CN103593652A - Cucumber disease identification method based on cucumber leaf symptom image processing - Google Patents

Abstract

A cucumber disease identification method based on cucumber leaf symptom image processing comprises the steps that cucumber disease leaf scab images are segmented; cucumber disease leaf image identification features are extracted, dimensionality reduction is carried out on feature vectors, and at last cucumber disease identification is carried out. The method resolves the problems that the identification rate of cucumber diseases based on leaves is not high and identification effects are instable due to the facts that according to an existing cucumber disease identification method and technique, cucumber disease leaf image components are complex, scabs on disease cucumber leaves are irregular in arrangement and color, and shapes and colors of scabs of leaves of different diseases are not the same, and has the advantages of being high in feature extraction speed and identification rate, stable in identification effect, higher in practicability and the like.

Description

A kind of cucumber disease recognition methods of processing based on cucumber leaves symptom image

Technical field

The present invention relates to image processes and the applied technical field of pattern-recognition in cucumber disease identification, particularly a kind of cucumber disease recognition methods of processing based on cucumber leaves symptom image.

Background technology

Cucumber is distributed widely in a plurality of areas of China and even the world, is one of edible vegetables of multinational resident, and human body is had to a lot of benefits.But cucumber is a kind of cucumber that easily catches an illness, common cucumber disease kind just has kind more than ten.Accurately judgement cucumber disease kind is the prerequisite of cucumber disease control.Traditional cucumber disease detects and substantially relies on agricultural producer's eye estimate to judge, this detection method has a lot of shortcomings, as strong in subjectivity, recognition speed is slow, identification intensity is large, misclassification rate is high, real-time is poor etc., is difficult to meet the demand of cucumber disease real-time monitoring system on a large scale.Cucumber leaves scab and correlated characteristic thereof be judgement cucumber disease kind and the extent of injury thereof important evidence [Tian Youwen, etc. the research of the cucumber in solar-greenhouse disease identification of processing based on image. agricultural research, 2006,2:151-153].From current result of study, adopt computing machine and image processing techniques can be objective, in time, the state of an illness of identifying and diagnosing cucumber exactly, thereby realize the control of cucumber disease and accurately applying pesticides [Cen Zhe is prosperous, Deng. the cucumber anthracnose based on color image color statistical nature and the Study of recognition of brown spot. gardening journal, 2007,34 (6): 1425-1430].But by leaf symptom, detecting cucumber disease is not a nothing the matter.Reason is: the kind of (1) current cucumber disease is a lot, causes disease blade to present diversified symptom; (2) along with the popularization of China's high-quality new breeds of cucumbers and diversification plantation, for creating of more kinds of cucumber diseases suitable condition (as greenhouse gardening etc.), caused serious ascendant trend of being of cucumber disease.These situations have brought challenge also to the cucumber disease detection method research based on cucumber leaves symptom.

Summary of the invention

In order to overcome the shortcoming of above-mentioned prior art, the object of the present invention is to provide a kind of cucumber disease recognition methods of processing based on cucumber leaves symptom image, the advantage such as have that feature extraction speed is fast, discrimination is high, recognition effect is stable and practical.

In order to achieve the above object, the technical scheme that the present invention takes is:

The cucumber disease recognition methods that cucumber leaves symptom image is processed, comprises the following steps:

The first step, cuts apart cucumber disease leaf spot lesion image: first, cucumber disease leaf image is converted to digital picture matrix; Then, utilize the level and smooth cucumber leaves of closure operation image border, fill the breach of leaf spot lesion inside, and the separating part of leaf spot lesion is linked together; Again computing is opened in the scab region obtaining, obtain leaf spot lesion region; Finally, cucumber leaves scab area image after mathematical morphology filter and former cucumber leaves coloured image are carried out to multiplying, obtain cucumber disease leaf spot lesion image, the scab image obtaining is converted to three kinds of primary colours matrix R of red, green, blue, G, B, establishes line number and columns that M, N are respectively matrix R, G, B;

Second step, cucumber disease leaf image recognition feature is extracted: utilize following formula (1) that R, G, B are converted to tone H, brightness S, tri-matrixes of saturation degree I:

$\left\{\begin{array}{c}H=\arccos \left\{\frac{[R-G]+[R-B]/2}{\sqrt{{(R-B)}^2+(R-B)(G-b)}}\right\}/360\\ S=\frac{R+G+B}{3}\left[\min (R,G,B)\right]\\ I=\frac{R+G+B}{3}\end{array}\right.---\left(1\right)$

Utilize following formula (2) by R, G, B, to be obtained the gray matrix Gray of scab:

Gray＝0.299R+0.587G+0.114B (2)

Utilize following formula (3) to calculate two kinds by R, G, B and stimulate color matrices X, Z:

$\left\{\begin{array}{c}X=0.607R+0.174G+0.200B\\ Z=0.066R+1.111B\end{array}\right.---\left(3\right)$

Utilize following formula (4) by R, G, B, to be calculated two kinds of color matrices Cb, Cr of YCbCr color space:

$\left\{\begin{array}{c}\mathrm{Cb}=-0.148R-0.291G+0.439B+128\\ \mathrm{Cr}=0.439R-0.368G-0.071B+128\end{array}\right.---\left(4\right)$

Any one M * N ties up matrix J _ijfirst moment μ ₁, second moment μ ₂with third moment μ ₂be expressed as:

$\left\{\begin{array}{c}{\mathrm{\μ}}_1=\frac{1}{M\·N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{J}_{\mathrm{ij}}\\ {\mathrm{\μ}}_2={\left[\frac{1}{M\·N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{(J_{\mathrm{ij}}-{\mathrm{\μ}}_1)}^2\right]}^{\frac{1}{2}}\\ {\mathrm{\μ}}_3={\left[\frac{1}{M\·N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(J_{\mathrm{ij}}-{\mathrm{\μ}}_1)}^3\right]}^{\frac{1}{3}}\end{array}\right.---\left(5\right)$

Utilize formula (5) to calculate respectively first moment, second moment and the third moment of 11 color component R, G of leaf spot lesion image, B, H, S, I, Gray, X, Z, Cb, Cr, obtain altogether 33 real numbers,

Utilize following formula (6) to calculate average, variance, the degree of bias, peak value, energy, the entropy of the tone matrix H of scab, be expressed as ω _i(i=1,2,3,4,5,6), obtain 6 real numbers altogether:

$\left\{\begin{array}{c}{\mathrm{\ω}}_1=\underset{b=0}{\overset{L-1}{\mathrm{\Σ}}}\mathrm{b\ρ}\left(b\right)\\ {\mathrm{\ω}}_2=\underset{b=0}{\overset{L-1}{\mathrm{\Σ}}}{(b-\mathrm{\μ})}^2\mathrm{\ρ}\left(b\right)\\ {\mathrm{\ω}}_3=\frac{1}{{\mathrm{\σ}}^3}\underset{b=0}{\overset{L-1}{\mathrm{\Σ}}}{(b-\mathrm{\μ})}^3\mathrm{\ρ}\left(b\right)\\ {\mathrm{\ω}}_4=\frac{1}{{\mathrm{\σ}}^4}\underset{b=0}{\overset{L-1}{\mathrm{\Σ}}}{(b-\mathrm{\μ})}^4\mathrm{\ρ}\left(b\right)\\ {\mathrm{\ω}}_5=\underset{b=0}{\overset{L-1}{\mathrm{\Σ}}}{\left[\mathrm{\ρ}\left(b\right)\right]}^2\\ {\mathrm{\ω}}_6=-\underset{b=0}{\overset{L-1}{\mathrm{\Σ}}}\mathrm{\ρ}\left(b\right)\log \left[\mathrm{\ρ}\left(b\right)\right]\end{array}\right.---\left(6\right)$

In formula,

the number that r (b) is b for element value in matrix H,

Utilize following formula (7) to calculate the p+q level square M that scab image converts gray level image Gray Leaf scab region to _pqwith center square μ _pq:

$\left\{\begin{array}{c}{M}_{\mathrm{pq}}=\underset{(x,y)\∈\mathrm{\Δ}}{\mathrm{\Σ}}{x}^p\·y^q\·\mathrm{Grap}(x,y)\\ {\mathrm{\μ}}_{\mathrm{pq}}=\underset{(x,y)\∈\mathrm{\Δ}}{\mathrm{\Σ}}{(x-\stackrel{-}{x})}^p\·{(y-\stackrel{-}{y})}^q\mathrm{Grap}(x,y)\end{array}\right.---\left(7\right)$

In formula, p, q are two positive integers, Δ=(x, y) | Grap (x, y) < 119} is leaf spot lesion region,

μ _pqafter regularization, be expressed as

${\mathrm{\&eta;}}_{\mathrm{pq}}=\frac{{\mathrm{\&mu;}}_{\mathrm{pq}}}{{{\mathrm{\&mu;}}_{00}}^{\mathrm{\&gamma;}}}---\left(8\right)$

In formula, γ=p+q+1 is the regularization factor,

Utilize regularization centre distance η _pq, by following formula (9), calculated 7 of scab region of Gray not bending moments, be expressed as Hu _i(i=1,2,3,4,5,6,7):

$\left\{\begin{array}{c}\mathrm{Hu}1={\mathrm{\&eta;}}_{20}+{\mathrm{\&eta;}}_{02}\\ \mathrm{Hu}2={({\mathrm{\&eta;}}_{20}-{\mathrm{\&eta;}}_{02})}^2+{4\mathrm{\&eta;}}_{11}^2\\ \mathrm{Hu}3={({\mathrm{\&eta;}}_{30}-{3\mathrm{\&eta;}}_{12})}^2+{({\mathrm{\&eta;}}_{03}-{3\mathrm{\&eta;}}_{21})}^2\\ \mathrm{Hu}4={({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2+{\left({\mathrm{\&eta;}}_{21}\right)}^2\\ \mathrm{Hu}5=({\mathrm{\&eta;}}_{30}-{3\mathrm{\&eta;}}_{12})({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-3{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+\\ ({3\mathrm{\&eta;}}_{21}-{\mathrm{\&eta;}}_{03})({\mathrm{\&eta;}}_{21}+{\mathrm{\&eta;}}_{03})[3{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{21})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]\\ \mathrm{Hu}6=({\mathrm{\&eta;}}_{20}-{\mathrm{\&eta;}}_{02})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{21})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+{4\mathrm{\&eta;}}_{11}({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})\\ \mathrm{Hu}7=({3\mathrm{\&eta;}}_{21}-{\mathrm{\&eta;}}_{03})({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-{3({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+\\ ({3\mathrm{\&eta;}}_{12}-{\mathrm{\&eta;}}_{30})({\mathrm{\&eta;}}_{21}+{\mathrm{\&eta;}}_{03})[3{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]\end{array}\right.---\left(9\right)$

By formula (1) above, to formula (9), obtain 46 eigenwerts of each width cucumber disease leaf image, these 46 eigenwerts are rearranged to a proper vector T according to sequencing;

The 3rd step, proper vector is carried out to Dimensionality Reduction: according to the first step above, calculate all cucumber disease leaf image characteristic of correspondence vectors, it is carried out to Dimensionality Reduction,

Be provided with K class n width cucumber disease leaf image { Im ₁, Im ₂..., Im _n, classification numbering is designated as C ₁, C ₂..., C _k, C wherein _iclass leaf image has n _iwidth, { Im ₁, Im ₂..., Im _nvectorial set of characteristic of correspondence be designated as { T ₁, T ₂..., T _n,

Calculate n proper vector { T ₁, T ₂..., T _nmean value T,

$\stackrel{-}{T}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{T}_i---\left(10\right)$

Calculate C _iclass n _ithe mean value of individual proper vector

$\stackrel{\&OverBar;}{T_{C_i}}=\frac{1}{n_i}\underset{X_j\&Element;C_i}{\mathrm{\&Sigma;}}{T}_j---\left(11\right)$

Between class scatter matrix S _bwith Scatter Matrix S in class _wbe defined as respectively

$\left\{\begin{array}{c}{S}_B=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}{N}_i\left|\right|\stackrel{\&OverBar;}{T_{C_i}}-\stackrel{\&OverBar;}{T}\left|\right|\\ S_W=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{X_j\&Element;C_i}{\mathrm{\&Sigma;}}\left|\right|T_j-\stackrel{\&OverBar;}{T_{C_i}}\left|\right|\end{array}\right.---\left(12\right)$

In formula, || || represent compute euclidian distances,

By S _band S _wsetting up objective optimization function is

Solve formula (13), calculate (S _b-S _w) yojan proper vector a corresponding to maximum d eigenwert of a=λ a ₁, a ₂..., a _d, wherein λ is eigenwert, the dimension that d is low dimensional feature vector, by a ₁, a ₂..., a _dform a projection matrix A=[a ₁, a ₂..., a _d],

Set of eigenvectors { T by A to n leaf spot lesion image ₁, T ₂..., T _ncarry out Dimensionality Reduction and obtain low-dimensional recognition feature vector collection { Y ₁, Y ₂..., Y _n, Y wherein _i=A ^{transposition}t _i(i=1,2 ..., n),

By following formula (14), the proper vector T of any width leaf image is carried out to Dimensionality Reduction, obtains low-dimensional recognition feature vector Y,

Y=A ^{transposition}t (14)

The low-dimensional recognition feature vector of the cucumber disease leaf image for training classifier obtaining is input to recognition template database, and the low-dimensional recognition feature vector of each leaf image is corresponding with the cucumber scab information that pre-deposits system;

The 4th step, cucumber disease identification: the data in template database are input to nearest neighbor classifier, train this sorter, then the low-dimensional recognition feature vector of cucumber leaves image to be tested is input to nearest neighbor classifier, finds out in 1 nearest neighbor point the same classification classification of maximum classifications as disease blade to be tested of counting.

Beneficial effect of the present invention is:

The object of the invention is to overcome existing cucumber disease recognition methods and technology arranges random and shade differs, also equal reason mutually not of the shape of different sick leaf spot lesions of planting and color because of the scab on cucumber disease leaf image complicated component, the sick leaf of cucumber, the problem such as make that the discrimination of the cucumber disease based on blade is not high and recognition effect is unstable, the advantage such as have that feature extraction speed is fast, discrimination is high, recognition effect is stable and practical.At present, patent of the present invention realizes on common PC computing machine, and operating system is not required.Be mainly used in computer based cucumber disease classification recognition system, can be for the cucumber disease identification in actual production.

Embodiment

Below in conjunction with example, the present invention is described in detail.

The cucumber disease recognition methods that cucumber leaves symptom image is processed, comprises the following steps:

The first step, cuts apart cucumber disease leaf spot lesion image: first, utilize the function ' imread ' in Matlab software to convert all cucumber disease leaf images to digital picture matrix; Then, design the square structure element of 3 * 3, utilize closure operation smoothing blade image boundary and fill the breach of leaf spot lesion inside, then the separating part of leaf spot lesion is linked together; Again computing is opened to eliminate scab noise around in the scab region obtaining, obtain leaf spot lesion region, finally, cucumber leaves scab area image after mathematical morphology filter and former cucumber leaves coloured image are carried out to multiplying, obtain cucumber disease leaf spot lesion image, the scab image obtaining is converted to three kinds of primary colours matrix R of red, green, blue, G, B, establishes line number and columns that M, N are respectively matrix R, G, B;

Second step, cucumber disease leaf image recognition feature is extracted: first, just R, G, B are converted to three color matrixes (tone H, brightness S, saturation degree I), gray matrix Gray, two stimulates color matrices (X and Z) and two color matrices (Cb and Cr).Describe in detail as follows:

R, G, B are converted to tone H, brightness S, tri-matrixes of saturation degree I, and conversion formula is:

$\left\{\begin{array}{c}H=\arccos \left\{\frac{[R-G]+[R-B]/2}{\sqrt{{(R-B)}^2+(R-B)(G-b)}}\right\}/360\\ S=\frac{R+G+B}{3}\left[\min (R,G,B)\right]\\ I=\frac{R+G+B}{3}\end{array}\right.---\left(1\right)$

R, G, B are converted to gray matrix Gray, and conversion formula is as follows:

Gray＝0.299R+0.587G+0.114B (2)

By R, G, B, calculating two kinds stimulates color matrices X, Z, and computing formula is as follows:

$\left\{\begin{array}{c}X=0.607R+0.174G+0.200B\\ Z=0.066R+1.111B\end{array}\right.---\left(3\right)$

Two kinds of color matrices Cb, Cr by R, G, B, being calculated YCbCr color space, computing formula is:

$\left\{\begin{array}{c}\mathrm{Cb}=-0.148R-0.291G+0.439B+128\\ \mathrm{Cr}=0.439R-0.368G-0.071B+128\end{array}\right.---\left(4\right)$

Calculate respectively matrix R, G, B, H, S, I, Gray, X, Z, the Cb of 11 M that 11 colors of scab image obtained above are corresponding * N dimension, first moment, second moment and the third moment of Cr, computing formula is:

$\left\{\begin{array}{c}{\mathrm{\&mu;}}_1=\frac{1}{M\&CenterDot;N}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{J}_{\mathrm{ij}}\\ {\mathrm{\&mu;}}_2={\left[\frac{1}{M\&CenterDot;N}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{(J_{\mathrm{ij}}-{\mathrm{\&mu;}}_1)}^2\right]}^{\frac{1}{2}}\\ {\mathrm{\&mu;}}_3={\left[\frac{1}{M\&CenterDot;N}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{(J_{\mathrm{ij}}-{\mathrm{\&mu;}}_1)}^3\right]}^{\frac{1}{3}}\end{array}\right.---\left(5\right)$

In formula, J _ijfor any one matrix, M, N are respectively matrix J _ijline number and columns, J _ijfirst moment be μ ₁, second moment is μ ₂with third moment be μ ₂.

The average, variance, the degree of bias, peak value, energy, the entropy that calculate tone matrix H, be expressed as ω _i(i=1,2,3,4,5,6), computing formula is:

$\left\{\begin{array}{c}{\mathrm{\&omega;}}_1=\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\mathrm{b\&rho;}\left(b\right)\\ {\mathrm{\&omega;}}_2=\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(b-\mathrm{\&mu;})}^2\mathrm{\&rho;}\left(b\right)\\ {\mathrm{\&omega;}}_3=\frac{1}{{\mathrm{\&sigma;}}^3}\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(b-\mathrm{\&mu;})}^3\mathrm{\&rho;}\left(b\right)\\ {\mathrm{\&omega;}}_4=\frac{1}{{\mathrm{\&sigma;}}^4}\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(b-\mathrm{\&mu;})}^4\mathrm{\&rho;}\left(b\right)\\ {\mathrm{\&omega;}}_5=\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{\left[\mathrm{\&rho;}\left(b\right)\right]}^2\\ {\mathrm{\&omega;}}_6=-\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\mathrm{\&rho;}\left(b\right)\log \left[\mathrm{\&rho;}\left(b\right)\right]\end{array}\right.---\left(6\right)$

In formula,

the number that r (b) is b for element value in matrix H.

Calculate the p+q level square M in the gray level image Gray Leaf scab region after the conversion of scab image _pqwith center square μ _pq:

$\left\{\begin{array}{c}{M}_{\mathrm{pq}}=\underset{(x,y)\&Element;\mathrm{\&Delta;}}{\mathrm{\&Sigma;}}{x}^p\&CenterDot;y^q\&CenterDot;\mathrm{Grap}(x,y)\\ {\mathrm{\&mu;}}_{\mathrm{pq}}=\underset{(x,y)\&Element;\mathrm{\&Delta;}}{\mathrm{\&Sigma;}}{(x-\stackrel{-}{x})}^p\&CenterDot;{(y-\stackrel{-}{y})}^q\mathrm{Grap}(x,y)\end{array}\right.---\left(7\right)$

In formula,

p, q are two positive integers, and Δ is leaf spot lesion region, Δ=(x, y) | Grap (x, y) < 119}.

To μ _pqcarry out obtaining after regularization

${\mathrm{\&eta;}}_{\mathrm{pq}}=\frac{{\mathrm{\&mu;}}_{\mathrm{pq}}}{{{\mathrm{\&mu;}}_{00}}^{\mathrm{\&gamma;}}}---\left(8\right)$

In formula, γ=p+q+1 is the regularization factor.Utilize η _pqcalculate 7 of scab region C of Gray not bending moment Hu _i(i=1,2,3,4,5,6,7), formula is as follows:

$\left\{\begin{array}{c}\mathrm{Hu}1={\mathrm{\&eta;}}_{20}+{\mathrm{\&eta;}}_{02}\\ \mathrm{Hu}2={({\mathrm{\&eta;}}_{20}-{\mathrm{\&eta;}}_{02})}^2+{4\mathrm{\&eta;}}_{11}^2\\ \mathrm{Hu}3={({\mathrm{\&eta;}}_{30}-{3\mathrm{\&eta;}}_{12})}^2+{({\mathrm{\&eta;}}_{03}-{3\mathrm{\&eta;}}_{21})}^2\\ \mathrm{Hu}4={({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2+{\left({\mathrm{\&eta;}}_{21}\right)}^2\\ \mathrm{Hu}5=({\mathrm{\&eta;}}_{30}-{3\mathrm{\&eta;}}_{12})({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-3{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+\\ ({3\mathrm{\&eta;}}_{21}-{\mathrm{\&eta;}}_{03})({\mathrm{\&eta;}}_{21}+{\mathrm{\&eta;}}_{03})[3{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{21})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]\\ \mathrm{Hu}6=({\mathrm{\&eta;}}_{20}-{\mathrm{\&eta;}}_{02})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{21})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+{4\mathrm{\&eta;}}_{11}({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})\\ \mathrm{Hu}7=({3\mathrm{\&eta;}}_{21}-{\mathrm{\&eta;}}_{03})({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-{3({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+\\ ({3\mathrm{\&eta;}}_{12}-{\mathrm{\&eta;}}_{30})({\mathrm{\&eta;}}_{21}+{\mathrm{\&eta;}}_{03})[3{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]\end{array}\right.---\left(9\right)$

46 eigenwerts of each width cucumber leaves image are rearranged to a proper vector T according to sequencing, be that T is comprised of three parts: first moment, second moment and the third moment of 11 components R of (1) scab image, G, B, H, S, I, Gray, X, Z, Cb, Cr, totally 33 data; (2) average, variance, the degree of bias, peak value, energy, the entropy of the tone matrix H of scab image, altogether to 6 data; (3) 7 of the scab region of the gray level image Gray of scab image bending moments not.

The 3rd step, carries out Dimensionality Reduction to proper vector: establish total K class n width cucumber disease leaf image { Im ₁, Im ₂..., Im _n, classification numbering is designated as C ₁, C ₂..., C _k, C _iclass has n _iindividual sample image,

{ Im ₁, Im ₂..., Im _nvectorial set of characteristic of correspondence be designated as { T ₁, T ₂..., T _n.

Calculate the mean value of n proper vector

$\stackrel{-}{T}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{T}_i---\left(10\right)$

Calculate C _iclass N _imean value in the class of individual proper vector

$\stackrel{\&OverBar;}{T_{C_i}}=\frac{1}{n_i}\underset{X_j\&Element;C_i}{\mathrm{\&Sigma;}}{T}_j---\left(11\right)$

Utilize Euclidean distance, by

with

calculating scab image is between class scatter matrix S _bwith Scatter Matrix S in class _w, computing formula is

$\left\{\begin{array}{c}{S}_B=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}{N}_i\left|\right|\stackrel{\&OverBar;}{T_{C_i}}-\stackrel{\&OverBar;}{T}\left|\right|\\ S_W=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{X_j\&Element;C_i}{\mathrm{\&Sigma;}}\left|\right|T_j-\stackrel{\&OverBar;}{T_{C_i}}\left|\right|\end{array}\right.---\left(12\right)$

By S _band S _wset up objective optimization function

In formula, A is projection matrix to be asked,

Calculate (S _b-S _w) yojan proper vector a corresponding to maximum d eigenwert of a=λ a ₁, a ₂..., a _d, wherein λ is eigenwert, the dimension that d is low dimensional feature vector.By a ₁, a ₂..., a _dform a projection matrix A=[a ₁, a ₂..., a _d].Set of eigenvectors { T by A to n leaf spot lesion image ₁, T ₂..., T _ncarry out Dimensionality Reduction and obtain low-dimensional recognition feature vector collection { Y ₁, Y ₂..., Y _n, i.e. Y _i=A ^{transposition}t _i(i=1,2 ..., n).For any one cucumber leaves image to be identified, calculate its 46 features, form a proper vector T _new, then calculate its low-dimensional recognition feature vector

Y _new=A ^{transposition}t _new(14)

By { the Y obtaining ₁, Y ₂..., Y _nbeing input to recognition template database, the low-dimensional recognition feature vector of each leaf image is corresponding with the cucumber scab information that pre-deposits system.

The 4th step, cucumber disease identification: by the data { Y in template database ₁, Y ₂..., Y _nbe input to nearest neighbor classifier, train this sorter, then by the low-dimensional recognition feature vector Y of cucumber leaves image to be tested _newbe input to nearest neighbor classifier, find out in 1 nearest neighbor point the same classification classification of maximum classifications as disease blade to be tested of counting.The assorting process of sorter is divided into three steps: first, produce training sample set, make the recognition feature template data of training set be divided into discrete type numerical value class according to existing criteria for classification; Secondly, with the basis that is categorized as of feature templates data, the low-dimensional recognition feature vector of test pattern is found to 1 arest neighbors, adopt Euclidean distance as the distinguishing rule of the similarity degree between sample, similarity is large is arest neighbors; Finally, cucumber disease classification to be tested is finally output as that maximum class of number in arest neighbors class.

Claims (1)

1. a cucumber disease recognition methods of processing based on cucumber leaves symptom image, is characterized in that, comprises the following steps:

The first step, cuts apart cucumber disease leaf spot lesion image: first, cucumber disease leaf image is converted to digital picture matrix; Then, utilize the level and smooth cucumber leaves of closure operation image border, fill the breach of leaf spot lesion inside, and the separating part of leaf spot lesion is linked together; Again computing is opened in the scab region obtaining, obtain leaf spot lesion region; Finally, just the cucumber leaves scab area image after mathematical morphology filter and former cucumber leaves coloured image carry out multiplying, obtain cucumber disease leaf spot lesion image, the scab image obtaining is converted to three kinds of primary colours matrix R of red, green, blue, G, B, establishes line number and columns that M, N are respectively matrix R, G, B;

Second step, cucumber disease leaf image recognition feature is extracted: utilize following formula (1) that R, G, B are converted to tone H, brightness S, tri-matrixes of saturation degree I:

$\left\{\begin{array}{c}H=\arccos \left\{\frac{[R-G]+[R-B]/2}{\sqrt{{(R-B)}^2+(R-B)(G-b)}}\right\}/360\\ S=\frac{R+G+B}{3}\left[\min (R,G,B)\right]\\ I=\frac{R+G+B}{3}\end{array}\right.---\left(1\right)$

Utilize following formula (2) by R, G, B, to be obtained the gray matrix Gray of scab:

Gray＝0.299R+0.587G+0.114B (2)

Utilize following formula (3) to calculate two kinds by R, G, B and stimulate color matrices X, Z:

$\left\{\begin{array}{c}X=0.607R+0.174G+0.200B\\ Z=0.066R+1.111B\end{array}\right.---\left(3\right)$

Utilize following formula (4) by R, G, B, to be calculated two kinds of color matrices Cb, Cr of YCbCr color space:

$\left\{\begin{array}{c}\mathrm{Cb}=-0.148R-0.291G+0.439B+128\\ \mathrm{Cr}=0.439R-0.368G-0.071B+128\end{array}\right.---\left(4\right)$

Any one M * N ties up matrix J _ijfirst moment μ ₁, second moment μ ₂with third moment μ ₂be expressed as:

$\left\{\begin{array}{c}{\mathrm{\&mu;}}_1=\frac{1}{M\&CenterDot;N}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{J}_{\mathrm{ij}}\\ {\mathrm{\&mu;}}_2={\left[\frac{1}{M\&CenterDot;N}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{(J_{\mathrm{ij}}-{\mathrm{\&mu;}}_1)}^2\right]}^{\frac{1}{2}}\\ {\mathrm{\&mu;}}_3={\left[\frac{1}{M\&CenterDot;N}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{(J_{\mathrm{ij}}-{\mathrm{\&mu;}}_1)}^3\right]}^{\frac{1}{3}}\end{array}\right.---\left(5\right)$

Utilize formula (5) to calculate respectively first moment, second moment and the third moment of 11 color component R, G of leaf spot lesion image, B, H, S, I, Gray, X, Z, Cb, Cr, obtain altogether 33 real numbers,

Utilize following formula (6) to calculate average, variance, the degree of bias, peak value, energy, the entropy of the tone matrix H of scab, be expressed as ω _i(i=1,2,3,4,5,6), obtain 6 real numbers altogether:

$\left\{\begin{array}{c}{\mathrm{\&omega;}}_1=\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\mathrm{b\&rho;}\left(b\right)\\ {\mathrm{\&omega;}}_2=\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(b-\mathrm{\&mu;})}^2\mathrm{\&rho;}\left(b\right)\\ {\mathrm{\&omega;}}_3=\frac{1}{{\mathrm{\&sigma;}}^3}\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(b-\mathrm{\&mu;})}^3\mathrm{\&rho;}\left(b\right)\\ {\mathrm{\&omega;}}_4=\frac{1}{{\mathrm{\&sigma;}}^4}\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(b-\mathrm{\&mu;})}^4\mathrm{\&rho;}\left(b\right)\\ {\mathrm{\&omega;}}_5=\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{\left[\mathrm{\&rho;}\left(b\right)\right]}^2\\ {\mathrm{\&omega;}}_6=-\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\mathrm{\&rho;}\left(b\right)\log \left[\mathrm{\&rho;}\left(b\right)\right]\end{array}\right.---\left(6\right)$

In formula,

the number that r (b) is b for element value in matrix H,

Utilize following formula (7) to calculate the p+q level square M that scab image converts gray level image Gray Leaf scab region to _pqwith center square μ _pq:

$\left\{\begin{array}{c}{M}_{\mathrm{pq}}=\underset{(x,y)\&Element;\mathrm{\&Delta;}}{\mathrm{\&Sigma;}}{x}^p\&CenterDot;y^q\&CenterDot;\mathrm{Grap}(x,y)\\ {\mathrm{\&mu;}}_{\mathrm{pq}}=\underset{(x,y)\&Element;\mathrm{\&Delta;}}{\mathrm{\&Sigma;}}{(x-\stackrel{-}{x})}^p\&CenterDot;{(y-\stackrel{-}{y})}^q\mathrm{Grap}(x,y)\end{array}\right.---\left(7\right)$ In formula,

p, q are two positive integers, Δ=(x, y) | Grap (x, y) < 119} is leaf spot lesion region,

μ _pqafter regularization, be expressed as

${\mathrm{\&eta;}}_{\mathrm{pq}}=\frac{{\mathrm{\&mu;}}_{\mathrm{pq}}}{{{\mathrm{\&mu;}}_{00}}^{\mathrm{\&gamma;}}}---\left(8\right)$ In formula, γ=p+q+1 is the regularization factor,

Utilize regularization centre distance η _pq, by following formula (9), calculated 7 of scab region of Gray not bending moments, be expressed as Hu _i(i=1,2,3,4,5,6,7):

$\left\{\begin{array}{c}\mathrm{Hu}1={\mathrm{\&eta;}}_{20}+{\mathrm{\&eta;}}_{02}\\ \mathrm{Hu}2={({\mathrm{\&eta;}}_{20}-{\mathrm{\&eta;}}_{02})}^2+{4\mathrm{\&eta;}}_{11}^2\\ \mathrm{Hu}3={({\mathrm{\&eta;}}_{30}-{3\mathrm{\&eta;}}_{12})}^2+{({\mathrm{\&eta;}}_{03}-{3\mathrm{\&eta;}}_{21})}^2\\ \mathrm{Hu}4={({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2+{\left({\mathrm{\&eta;}}_{21}\right)}^2\\ \mathrm{Hu}5=({\mathrm{\&eta;}}_{30}-{3\mathrm{\&eta;}}_{12})({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-3{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+\\ ({3\mathrm{\&eta;}}_{21}-{\mathrm{\&eta;}}_{03})({\mathrm{\&eta;}}_{21}+{\mathrm{\&eta;}}_{03})[3{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{21})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]\\ \mathrm{Hu}6=({\mathrm{\&eta;}}_{20}-{\mathrm{\&eta;}}_{02})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{21})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+{4\mathrm{\&eta;}}_{11}({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})\\ \mathrm{Hu}7=({3\mathrm{\&eta;}}_{21}-{\mathrm{\&eta;}}_{03})({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})[{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-{3({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]+\\ ({3\mathrm{\&eta;}}_{12}-{\mathrm{\&eta;}}_{30})({\mathrm{\&eta;}}_{21}+{\mathrm{\&eta;}}_{03})[3{({\mathrm{\&eta;}}_{30}+{\mathrm{\&eta;}}_{12})}^2-{({\mathrm{\&eta;}}_{03}+{\mathrm{\&eta;}}_{21})}^2]\end{array}\right.---\left(9\right)$

46 eigenwerts that can be accessed each width cucumber disease leaf image by formula (1) to (9) above, rearrange a proper vector T these 46 eigenwerts according to sequencing;

The 3rd step, proper vector is carried out to Dimensionality Reduction: according to the first step above, calculate all cucumber disease leaf image characteristic of correspondence vectors, it is carried out to Dimensionality Reduction,

Be provided with K class n width cucumber disease leaf image { Im ₁, Im ₂..., Im _n, classification numbering is designated as C ₁, C ₂..., C _k, C wherein _iclass leaf image has n _iwidth, { Im ₁, Im ₂..., Im _nvectorial set of characteristic of correspondence be designated as { T ₁, T ₂..., T _n,

N proper vector { T ₁, T ₂..., T _nmean value be

$\stackrel{-}{T}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{T}_i---\left(10\right)$

C _iclass n _ithe mean value of individual proper vector is

$\stackrel{\&OverBar;}{T_{C_i}}=\frac{1}{n_i}\underset{X_j\&Element;C_i}{\mathrm{\&Sigma;}}{T}_j---\left(11\right)$

Between class scatter matrix S _bwith Scatter Matrix S in class _wbe defined as respectively

$\left\{\begin{array}{c}{S}_B=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}{N}_i\left|\right|\stackrel{\&OverBar;}{T_{C_i}}-\stackrel{\&OverBar;}{T}\left|\right|\\ S_W=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{X_j\&Element;C_i}{\mathrm{\&Sigma;}}\left|\right|T_j-\stackrel{\&OverBar;}{T_{C_i}}\left|\right|\end{array}\right.---\left(12\right)$ In formula, || || represent compute euclidian distances,

By S _band S _wset up objective optimization function

Solve formula (13), calculate (S _b-S _w) yojan proper vector a corresponding to maximum d eigenwert of a=λ a ₁, a ₂..., a _d, wherein λ is eigenwert, the dimension that d is low dimensional feature vector, by a ₁, a ₂..., a _dform a projection matrix A=[a ₁, a ₂..., a _d],

Set of eigenvectors { T by A to n leaf spot lesion image ₁, T ₂..., T _ncarry out Dimensionality Reduction and obtain low-dimensional recognition feature vector collection { Y ₁, Y ₂..., Y _n, Y wherein _i=A ^{transposition}t _i(i=1,2 ..., n),

By following formula (14), the proper vector T of any width leaf image is carried out to Dimensionality Reduction, obtains low-dimensional recognition feature vector Y,

Y=A ^{transposition}t (14)

The low-dimensional recognition feature vector of the cucumber disease leaf image for training classifier obtaining is input to recognition template database, and the low-dimensional recognition feature vector of each leaf image is corresponding with the cucumber scab information that pre-deposits system;

The 4th step, cucumber disease identification: the data in template database are input to nearest neighbor classifier, train this sorter, then the low-dimensional recognition feature vector of cucumber leaves image to be tested is input to nearest neighbor classifier, finds out in 1 nearest neighbor point the same classification classification of maximum classifications as disease blade to be tested of counting.