同时平衡一下大棚内温度和湿度，还要使其接受一定时间的日光照射，以便进行光合作用，促进蔬菜生
蔬菜病虫害防治技术(汇总)讲义
番茄病虫害防治 一、主要病害及防治 1.番茄灰霉病 防治措施主要是采用以药剂防治为主，结合生态防治的综合防治。在配好的蘸花药液里加入保果宁或50％利得（异菌.福)可湿性粉剂进行蘸花或喷花。加强通风换气，调节温、湿度，避免结露。苗期对床土表面进行灭菌，收获后和种植前彻底清除温室内病残体并集中烧毁或深埋，并对棚室进行消毒处理。 2.番茄菌核病 防治措施主要采用栽培措施和药剂防治相结合。深翻地，将菌核翻至10厘米以下，使其不能萌发；在夏季高温季节利用换茬时用稻草、生石灰或马粪等混合耕地，做起垄小畦，灌满水后铺上地膜，密闭大棚，使土温升高，保持20天。可杀死病菌；加强管理，及时摘除老叶、病叶，清除田间杂草，注意通风排湿，采取滴灌、暗灌；药剂防治在发病初可喷洒40％菌核净可湿性粉剂500倍液或50％农利灵（乙烯菌核利）可湿性粉剂1000～1500倍液等。棚室采用烟雾法或粉尘法施药，亩用10％速克灵（腐霉利）烟剂250～300克，也可于傍晚喷撒6.5％甲霉灵粉尘、百菌清粉尘剂或10％灭克粉尘剂，每亩次1公斤。 3.番茄晚疫病 防治措施主要通过选用耐病品种，结合药剂进行综合防治措施。植株茎、叶茸毛多的品种较耐晚疫病。施足底肥，增施磷、钾肥，提高植株抗病力。在发现中心病株后，要立即全面喷药，集中消灭发病中心，发病中心地块要反复3～4次喷药封锁，并将病叶、病枝、病果和重病株带出田外烧毁。药剂可选用50％安克可湿性粉剂1500～2000倍液或65％安克锰锌可湿性粉剂600～800倍液等。 4.番茄叶霉病 防治措施要以抗病品种为主，结合栽培措施，辅以药剂控制的综合防治措施。选用抗病品种佳粉18号、硬粉8号、仙克1号，金棚1号，中杂105，京丹4、6号，京丹绿宝石和京丹黄玉等对叶霉病菌1.2.3和1.2.3.4生理小种群高抗；栽培措施上，前期作好保温，后期加强通风，降低湿度，防止叶面结露。及时整枝打杈，去掉老病叶；在发病初期可用45％百菌清烟剂每亩250～300克，熏一夜或于傍晚喷撒7％叶霉净粉尘剂，或5％百菌清粉尘剂或10％敌托粉尘剂，每亩1公斤，隔8～10天1次，连续或交替轮换施用。 5.番茄溃疡病 防治措施要采取严格的种子检疫措施，加强栽培管理，结合药剂防治的综合防治措施。对种子进行温汤浸种或药剂浸种，催芽播种；采用高垄栽培。及时清除病株并销毁，病穴要进行生石灰消毒。避免偏施氮肥，禁止大水漫灌，整枝打杈时避免带露水操作；发病时，全田喷洒14％络氨铜水剂300倍、或77％可杀得可湿性微粒粉剂500倍液等。 6.番茄花叶病毒 防治措施上采取以抗病品种为主，结合栽培措施的综合防治。目前国内推广的番茄品种多数具有抗番茄花叶病毒的能力。避蚜防病，大棚采用网纱覆盖风口或利用银灰膜反光，减少室外蚜虫进入，以减轻病毒病的发生。拔除病苗，田间整枝打杈时要病、健株分开操作，
浅析电力系统模型参数辨识
浅析电力系统模型参数辨识 （贵哥提供） 一、现状分析 随着我国电力事业的迅猛发展, 超高压输电线路和大容量机组的相继投入, 对电力系统稳定计算、以及其安全性、经济性和电能质量提出了更高的要求。现代控制理论、计算机技术、现代应用数学等新理论、新方法在电力系统的应用，正在促使电力工业这一传统产业迅速走向高科技化。 我国大区域电网的互联使网络结构更复杂，对电力系统安全稳定分析提出了更高的要求，在线、实时、精确的辨识电力系统模型参数变得更加紧迫。由于电力系统模型的基础性、重要性，国外早在上世纪三十年代就开始了这方面的分析研究，[1,2]国内外的电力工作者在模型参数辨识方面做了大量的研究工作。[3]随后IEEE相继公布了有关四大参数的数学模型。1990年全国电网会议上的调查确定了模型参数的地位，促进了模型参数辨识的进一步发展，并提出了研究发电机、励磁、调速系统、负荷等元件的动态特性和理论模型，以及元件在极端运行环境下的动态特性和参数辨识的要求。但传统的测量手段，限制了在线实时辨识方法的实现。 同步相量测量技术的出现和WAMS系统的研究与应用，使实现在线实时的电力系统模型参数辨识成为可能。同步相量是以标准时间信号GPS作为同步的基准，通过对采样数据计算而得的相量。相量测量装置是进行同步相量测量和输出以及动态记录的装置。PMU的核心特征包括基于标准时钟信号的同步相量测量、失去标准时钟信号的授时能力、PMU与主站之间能够实时通信并遵循有关通信协议。 自1988年Virginia Tech研制出首个PMU装置以来，[4]PMU技术取得了长足发展，并在国内外得到了广泛应用。截至2006年底，在我国范围内，已有300多台P MU装置投入运行，并且可预计，在不久的将来PMU装置会遍布电力系统的各个主要电厂和变电站。这为基于PMU的各种应用提供了良好的条件。 二、系统辨识的概念 系统模型是实际系统本质的简化描述。[5]模型可分为物理模型和数学模型两大类。物理模型是根据相似原理构成的一种物理模拟，通过模型试验来研究系统的
主要病虫害防治方法分析
灵台县无公害蔬菜常见病虫害防治技术 一、十字花科类蔬菜主要病害 白菜软腐病症状：白菜软腐病又称白菜腐烂病、烂疙瘩、酱桶、脱帮等，属细菌性病害一般通过雨水和灌溉水传播,病菌只有与白菜伤口接触,才能侵染发病。多从白菜包心期开始发病，先在菜帮基部出现半透明状浸润斑，逐渐扩大为灰白色，嫩组织腐烂，老组织干缩。后期，由叶帮基部向短缩茎发展，引起根髓腐烂，并溢出灰黄色粘液。植株外叶中午萎蔫，早晚恢复，持续几天后，病株外叶平贴地面，心叶或叶球部分外露，叶柄茎或根茎处髓组织溃烂，流出灰褐色或白色粘稠状物，并散出臭味，轻碰病株即倒折溃烂。潜伏有病菌的组织，入窖后可发生腐烂。较坚实少汁的组织受害时，也是先呈水浸状，逐渐腐烂，但最后病部组织干缩，变为干腐状。此病发生危害期长，在田间、贮运期及市场上都能发生腐烂，造成重大损失。在包心期，遇低温多雨，地势低洼，排水不良时发病重。除白菜外，甘蓝、萝卜等受害也较重。此病除危害十字花科蔬菜外，还可危害马铃薯、番茄、辣椒、洋葱、胡萝卜、芹菜、莴苣等多种蔬菜。 防治：可用72%农用链霉素2000-3000倍，每隔7-10天喷一次，连喷2-3次。 白菜霜霉病症状：真菌性病害，主要发生在叶部。先从外部叶片开始，在病叶正面出现淡黄色小病斑，逐渐扩大成黄绿色不规则病斑。天气潮湿时，叶背面产生白色霜层，气候干旱时，病斑干枯。病原为鞭毛菌寄生霜霉真菌。田间高温多雨是病害常严重发生。早播、播种过密、通风不良、连茬、包心期缺肥易引发病害。 防治：喷施1:0.5:240的波尔多液或75％百菌清可湿性粉剂800倍液、72%霜脲锰锌（杜邦克露、克抗灵、克霜氰）800～1000倍液、每5～7天喷1次。霜霉病的发生与病毒病关系密切，防治应综合考虑。 二、茄科类蔬菜主要病害 茄科晚疫病症状：又称茄科疫病，是番茄、辣椒主要病害之一，该病主要危害叶片及果实，也可危害茎部。一般是中下部叶片先发病。叶片发病首先在叶尖，叶缘处出现不规则暗绿色水渍状病斑，然后病斑扩大并变成褐色。在高湿条件下，病势发展迅速，叶背面的病，病健交界处发生白色霉状物。在干燥条件下，病部干枯，呈清白色，脆而易碎。危害茎部的病斑开始呈暗绿色，后变成黑褐色，稍凹陷，腐败状，病部边缘生有较重的白色霉层，最后表皮腐烂，植株易从腐烂处弯折。果实多发病在青果期表面开始呈灰绿油渍状硬斑块，逐渐变为暗褐色或棕褐色，病斑稍凹陷，边缘
基于最小二乘模型的Bayes参数辨识方法
基于最小二乘模型的Bayes 参数辨识方法 王晓侃1，冯冬青2 1 郑州大学电气工程学院，郑州（450001） 2 郑州大学信息控制研究所，郑州（450001） E-mail ：wxkbbg@https://www.wendangku.net/doc/ba7196474.html, 摘 要：从辨识定义出发,首先介绍了Bayes 基本原理及其两种常用的方法,接着重点介绍了基于最小二乘模型的Bayes 参数辨识,最后以实例用MATLAB 进行仿真,得出理想的辨识结果。 关键词:辨识定义；Bayes 基本原理；Bayes 参数辨识 中国图书分类号：TP273+.1 文献标识码：A 0 概述 系统辨识是建模的一种方法。不同的学科领域，对应着不同的数学模型，从某种意义上讲，不同学科的发展过程就是建立它的数学模型的过程。建立数学模型有两种方法:即解析法和系统辨识。L. A. Zadehll 于1962年曾对”辨识”给出定义[1]:系统辨识是在对输入和输出观测的基础上，在指定的一类系统中，确定一个与被识别的系统等价的系统。一般系统输出y(n)通常用系统过去输出y(n-m)和现在输入u(n)及过去输入u(n-m)的函数描述 y(n)=f(y(n-1),y(n-2)，...，y(n-m y ), u(n),u(n-1),... ,u(n-m u ))=f(x(n),n) x(n)=[y(n-1),y(n-2),...y(n-m y ), u(n),u(n-1),...,u(n-m u )]’ 这里f(,)为未知函数关系，一般情况为泛函数，可以是线性函数或非线性函数，分别对应于线性或非线性系统，通常这个函数未知，但是局部输入输出数据可以测出，系统辨识的任务就是根据这部分信息寻找确定函数或确定系统来逼近这个未知函数。但实际上我们不可能找到一个与实际系统完全等价的模型。从实用的角度来看，系统辨识就是从一组模型中选择一个模型，按照某种准则，使之能最好地拟合由系统的输入输出观测数据体现出的实际系统的动态或静态特性。接下来本文就以最小二乘法为基础的Bayes 辨识方法为例进行分析介绍并加以仿真[4]。 1 Bayes 基本原理 Bayes 辨识方法的基本思想是把所要估计的参数看做随机变量,然后设法通过观测与该参数有关联的其他变量,以此来推断这个参数。 设μ是描述某一动态系统的模型，θ是模型μ的参数，它会反映在该动态系统的输入输出观测值中。如果系统的输出变量z(k)在参数θ及其历史纪录(1) k D ?条件下的概率密度函 数是已知的，记作p(z(k)｜θ，(1) k D ?)，其中(1) k D ?表示（k-1）时刻以前的输入输出数据集 合，那么根据Bayes 的观点参数θ的估计问题可以看成是把参数θ当作具有某种先验概率密 度p （θ，(1) k D ?）的随机变量，如果输入u(k)是确定的变量，则利用Bayes 公式，把参数θ 的后验概率密度函数表示成[2] p (θ，k D ）= p (θ｜z (k ),u(k ), (1) k D ?)=p (θ｜z (k ),(1) k D ?) = (k-1) (k-1) p(z(k)/,D )p(/D ) (k-1)(k-1)p(z(k)/,D )p(/D )d θθθθθ∞∫?∞ (1) 在式(1)中,参数θ的先验概率密度函数p(θ｜(1) k D ?)及数据的条件概率密度函数p(z(k)｜θ，
蔬菜常见虫害防治的化学农药
蔬菜——常见虫害防治的化学农药 1、菜青虫：311菜蛾敌600-1000倍液、2.5%功夫乳油2000-5000倍液、2.5%敌杀死乳油2000-5000倍液、5%卡死克乳油2000-3000倍液、5%抑太保乳油2000-3000倍液、50%辛硫磷乳油1000倍液、10%天王星乳油1000倍液、21%灭杀毙乳油3000倍液。注意：抑太保要比一般农药提早3天施药。 2、小菜蛾：311菜蛾敌600-1000倍液、2.5%功夫乳油2000-5000倍液、2.5%敌杀死乳油2000-5000倍液、5%卡死克乳油2000-3000倍液、5%抑太保乳油2000-3000倍液、5%锐劲特悬浮剂3000倍液、5%农梦特2000倍液。 3、斜纹夜蛾、甜菜夜蛾、银纹夜蛾、甘蓝夜蛾：25%辛硫灭扫利乳油2500-4000倍液、夜蛾灵800-1200倍液、夜蛾净1500-2000倍液、克蛾宝2000-3000倍液、10%氯氰菊酯乳油2000-5000倍液、2.5%功夫乳油5000倍液、2.5%木虱净乳油1500倍液。注意：辛硫灭扫利、夜蛾灵、夜蛾净不能与碱性药混用。 4、黄曲条跳甲成虫、幼虫：80%敌百虫可湿性粉剂800-1000倍液、农地乐1000-1500倍液、扑甲灵600-1000倍液、50%辛硫磷乳油1000倍液、21%增效氰乌乳油4000倍液。注意：农地乐应注意标签说明。喷施扑甲灵时应从田块四周向中间喷药，防止虫逃跑。 5、蚜虫（瓜蚜）：快杀灵3000-5000倍液、一遍净3000-5000倍液、杀灭菊酯8000-10000倍液、灭百可5000-8000倍液、40%乐果乳油1000-2000倍液、10%大功臣可湿性粉剂2500倍液。注意：快杀灵、杀灭菊酯不与碱性药混用。盛装灭百可的容器要冲洗3次后埋掉。 6、红蜘蛛：73%克螨特乳油3000-5000倍液、爱福丁3000-4000倍液、5%卡死克乳油1000-1500倍液、20%螨克乳油1000-1500倍液。 7、黄守瓜、瓜绢螟：80%敌百虫可湿性粉剂800-1000倍液、50%辛硫磷乳油1000-1500倍液、10%氯氰菊酯2000-5000倍液、灭杀毙8000倍液。注意：瓜豆对50%辛硫磷敏感，高温慎用。 8、蓟马：爱卡士1000-2000倍液、拉硫磷1500-2000倍液、0%辛硫磷乳油1000-2000倍液、0%氯马乳油1000-2000倍液、0%乐果乳油1000倍液。注意：20%氯马乳油不与碱性药混用。 9、瓜食蝇：2.5%溴氰菊酯乳油2000-3000倍液、香蕉或菠萝皮40份，加80%敌百虫晶体（其它农药）0.5份，加香精1份，加水调成糊状毒饵诱杀。每亩20个点，每点25克、50%敌敌畏乳油1000倍液。 10、潜叶蝇：2.5%溴氰菊酯1500-3000倍液、爱卡士1000-2000倍液、抑太保乳油 2000-3000倍液、21%灭杀毙乳油8000倍液、25%喹硫磷乳油1000倍液、80%敌百虫可湿性粉剂1000倍液。 11、豆秆蝇：⑴50%辛硫磷乳油1000倍液。⑵2.5%溴氰菊酯1500-3000倍液。注意：用时应清除
蔬菜常见病虫害防治方法
（一）黄瓜 一. 霜霉病疫病： 1.烯酰吗啉 2. 霜脲锰锌 3.菌霜杰 4.甲霜灵锰锌 二. 细菌性角斑病.叶枯病：1.中生菌素 2. 春雷霉素 3.噻唑锌 三. 白粉病黑星病：1. 乙嘧酚 2. 3.醚菌酯 4. 新秀戊唑醇 5. 白粉速净 6. 斑星杰 四. 炭疽病叶斑病：1. 苯醚甲环唑 2. 克菌杰 3.醚菌酯 五蔓枯病枯萎病根腐病：1. 枯草芽孢杆菌 2. 青枯立克 4. 中生菌素 5.络氨铜灌根 六. 灰霉病菌核病：1. 嘧霉胺 2. 丁子香酚 3. 统防统治 4. 异菌脲 5.腐霉利 6. 烟酰胺 (二）西瓜甜瓜茭瓜 一. 炭疽病.叶斑病：1.苯醚甲环唑 2. 克菌杰 3.斑星杰 4. 醚菌酯 5. 新秀戊唑醇 6.乙嘧酚 二. 蔓枯病枯萎病：1. 克菌杰 2. 青枯立克 3. 枯草芽孢
4. 中生菌素（上述产品喷雾或者少兑水拌成糊状涂抹病部） 6. 络氨铜灌根 三. 细菌性叶枯病.软腐病果腐病：1.中生菌素 2. 春雷霉素 3.噻唑锌 四. 疫病霜霉病：1.烯酰吗啉 2. 霜脲锰锌 3.菌霜杰 4.甲霜灵锰锌 五.病毒病：1. 菌毒清.吗啉胍 2. 氮苷.吗啉胍 3. 吗啉胍.乙酸铜 （三）番茄辣椒茄子 一. 晚疫病： 1.烯酰吗啉 2. 霜脲锰锌 3.菌霜杰 4. 甲霜灵锰锌 二.早疫病： 1. 苯醚甲环唑 2. 戊唑醇 3.醚菌酯 三. 灰霉病菌核病： 1. 嘧霉胺 2. 丁子香酚 3. 统防统治 4. 异菌脲 5.腐霉利 6. 烟酰胺 四. 叶霉病： 1. 春雷霉素 2. 新秀戊唑醇 3. 异菌脲
五病毒病： 1. 菌毒清.吗啉胍 2. 氮苷.吗啉胍 3. 吗啉胍.乙酸铜 六. 青枯病溃疡病茎基腐黄萎病枯萎病：1. 中生菌素 2. 青枯立克 3. 枯草芽孢杆菌 4.春雷霉素 5.络氨铜灌根 七. 细菌性叶斑病疮痂病：1.中生菌素 2. 春雷霉素 3.噻唑锌 （四）菜豆芸豆 一. 锈病炭疽叶斑病：1. 苯醚甲环唑 2.斑星杰 3. 新秀戊唑醇 4. 乙嘧酚 5. 克菌杰 6. 醚菌酯 二.细菌性疫病：1. 中生菌素 2 噻唑锌 三.花叶病毒病：1. 菌毒清.吗啉胍 2. 氮苷.吗啉胍 3. 吗啉胍.乙酸铜 四.枯萎病根腐病茎基腐：1.中生菌素 2.青枯立克 4. 枯草芽孢杆菌 5.络氨铜灌根
Bouc-Wen 滞回模型的参数辨识
上海交通大学 硕士学位论文 Bouc-Wen滞回模型的参数辨识及其在电梯振动建模中的应用 姓名：周传勇 申请学位级别：硕士 专业：机械设计及理论 指导教师：李鸿光 20080201
Bouc-Wen滞回模型的参数辨识 及其在电梯振动建模中的应用 摘 要 电梯导靴是连接轿箱系统与导轨的装置，它能起到导向和隔振减振的作用。同时，在电梯的运行过程中它又将导轨由于制造或安装所造成的表面不平顺度传递给轿箱系统，从而引起轿箱系统的水平振动。国内外学者在电梯水平振动的建模和分析中，往往把导靴视为线性弹簧-阻尼元件来建模而忽略了非线性因素。事实上导靴与导轨之间存在非线性的迟滞摩擦力，本文通过实验的方法，采用Bouc-Wen 滞回模型来建立导靴-导轨非线性摩擦力模型。 Bouc-Wen滞回模型因其微分形式的非线性表达式而使得其参数辨识存在较大的困难，本文利用模型中部分参数的不敏感性，通过数学变换将非线性参数辨识问题转化为线性参数辨识问题，从而使得问题大大简化，参数辨识的效果也能满足要求。 基于以上导靴-导轨间摩擦力模型，本文进而建立了轿箱-导轨耦合水平振动动力学模型，该模型将轿箱系统等效为2自由度的平面运动刚体，将导靴等效为质量-弹簧-阻尼单元，同时考虑了导靴-导轨间的非线性摩擦力，以及导靴靴衬与导轨间接触的不连续性等。 在建立了轿箱-导轨耦合水平振动动力学模型后，利用Matlab/Simulink，建立了相应的仿真模型，开展了几种典型导轨不
平顺度激励（弯曲、失调和台阶）下的仿真分析。研究结果表明，这些分析对于电梯结构优化设计和动力学建模与分析有理论指导意义。 关键词：迟滞，参数辨识，非线性，动力学建模，系统仿真
主要蔬菜常见病虫害防治
主要蔬菜常见病虫害防治 日期：2013-10-14 10:40 作者：来源：湖南湘阴县农业局点击：2103 一、辣椒主要病虫害的防治 1、苗床期主要病虫害。①主要病害有灰霉病、炭疽病。分别用速克宁和甲基托布津防治，每隔15天喷一次药液。②主要虫害是蚜虫，可用敌蚜螨、克螨特等药物进行防治。 2、生长结果期的主要病虫害。病害主要有：真菌性疮痂病、炭疽病、疫病可选用可杀得或甲基托布津、瑞毒霉或百菌清防治；病毒病一般引起花叶、卷叶，可用病毒A或辣椒卷叶灵喷雾预防；白绢病、青病重在预防，加强土壤消毒，多施石灰。在发病初期分别5%多菌灵500倍液淋蔸进行防治。 二、茄子的主要病虫害防治 1、主要病害。①育苗初期易发生猝倒病和真菌性病害，防病药剂有速克宁、甲基托布津、百菌清等。 ②生长发育期后的病害有黄萎病、绵疫病。防治重在轮作，严格土壤消毒，重施生石灰。发病始期用可杀得喷雾防治。 2、主要虫害。主要有蚜虫、棉铃虫、十八星瓢虫、茶蟥螨等。蚜虫可用敌蚜螨、乐果等药防治；棉铃虫、十八星瓢虫用功夫或甲维盐等药防治；茶蟥螨用哒螨酮、克螨特防治。 三、白菜类病虫害防治 1、病害有很多，但发生最普遍、最严重的有软腐病、霜霉病和病毒病。①软腐病，发病初期及时拨除病株，并在周围土壤撒少量石灰；②霜霉病，发病初期可用以下药剂防治：75%甲霜灵800倍喷雾，40%乙磷铝150-200倍喷雾；③病毒病，苗期及时防治蚜虫，用吡蚜酮1000倍喷雾。 2、虫害的防治。虫害主要是菜青虫、甜菜夜蛾、蚜虫等虫害，可用敌杀死、甲维盐等药剂防治。 四、莴苣类病虫害防治 1、莴苣的病害主要有：霜霉病和软腐病。①霜霉病，在发病初期用25%甲霜灵800倍或45%代森铵900-1000倍喷施，每隔5-7天喷一次，连续2-3次。②软腐病，发病初期及时拨除病株，并在病穴四周撒少量石灰，同时，可用农链霉素200克/升或敌克松500-1000倍液或50%代森铵800-1000倍液喷施，每5-7天喷一次，连续2-3次。 2、莴苣的主要虫害是蚜虫和蜗牛。分别用敌蚜螨、蜗牛灵进行防治。 五、芹菜的病虫害防治 1、病害。苗期主要有猝倒病，发现病株，及时清除，并撒施药土。可用35%多菌灵拌细土25千克撒施，或用50%多菌灵1000倍液或用50%代森铵1000倍液或70%百菌清800倍液喷雾。成株病害主要有叶枯病和早疫病。可用70%代森锌可湿性粉剂500倍液、40%甲基托布津600-800倍液喷雾，隔5-7天交替用药。 2、主要虫害。芹菜虫害主要有斜纹夜蛾和蚜虫。斜纹夜蛾可用25%灭幼脲3号500倍液或功夫或甲维等药防治。蚜虫可用蚍蚜西酮、吡虫啉单剂防治。 六、黄瓜病虫害防治。 1、黄瓜的主要病害有枯萎病、疫病、霜霉病、细菌性角斑病，可分别用代森铵、可杀得、甲霜灵、农用链霉素防治。 2、虫害主要有黄守瓜虫、蚜虫等，可用敌蚜螨或拟除虫菊酯类等药物喷雾防治。
蔬菜虫害分类
蔬菜主要病虫害种类 十字花科蔬菜病虫：主要有小菜蛾、菜青虫、黄曲条跳甲、斜纹夜蛾、甜菜夜蛾、蚜虫、菜螟、软腐病、黑腐病、菌核病、霜霉病、炭疽病、白锈病等。 豆科蔬菜病虫：主要有豆荚螟、豆蚜、斜纹夜蛾、豆芫菁、斑潜蝇、烟粉虱、朱砂叶螨、白粉病、炭疽病、枯萎病、锈病、轮纹病、病毒病等。 葫芦科蔬菜病虫：主要有黄守瓜、芫菁、瓜娟螟、蚜虫、蓟马、斑潜蝇、烟粉虱、灰霉病、疫病、霜霉病、白粉病、炭疽病、黑斑病、叶斑病、蔓枯病、枯萎病、细菌性角斑病、病毒病等。 茄科蔬菜病虫：主要有蓟马、棉铃虫、二十八星瓢虫、茶黄螨、青枯病、早疫病、晚疫病、褐纹病、炭疽病、白粉病、灰霉病、根结线虫病等。 葱蒜类蔬菜病虫：主要有蓟马、潜叶蝇、斜纹夜蛾、甜菜夜蛾、灰霉病、疫病、霜霉病等。 蔬菜主要病虫害种类及适用农药品种 乐斯本基立甲霜灵百菌清农药杂谈分类：农业 病虫种类： 1、苗期病害（猝倒病、立枯病）发生作物：叶菜类、瓜类、茄果类苗期。适用农药：恶习霉灵、甲基立枯磷、苗菌净、多氧霉素、甲霜灵锰锌。 2、霜霉病类发生作物：瓜类、葱类、叶菜类。适用农药：甲霜灵锰锌、百菌清、氰霜唑、杀毒矾、克露、氟吗锰锌、疫霜灵。 3、疫病类型发生作物：瓜类、葱类、叶菜类。适用农药：甲霜灵锰锌、百菌清、氰霜唑、杀毒矾、克露、氟吗锰锌、疫霜灵。 4、早疫病类发生作物：辣（甜）椒、黄瓜、葱、芋等。适用农药：代森锰锌、可杀得、百菌清、甲霜灵锰锌、杀毒矾、异菌。 5、晚疫病类发生作物：番茄、马铃薯。适用农药：甲霜灵锰锌、氟吗锰锌、氰霜唑、金雷多米尔、克露、蓝保、可杀得。 6、炭疽病类发生作物：甜（辣）椒、白菜、黄瓜、菜豆。适用农药：炭特
瓜类蔬菜主要病虫害防治技术
瓜类蔬菜主要病虫害防治技术 夏季高温高湿有利于多种瓜类蔬菜病虫的发生，加上春种蔬菜收获后田间遗留大量病虫源，使防治工作更加困难。因此，夏季种植瓜类蔬菜应注意采取以下病虫害防治措施： 一、春种蔬菜收获后彻底清除田间残枝败叶，铲除田边杂草，搞好田园清洁。抓住时机赶晴天翻土晒透，以杀灭部分病菌，减少田间病虫源。 二、注意选用前茬非瓜类的农田，避免瓜类连作。育苗前要进行苗床土壤消毒，可有效防治瓜类蔬菜苗期病害。方法有：①每平方米可用1.0-1.5克绿亨一号（≥99%噁霉灵可溶性粉剂）和4克绿亨2号（80%多·福·锌可湿性粉剂），与细土20千克充分掺匀后取1/3撒在畦面，余下2/3播种后做盖土；②用绿亨8号（15%噁霉灵水剂）稀释800-1000倍苗床喷洒或淋施；③用绿亨4号（3％甲霜· 噁霉灵水剂）12-18毫升/平方米，300-500倍液喷雾或淋湿；④用绿亨一号3000-4000倍液苗床淋湿。 三、种植规格不宜过密，比春种稍疏。上棚后注意引蔓整形，保证枝叶间通透性。施肥要注意合理配方，防止偏施过量氮肥，适当提高磷钾比例，并适时补充叶面喷施微肥，如绿亨天宝等，以提高植株抗病力。 四、夏季瓜类主要病害有：霜霉病、疫病、白粉病、炭疽病、叶斑病、枯萎病、蔓枯病等。要注意使用防病杀菌剂喷洒，一般每10天左右喷一次，针对不同病害，选择相应杀菌剂。 1、疫病和霜霉病： 疫病和霜霉病是瓜类生产上的重要病害，在瓜类各生长期均可发生。其有效防治药剂有绿亨80%烯酰吗啉水分散粒剂1500-2000倍液、绿亨铜师傅86.2%氧化亚铜1500倍液，绿亨飓风70%烯酰·嘧菌酯1500倍液、72%霜脲·锰锌可湿性粉剂500-1000倍液、58%甲霜· 锰锌可湿性粉剂600-800倍液等。交替轮换使用上述药剂，可延缓病菌抗药性的产生，取得更好的防治效果。在喷施上述药剂的同时，加入绿亨天宝可在防治蔬菜病害的同时增强瓜类自身的免疫力，提高品质，增加产量。 2、白粉病： 瓜类白粉病分布广泛，全球及我国南北菜区均有发生，是危害瓜类生产的重要病害之一，黄瓜、甜瓜、南瓜、西葫芦等发生较重。可选择以下药剂进行喷雾防治：绿亨8%氟硅唑微乳剂1000-1500倍液、绿亨80%戊唑醇3000-4000倍液、250克/升苯醚甲环唑乳油3000-5000倍液，严重时配合绿亨铜师傅86.2%氧化亚铜1500倍液可解除病害抗性，通知可防治霜霉和多数细菌性病害。 3、枯萎病：
蔬菜主要病虫害综合防治技术
蔬菜主要病虫害综合防治技术 随着农村种植业结构的不断调整，蔬菜生产已成为我县农业的重要产业之一。蔬菜的产量、产值、经济效益均居其他农作物之首，是我县广大蔬菜种植户的主要经济来源。 近年来，我县蔬菜面积迅速扩大，品种日益增多，栽培方式多样，复种指数提高，特别是大量种植保护地和反季节蔬菜，使得多数蔬菜得以周年生产，很大程度上打破了原有蔬菜作物的生态系统的平衡，加之其它经济作物、观赏植物面积的不断增加，为多种害虫发生提供了丰富的食料条件，常规性病虫频频暴发，突发性病虫威胁增大，病虫发生种类逐步增多，几乎任何一种蔬菜都可遭受病虫害的侵染。另外，由于蔬菜种类增加、保护地面积扩大、栽培方式多样、反季节栽培及调运频繁等缘故，造成原来一些次要的病虫害逐渐上升为主要病虫害；还有因长期、过量、单一使用农药，导致病虫抗药性越来越强，防治难度越来越大。据统计,我县常年蔬菜病虫发生面积80多万亩次，防治面积90多万亩次，通过防治挽回损失1.5万吨以上。 我县蔬菜病虫害发生特点 据调查,我县发生的蔬菜病虫害有200余种,其中常年发生的虫害有5O多种,病害有70多种。在日光温室大棚蔬菜病虫中,病害重于虫害,冬春季节的霜霉病、灰霉病偏重发生，其它病虫较轻。在露地菜病虫中,虫害重于病害,病虫发生季节性明显,冬春病害重,夏季和春夏之交病虫并重,秋季虫害重。全年病害发生高峰为苗期的立枯病、猝倒病和6月份的疫病、枯萎病等，特别是辣椒、豇豆、黄瓜的疫病和瓜类的枯萎病等。全年虫害发生高峰在7～9月份，主要是小菜蛾、斜纹夜蛾、甜菜夜蛾、蚜虫、美洲斑潜蝇。 当前蔬菜病虫防治中存在的主要问题 一、对病虫害缺乏了解：目前，蔬菜病虫害种类较多，尤其是保护地蔬菜，在高温高湿条件下，病虫害发生危害严重，而不少菜农不能正确识别病虫害，缺乏植物保护基本知识和病虫害综合防治技能，不懂得贯彻“预防为主，综合防治”的植保方针，依赖、误用和不合理使用化学农药的情况较为普遍。在病虫害暴发时"病急乱投医"，不能做到对症下药，经常会出现打“保险药”、打“马前炮药”、“马后炮药”、“乱打药”、打“高毒剧毒药”及“打错药”或不严格控制农药安全间隔期等现象。长期使用单一种农药:有的菜农发现某种农药效果好，就长期使用，即使发现该药对病虫害的防治效果下降，也不换用其它农药产品，而是采取加大药量的方法，结果随着用药量的增加，病虫害的抗药性也不断增强。如某菜农在防治番茄病毒病时，长期使用20%病毒A，而不更换其它的农药，导致防治病毒病效果降低。 二、缺乏无公害生产关键技术：蔬菜病虫害的防治，长期以来依赖于化学农药防治，其弊端越来
参数辨识示例 报告
参数辨识 参数辨识的步骤 飞行器气动参数辨识是一个系统工程，包括四部分：①试验设计，使试验能为辨识提供含有足够信息量且信息分布均匀的试验数据；②气动模型结果确定，即从候选模型集中，根据一定的准则和经验，选出最优的气动模型构式；③气动参数辨识，根据辨识准则和数据求取模型中待定参数，这是气动辨识定量研究的核心阶段；④模型检验，确认所得气动模型是否确实反映了飞行器动力学系统中气动力的本质属性。这四个部分环环相扣，缺一不可，要反复进行，直到对所得气动模型满意为止。 参数辨识的方法 参数辨识方法主要有最小二乘算法、极大似然法、集员辨识法、贝叶斯法、岭估计法、超椭球法和鲁棒辨识法等多种辨识方法。虽然目前参数辨识的领域己经发展了多种算法，但是用于气动参数估计的算法主要有：极大似然法(ML)，广义Kalman滤波(EKF)法，模型估计法(EBM )、分割及多分割算法(PIA及MPIA)、最小二乘法，微分动态规划法等。 因为最小二乘法和极大似然法是两种经典的算法，目前己经发展得相当成熟。最小二乘法适于线性模型的参数辨识，可以用于飞行器系统辨识中很多的线性模型，如惯性仪表误差系数的辨识，线性时变离散系统初始状态的辨识及多项式曲线拟合等。目前最小二乘法已经广泛应用于工程实际中。而极大似然算法因其具有渐进一致性、估计的无偏性、良好的收敛特性等特点而被广泛应用于飞行器参数辨识领域。 最小二乘法大约是1975年高斯在其著名的星体运动轨道预报研究工作中提出来的。后来，最小二乘法就成了估计理论的奠基石。由于最小二乘法原理简单，编程容易，所以它颇受人们重视，应用相当广泛。 极大似然估计算法在实践中不断地被加以改进，这种改进主要表现在三个方
蔬菜主要病虫害种类及适用农药品种
蔬菜主要病虫害种类及适用农药品种 1、苗期病害发生作物：叶菜类、瓜类、茄果类苗期。合用农药：恶习霉灵、甲基立枯磷、苗菌净、多氧霉素、甲霜灵锰锌。 2、霜霉病类发生作物：瓜类、葱类、叶菜类。合用农药：甲霜灵锰锌、百菌清、氰霜唑、杀毒矾、克露、氟吗锰锌、疫霜灵。 3、疫病类型发生作物：瓜类、葱类、叶菜类。合用农药：甲霜灵锰锌、百菌清、氰霜唑、杀毒矾、克露、氟吗锰锌、疫霜灵。 4、早疫病类发生作物：辣椒、黄瓜、葱、芋等。合用农药：代森锰锌、可杀得、百菌清、甲霜灵锰锌、杀毒矾、异菌。 5、晚疫病类发生作物：番茄、马铃薯。合用农药：甲霜灵锰锌、氟吗锰锌、氰霜唑、金雷多米尔、克露、蓝保、可杀得。 6、炭疽病类发生作物：甜椒、白菜、黄瓜、菜豆。合用农药：炭特灵、施保功、农抗120、年夜生、绿亨一号、多福、多氧霉素。 7、白粉病、锈病类发生作物：瓜、豆、葱类。合用农药：晴菌唑、烯唑醇、三唑酮、百菌清、武夷霉素、农抗120、多氧霉素。 8、黑斑病类发生作物：白菜、甘蓝、花椰菜、萝等。合用农药：百菌清、杀毒矾、代森锰锌、农抗120。 9、根腐病类发生作物：瓜类、茄果类。合用农药：根腐灵、多菌灵、双效灵、壮生、百菌通、多氧霉素。 菌灵、根病净、绿亨二号、克菌、多氧霉素。 11、细菌性叶斑类发生作物：黄瓜、菜豆、甜椒、菜豆。合用农药：杀菌王、龙克菌、克菌康、硫酸、链霉素、新植霉素、波尔多液。 12、细菌性青枯、软腐、黑腐、黑斑病类发生作物：茄果类、十字花科。合用农药：杀菌王、龙克菌、克菌康、硫酸、链霉素、新植霉素、波尔多液。 13、病毒类发生作物：各类蔬菜。合用农药：病毒A、菌毒清、菌毒必克、素灭星植病灵、毒畏、镇毒。 14、小菜蛾、菜青虫等害虫发生作物：十字花科蔬菜。合用农药：菜喜、阿巴虫净、锐劲特、卫灵安打、阿维菌素、米满、千虫克。
遗传算法工具箱识别(GA)Bouc-Wen模型参数辨识_识别
Bouc-Wen模型因数字处理方便简单而得到较为广泛的应用，力可以表示为： 利用遗传算法工具箱（GA）对Bouc-Wen模型进行参数识别。 实验数据来源于对磁流变阻尼器（MR damper）进行性能测试，试验获得的数据包括力F，位移x，采用频率已知，速度和加速度可以由位移求导得出。 参数识别出现程序如下：（文件名：Copy_0_of_BoucWen） function j=myfung(x) y0=[0]; yy=y0; tspan=[]'; s=[]'; v=[]'; Ft=[]'; rr=max(size(s));%计算数据个数 i=1; while (i
蔬菜种类及主要病虫害防治
蔬菜种类、主要虫害及其防治 一、十字花科蔬菜 （一）蔬菜种类：包括大白菜、甘蓝、花椰菜、萝卜、小青菜等 （二）主要虫害及其防治 1、甜菜夜蛾：1%甲维盐、15%茚虫威、5%氟虫晴、5%氟铃脲、5%氟啶脲等可以防止。 2、菜青虫：5%氟啶脲、1%甲维盐、1.8%阿维等可以防止。 3、小菜蛾：15%茚虫威、1%甲维盐、5%氟虫晴、1.8%阿维、5%氟啶脲等可以防止。 4、甘蓝夜蛾：5%氟啶脲 5、甘蓝蚜：10%吡虫啉、48%毒死蜱、4.5%高氯等 6、大猿叶虫：50%辛硫磷80%敌敌畏， 7、黄翅菜叶蜂：50%辛硫磷80%敌敌畏， 8、同型巴蜗牛：30%除涡特、50%涡克星 9、萝卜蚜：48%毒死蜱、10%吡虫啉、20%啶虫脒 10、萝卜地种蝇：50%辛硫磷 （三）主要病害及其防治 1、霜霉病：甲霜灵、氟吗啉、75%百菌清 2、软腐病：链霉素、氢氧化铜、 3、病毒病： 4、细菌性黑腐病：10%吡虫啉、链霉素、75%百菌清、甲霜灵、福美双 5、黑斑病：百菌清、腐霉利、福美双、多菌灵 6、炭疽病：百菌清、多菌灵、苯醚甲环唑、福美双 二、茄科 （一）蔬菜种类：番茄、茄子、辣椒、马铃薯 （二）主要虫害及其防治 1、二十八星瓢虫：48%毒死蜱、90%茚虫威、5%氟虫晴等可以防止。 2、茶黄螨：15%哒螨灵、48%毒死蜱、73%炔螨特、1.8%阿维等可以防止。 3、烟青虫：15%茚虫威、5%氟虫晴、1.8%阿维、5%氟啶脲等可以防止。 4、班须蝽：40%氧化乐果、20%灭多威、80%敌敌畏 （三）主要病害及其防治 1、猝倒病：甲霜灵、75%百菌清、霜霉威、恶霉威 2、灰霉病：30%百菌清·15腐霉利烟剂 3、病毒病：10%吡虫啉、5%菌毒清 4、番茄、马铃薯晚疫病：50%多菌灵、75%百菌清、甲霜灵、氢氧化铜、腐霉利 5、番茄、马铃薯早疫病：60%多菌灵、75%百菌清、甲霜灵、氢氧化铜 6、番茄叶霉病：45%百菌清、50%腐霉利烟剂 7、番茄根结线虫病：80%敌敌畏、50%辛硫磷 8、茄子绵疫病：甲霜灵、霜霉威、百菌清 9、茄子褐纹病：初期可用10%百菌清和20%腐霉利烟剂 10、茄子黄萎病：多菌灵、苯菌灵、苯醚甲环唑 11、辣椒疫病：霜霉威、甲霜灵、烯酰吗啉、氟吗啉、恶霜灵 12、辣椒疮痂病：农用链霉素、氢氧化铜、加瑞农
白菜类蔬菜病虫害识别与防治
白菜类蔬菜病虫害识别与防治 白菜类蔬菜属于十字花科芸苔属一年生或两年生草本植物，按照它们的形态，可以分为大白菜、小白菜、菜薹、乌塌菜以及芜菁等。 大白菜就是结球白菜，在全国都有分布，以北方栽培为主，是华北秋季栽培的主要蔬菜。 小白菜就是不结球白菜，我国北方俗称它为小油菜，南方地区成为青菜。 菜薹又称菜心，是我国著名的特菜之一。 乌塌菜含有丰富的矿物质和维生素，所以人们又叫它“维生素菜”。 总的来说，白菜类蔬菜的叶片、叶球、花薹和嫩茎等，富含各种维生素和矿物质，符合人们的食用习惯，再加上它们栽培面积广、产量高、耐于贮运、供应期长，所以，白菜类蔬菜称得上我国最重要的蔬菜之一。 在白菜类蔬菜的生长过程中，经常会出现各种各样的病虫害，如果不及时地进行防治、或是防治方法不对的话，会对蔬菜的产量和品质造成严重影响。接下来的节目中，我们首先来认识几种典型的病虫害，然后再看看如何防治。 病害 白菜类蔬菜的病害可以分为侵染性病害和生理性病害，而侵染性病害又分为细菌性病害、真菌性病害和病毒性病害。 细菌性病害
细菌性病害是植株由于受到细菌侵染而引起的一种病害，一般表现出坏死、腐烂、萎蔫、畸形的特点。 软腐病 您瞧这颗大白菜，叶球直接露出来了，叶柄基部和根茎处的心髓部组织已经完全腐烂，充满了灰黄色的粘稠物，还散发出很大的臭味，这就是软腐病的典型症状。 软腐病又叫做烂疙瘩、烂葫芦、腐烂病、水烂病等，发生极为普遍。它的主要特点是轻轻一掰，植株就倒了，病部呈黏滑软腐状．并伴有恶臭味。小白菜、菜心等白菜类蔬菜发生软腐病时，症状与大白菜基本相似。 拿大白菜来说，大白菜定植后直到形成心叶的这个过程是长外叶的过程，这个过程中软腐病不会发生。而当植株外叶即将罩严地面的时候，大白菜渐渐进入壮心期，这时软腐病开始发生，从壮心开始至收获的整个过程中都有发病的可能，如果在这个时期，一开始植株外围的叶片在烈日下表现出萎蔫，但早晚尚能恢复，慢慢儿地外叶不能恢复的话，那您就要注意了，这有可能是得软腐病的早期症状。 黑腐病 您瞧这棵大白菜，从叶片的边缘往两侧和里边扩展，形成“V”字形黄褐色枯斑，病斑的周边呈淡黄色，这就是黑腐病的症状。以后，病原菌还会沿着叶脉向里扩展，形成大块黄褐色病斑或网状黑脉，并感染叶柄。大白菜一般在莲座期以后容易得这种病，它也是由细菌引起的。
蔬菜病虫害防治
蔬菜——不用农药怎么防止病虫害？ ?A+ ?A- 2017-04-05 10:01:47农产信息网关注 说实话，作物病虫害防治不用农药很不现实，比较麻烦且费人力，但要有想学习如何不用农药来防治病虫害的朋友可以来看一下。 蔬菜虫害是蔬菜种植户们非常头疼的问题，若是用传统的农药喷洒方式解决虫害，会因为农药残留影响蔬菜的品质。这里和大家分享一下蔬菜虫害的科学防治方法。
一、伴生植物法： 1.青椒和大蒜间作。由于大蒜有一种特殊气味，能使为害青椒的害虫闻之即逃，避免青椒受害。 2.番茄和甘蓝套种。番茄的叶片会散发一种特殊的气味，可驱赶走为害甘蓝的菜青虫和蚜虫。除此之外，这两种蔬菜吸收的营养有很强的互补性，能充分发挥地力。 3.葱头与胡萝卜间作。它们各自散发的气味能驱走相互间的害虫。若单一种植胡萝卜，为防止虫害，可在地内或四周种上几棵葱头，这也能起到驱虫的作用。 并非所有的蔬菜都可以间作，如甘蓝和芹菜、黄瓜和番茄等不宜间作在一起，因为它们各自的分泌物能抑制对方的生长。 这种方法对适用于种植户和家庭小面积种植者。 二、自然材料治虫
1.草木灰液治虫。草木灰10千克对水50千克浸泡24小时，取滤液喷洒可有效地防治蚜虫、黄守虫。若葱、蒜、韭菜受种蝇、葱蝇的蛆虫危害，每亩沟施或撒施草木灰20~30千克，既治蛆又增产。 2.红糖液防治病。害红糖300克溶于500毫升清水中，加入10克白衣酵母，置于温室或大棚内，每天搅拌1次，发酵15~20天，待其表面出现白膜层为止。然后将此发酵液再加入米醋、烧酒各100克，对入100千克水。每隔10天1次，连喷4~5次，防治黄瓜细菌性斑点病和灰霉病有良好效果。 3.兔粪治地老虎每10千克水加兔粪1千克，装入瓦缸内密封沤15~20天，用时搅拌均匀，浇于瓜菜根部，可防治地老虎。 4.尿洗合剂治菜蚜用洗衣粉、尿素、水按1∶4∶400的比例制成混合液，可防治菜蚜，杀虫率达90％以上。 5.猪胆液治病虫10％浓度的猪胆液加适量小苏打、洗衣粉，能防治茄子立枯病、辣椒炭疽病，能驱赶长豆角、四季豆、瓜类等蔬菜上的蚜虫、菜青虫、蜗牛等多种害虫。稀释液可保持10天有效。 6.大蒜、番茄叶巧杀红蜘蛛用大蒜（捣烂成泥状）2份，水1份混拌均匀，取其滤液喷治。或用新鲜的番茄叶（捣烂成浆）加清水2倍并浸泡5小时然后取滤液喷洒果树、花木或蔬菜，都可有效将红蜘蛛杀死。 7.糖醋、烂果诱捕金龟子选用熟烂酸臭的无花果、烂西瓜等，与糖醋液（红糖、醋、水比为1：3：16），一起放入陶钵，支撑分布在果园或菜园中，每2—3天收集钵中的金龟子即可。 8.三合板涂漆聚捕微型害虫在较大的三合板两面涂上橙黄色油漆，干后再涂一层机油、黄油混合油，分布挂在果园或菜园中，蚜虫、白粉虱、美洲斑潜蝇等害虫就会自投罗网。1周后更换涂刷油漆、混合效果更好。