CN103336974A - Flower and plant category recognition method based on local constraint sparse characterization - Google Patents

Abstract

The invention discloses a flower and plant category recognition method based on local constraint sparse characterization, which comprises the following steps: firstly a flower and plant image database is collected, a flower and plant popular science knowledge base is built, and then various image characteristics of flower and plant pictures are exteracted; then the linear representation of training data characetristics and test data characteristics is built by utilizing the sparse coding theory; when the linear representation is built, the minimum error of the linear representation of the test image characteristics and the training image characteristics is taken into account, the weight constraint of similar structures of local characteristics of the test data characteristics and the traning data characteristics is increased at the same time, and the efficient solving process is performed by utilizing the statistical gradient descent method of kernel function extension, so that flower and plant category recognition and studying processes are completed; finally, the to-be-recognized flower and plant image extraction characteristics are substituted into the flower and plant category judgment formula so as to obtain the recognition result, and corresponding explanatory notes are called out of the flower and plant popular science knowledge base. The method has the advantage of high recognition performance.

Description

A kind of flowers classification discrimination method based on the sparse sign of local restriction

Technical field

The present invention relates to pattern-recognition and field of artificial intelligence, particularly a kind of flowers classification discrimination method based on the sparse sign of local restriction.

Background technology

The identification of flowers classification namely refers to utilize computing machine that the image information of flowers is carried out feature extraction, computing machine according to people's understanding and the mode of thinking in addition kind sort out and understand, and then can be the flowers knowledge that the user provides some science popularization, it belongs to the category of computing machine automatic object identification, patent is fewer aspect the identification of flowers classification at present, in academia a small amount of paper publishing is arranged, as paper (Yuning CHAI, Victor LEMPITSKY, Andrew ZISSERMAN.BiCoS:A Bi-level Co-Segmentation Method for Image Classification.ICCV, 2011), it utilizes image partition method that image is divided into prospect and background, extract color distribution and the super Pixel Information of image then, and utilize the effective algorithm of deducing to identify.Paper (Nilsback, M-E.and Zisserman, A.Automated flower classification over a large number of classes, Proceedings of the Indian Conference on Computer Vision, Graphics and Image Processing2008), it extracts the color histogram of image, SIFT feature, after three kinds of features of HOG feature, carry out the classification of flowers classification with svm classifier.Paper (Nilsback, M-E.and Zisserman, A, A Visual Vocabulary for Flower Classification, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition2006), it utilizes the BOW feature to carry out the classification of flower chart picture.

Summary of the invention

Above-mentioned shortcoming and deficiency in order to overcome prior art the object of the present invention is to provide a kind of flowers classification discrimination method based on the sparse sign of local restriction, the recognition performance height.

Purpose of the present invention is achieved through the following technical solutions: a kind of flowers classification discrimination method based on the sparse sign of local restriction may further comprise the steps:

(1) collects the flowers image data base and set up flowers popular science knowledge storehouse: the title of utilizing S flowers classification commonly used of existing plants and flowers wikipedia definition, utilize network search engines to search for flowers character introduction and the picture of each flowers classification correspondence, picture is formed flower chart as database; Explanatory note is included into flowers popular science knowledge storehouse;

(2) all pictures in the flowers image data base that step (1) is obtained carry out feature extraction, and what every pictures extracted is characterized as m;

(3) flowers classification identification learning process:

(3-1) from the flowers image data base, select the p pictures, with its feature as test data set Y={y ^k, wherein k=1...m represents m the feature that every pictures extracts respectively, and p＜N, N are the picture number in the flowers image data base; The feature of remaining N-p pictures is as training dataset in the flowers image data base

J=1 wherein, 2 ... S, the classification of expression flowers;

(3-2) with test data feature y ^kUse the training data feature

Linear expression is:

Wherein

Be the weight coefficient of training dataset characterization test data set, it is worth greater than 0;

(3-3) for training dataset and test data set linear expression error minimum, increase the weight constraints of the locality characteristic similarity structure of test data feature and training data feature simultaneously, the optimization learning criterion below setting up:

$\underset{W_j^k}{\min }\frac{1}{2}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left|\right|y^k-\underset{j=1}{\overset{S}{\mathrm{\Σ}}}{X}_j^k{w}_j^k\left|\right|}_2^2+\mathrm{\λ}\underset{j=1}{\overset{S}{\mathrm{\Σ}}}{\left|\right|D_j^k\mathrm{\Θ}{w}_j^k\left|\right|}_2---\left(1\right)$

Wherein Θ represents vector dot; λ is the bound term weight, is the constant variables between balance linear expression error and the weight coefficient;

Be test data feature y ^kWith the training data feature

Euclidean distance;

(3-4) adopt the statistical gradient descent method right Carry out iteration and upgrade, renewal equation is:

$\left\{\begin{array}{c}{w}_j^{k,t+1}=w_j^{k,t}-{\mathrm{\η}\▿}_w\\ {\▿}_w=-X_j^k{y}^k+X_j^k{X}_j^k{w}_j^t+\mathrm{\λ}{\left|\right|y^k-X_j^k\left|\right|}_2^2\·w_j^{k,t}\end{array}\right.----\left(2\right)$

Wherein, t is the iterations of SGD iterative process, and η is the learning rate of statistical gradient decline iterative process;

(3-5) utilize nonlinear function φ that the feature of training and testing data is carried out Nonlinear Mapping to the reproducing kernel Hilbert space (reproducing kernel hilbert space is called for short RKHS) of higher-dimension, i.e. φ (x _i) ^Tφ (x _j)=g (x _i, x _j), g (x wherein _ix _j) be χ ²Kernel function, x _iAnd x _jIt is data characteristics; Formula (2) is transformed into:

$\left\{\begin{array}{c}{w}_j^{k,t+1}=w_j^{k,t}-{\mathrm{\η}\▿}_w\\ {\▿}_w={-h}^k+G^k{w}_j^{k,t}+\mathrm{\λ}(P^k-{2h}^h+G^k)\·w_j^{k,t}\end{array}\right.---\left(3\right)$

Wherein It is the training data feature

With test data feature y ^kThe dot product kernel function,

It is the training data feature

With the dot product kernel function of self, P ^k=φ (y ^k) φ (y ^k) be test data feature y ^kWith the dot product kernel function of self; Through the repeatedly iteration of formula (3), obtain optimum sign weight coefficient

(4) identification flowers: the user takes the image of flowers to be identified, and flower chart picture to be identified is extracted characteristic Z ^k, wherein k=1...m represents m the feature that every pictures extracts respectively;

According to Z ^k, by flowers classification judgement formula identification flowers classification, and from flowers popular science knowledge storehouse, access this flowers classification corresponding character explanation;

Wherein flowers classification judgement formula is:

J wherein ^*Expression utilizes linear expression and the minimum error values of test data of the training data of certain j classification, selects by minimum, and j is and identifies the flowers classification that obtains.

In the step (3-4), η reduces along with the increase of iterations t,

The described χ of step (3-5) ²The expression formula of kernel function is exp (χ ²(x, x)/μ), χ wherein ²Be card side (Chi-squared) distance of symmetrical expression, μ is the χ of current training dataset ²The average of distance.

The described title of utilizing S flowers classification commonly used of existing plants and flowers wikipedia definition of step (1), utilize network search engines to search for flowers character introduction and the picture of each flowers classification correspondence, be specially: the title of utilizing 313 herbage flower classifications commonly used of existing plants and flowers wikipedia definition, utilize network search engines to search for flowers character introduction and the picture of each flowers classification correspondence, each flowers classification is downloaded 100 pictures.

The feature that every pictures extracts comprises color histogram, SIFT feature (Scale-invariant feature transform, yardstick invariant features converting characteristic), HOG feature (Histogram of oriented gradients, the direction gradient histogram feature), BOW feature (Bag of word, dictionary bag feature), SSIM feature (Structural Similarity, the structural similarity feature), GB feature (Structural Similarity, structural similarity feature).

Compared with prior art, the present invention has the following advantages and beneficial effect:

(1) in the flowers classification identification learning process, the optimization learning criterion has not only been considered current linear expression error minimum principle, but also with image based on the similar features structural modeling of the part bound term to learning criterion, thereby image is when linear expression, system preferably selects the most similar characteristics of image to carry out linear expression, when guaranteeing the sparse property of linear coefficient again, improved the identification performance of system again.

(2) framed structure of the present invention can seamlessly incorporate the more images feature, and for follow-up system upgrading facilitates, and the characteristics of image that identification arranged also can further improve the identification performance of system more.

(3) identification method of the present invention can be applied in the actual flowers science popularization system well, because identification modeling method recognition performance height of the present invention, thereby robustness and the stability of assurance flowers science popularization system.

Description of drawings

Fig. 1 is the process flow diagram based on the flowers classification discrimination method of the sparse sign of local restriction of present embodiment.

Fig. 2 is the main process flow diagram of statistical gradient descending method of present embodiment.

Embodiment

Below in conjunction with embodiment, the present invention is described in further detail, but embodiments of the present invention are not limited thereto.

Embodiment

As shown in Figure 1, the flowers classification discrimination method based on the sparse sign of local restriction of present embodiment may further comprise the steps:

(1) collects the flowers image data base and set up flowers popular science knowledge storehouse: the title of utilizing 313 herbage flower classifications commonly used of existing plants and flowers wikipedia definition, utilize network search engines such as google and Baidu to search for flowers character introduction and the picture of each flowers classification correspondence, each flowers classification is downloaded 100 pictures.Picture is formed flower chart as database; Explanatory note is included into flowers popular science knowledge storehouse.

(2) all pictures in the flowers image data base that step (1) is obtained carry out feature extraction, what every pictures extracted is characterized as 6, comprise color histogram, the SIFT feature (realizes details reference Lowe, D.G., Distinctive Image Features from Scale-Invariant Keypoints, International Journal of Computer Vision, 60,2, pp.91-110,2004), the HOG feature (realizes details reference N.Dalal and B.Triggs.Histogram of oriented gradients for human detection.In Computer Vision and Pattern Recognition, pp.887 – 893.IEEE, 2005), the BOW feature (realizes details reference S.Lazebnik, C.Schmid, and J.Ponce.Beyond bag of features:Spatial pyramid matching for recognizing natural scene categories.In Computer Vision and Pattern Recognition, pp.2169 – 2178.IEEE, 2006), the SSIM feature (realizes details reference E.Shechtman and M.Irani., Matching local self-similarities across images and videos.In Proc.CVPR, 2007), the GB feature (realizes details reference A.C.Berg, T.L.Berg, and J.Malik., Shape matching and object recognition using low distortion correspondences.In Proc.CVPR, 2005).

(3) flowers classification identification learning process:

(3-1) from the flowers image data base, select 30 pictures, with its feature as test data set Y={y ^k, wherein k=1...6 distinguishes 6 kinds of features of representative image: color histogram, SIFT feature, HOG feature, BOW feature, SSIM feature and GB feature; The feature of the 31270 remaining pictures in the flowers image data base is as training dataset

J=1...313 wherein, the classification of expression flowers;

(3-2) with test data feature y ^kUse the training data feature Linear expression is:

Wherein Be the weight coefficient of training dataset characterization test data set, it is worth greater than 0;

(3-3) for training dataset and test data set linear expression error minimum, increase the weight constraints of the locality characteristic similarity structure of test data feature and training data feature simultaneously, the optimization learning criterion below setting up:

$\underset{W_j^k}{\min }\frac{1}{2}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left|\right|y^k-\underset{j=1}{\overset{S}{\mathrm{\Σ}}}{X}_j^k{w}_j^k\left|\right|}_2^2+\mathrm{\λ}\underset{j=1}{\overset{S}{\mathrm{\Σ}}}{\left|\right|D_j^k\mathrm{\Θ}{w}_j^k\left|\right|}_2---\left(1\right)$

Wherein Θ represents vector dot; λ is the bound term weight, is that constant variables between balance linear expression error and the weight coefficient (is set at 0 in the present embodiment.01);

Expression test data feature y ^kThe training data feature similar with it Locality describe, concrete definition is

Be test data feature y ^kWith the training data feature

Euclidean distance; Under the optimization learning criterion, can guarantee that similar test feature can select similar training characteristics to carry out linear expression; In order to make that final linear expression coefficient can be sparse, generally can be with weight coefficient wherein

Be lower than certain threshold value (present embodiment 0.005) and be made as 0.

(3-4) adopt statistical gradient descent method (Stochastic gradient descent Method is called for short SGD) right Carry out iteration and upgrade, renewal equation is:

$\left\{\begin{array}{c}{w}_j^{k,t+1}=w_j^{k,t}-{\mathrm{\η}\▿}_w\\ {\▿}_w=-X_j^k{y}^k+X_j^k{X}_j^k{w}_j^t+\mathrm{\λ}{\left|\right|y^k-X_j^k\left|\right|}_2^2\·w_j^{k,t}\end{array}\right.----\left(2\right)$

Wherein, t is the iterations of SGD iterative process, and η is the learning rate of statistical gradient decline iterative process; In order to guarantee that learning criterion can effectively restrain, present embodiment is set learning rate η and is reduced along with the increase of iterations t, $\mathrm{\η}=\frac{1}{1+100t};$

The main flowchart process of statistical gradient descending method at first is the initializes weights coefficient as shown in Figure 2 Select the part training dataset then at random

Calculate the kernel function of its training dataset and test data correspondence

With

Then allow i=1, ..., n utilizes formula (3) to carry out n iteration of weight coefficient, after iteration is intact, computing formula (2) optimization aim functional value, if it is littler than the duration of last iteration process, the randomization training dataset carries out the next iteration circulation again, shows that perhaps iteration has searched out the optimal weight coefficient, iteration finishes, and the output optimum linearity is represented weight coefficient;

(3-5) utilize nonlinear function φ that the feature of training and testing data is carried out Nonlinear Mapping to the reproducing kernel Hilbert space (reproducing kernel hilbert space is called for short RKHS) of higher-dimension, i.e. φ (x _i) ^Tφ (x _j)=g (x _i, x _j), g (x wherein _i, x _j) be χ ²Kernel function, x _iAnd x _jIt is data characteristics; χ ²The expression formula of kernel function is exp(-χ ²(x, x)/μ, wherein χ ²Be card side (Chi-squared) distance of symmetrical expression, μ is the χ of current training dataset ²The average of distance.

Formula (2) is transformed into:

$\left\{\begin{array}{c}{w}_j^{k,t+1}=w_j^{k,t}-{\mathrm{\η}\▿}_w\\ {\▿}_w=-X_j^k{y}^k+X_j^k{X}_j^k{w}_j^t+\mathrm{\λ}{\left|\right|y^k-X_j^k\left|\right|}_2^2\·w_j^{k,t}\end{array}\right.----\left(2\right)$

Wherein

It is the training data feature

With test data feature y ^kThe dot product kernel function,

It is the training data feature

With the dot product kernel function of self, P ^k=φ (y ^k) φ (y ^k) be test data feature y ^kWith the dot product kernel function of self; Through the repeatedly iteration of formula (3), obtain optimum sign weight coefficient

(4) identification flowers: the user takes the image of flowers to be identified, and flower chart picture to be identified is extracted characteristic Z ^k, wherein k=1...6 distinguishes representative color histogram, SIFT feature, HOG feature, BOW feature, SSIM feature, GB feature;

According to Z ^k, by flowers classification judgement formula identification flowers classification, and from flowers popular science knowledge storehouse, access this flowers classification corresponding character explanation;

Wherein flowers classification judgement formula is:

J wherein ^*Expression utilizes linear expression and the minimum error values of test data of the training data of certain j classification, selects by minimum, and j is and identifies the flowers classification that obtains.

Above-described embodiment is preferred implementation of the present invention; but embodiments of the present invention are not limited by the examples; other any do not deviate from change, the modification done under spiritual essence of the present invention and the principle, substitutes, combination, simplify; all should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (5)

1. the flowers classification discrimination method based on the sparse sign of local restriction is characterized in that, may further comprise the steps:

(1) collects the flowers image data base and set up flowers popular science knowledge storehouse: the title of utilizing S flowers classification commonly used of existing plants and flowers wikipedia definition, utilize network search engines to search for flowers character introduction and the picture of each flowers classification correspondence, picture is formed flower chart as database; Explanatory note is included into flowers popular science knowledge storehouse;

(2) all pictures in the flowers image data base that step (1) is obtained carry out feature extraction, and what every pictures extracted is characterized as m;

(3) flowers classification identification learning process:

(3-1) from the flowers image data base, select the p pictures, with its feature as test data set Y={y ^k, wherein k=1...m represents m the feature that every pictures extracts respectively, and p＜N, N are the picture number in the flowers image data base; The feature of remaining N-p pictures is as training dataset in the flowers image data base

J=1 wherein, 2 ... S, the classification of expression flowers;

(3-2) with test data feature y ^kUse the training data feature

Linear expression is: Wherein

Be the weight coefficient of training dataset characterization test data set, it is worth greater than 0;

(3-3) for training dataset and test data set linear expression error minimum, increase the weight constraints of the locality characteristic similarity structure of test data feature and training data feature simultaneously, the optimization learning criterion below setting up:

$\underset{W_j^k}{\min }\frac{1}{2}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left|\right|y^k-\underset{j=1}{\overset{S}{\mathrm{\Σ}}}{X}_j^k{w}_j^k\left|\right|}_2^2+\mathrm{\λ}\underset{j=1}{\overset{S}{\mathrm{\Σ}}}{\left|\right|D_j^k\mathrm{\Θ}{w}_j^k\left|\right|}_2---\left(1\right)$

Wherein Θ represents vector dot; λ is the bound term weight, is the constant variables between balance linear expression error and the weight coefficient;

Be test data feature y ^kWith the training data feature

Euclidean distance;

(3-4) adopt the statistical gradient descent method right

Carry out iteration and upgrade, renewal equation is:

$\left\{\begin{array}{c}{w}_j^{k,t+1}=w_j^{k,t}-{\mathrm{\η}\▿}_w\\ {\▿}_w=-X_j^k{y}^k+X_j^k{X}_j^k{w}_j^t+\mathrm{\λ}{\left|\right|y^k-X_j^k\left|\right|}_2^2\·w_j^{k,t}\end{array}\right.----\left(2\right)$

Wherein, t is the iterations of SGD iterative process, and η is the learning rate of statistical gradient decline iterative process;

(3-5) utilize nonlinear function φ that the feature of training and testing data is carried out Nonlinear Mapping to the reproducing kernel Hilbert space of higher-dimension, i.e. φ (x _i) ^Tφ (x _j)=g (x _i, x _j), g (x wherein _i, x _j) be χ ²Kernel function, x _iAnd x _jIt is data characteristics; Formula (2) is transformed into:

$\left\{\begin{array}{c}{w}_j^{k,t+1}=w_j^{k,t}-{\mathrm{\η}\▿}_w\\ {\▿}_w=-h^k+G^k{w}_j^{k,t}+\mathrm{\λ}(P^k-{2h}^h+G^k)\·w_j^{k,t}\end{array}\right.---\left(3\right)$

Wherein

It is the training data feature

With test data feature y ^kThe dot product kernel function,

It is the training data feature

With the dot product kernel function of self, P ^k=φ (y ^k) φ (y ^k) be test data feature y ^kWith the dot product kernel function of self; Through the repeatedly iteration of formula (3), obtain optimum sign weight coefficient

(4) identification flowers: the user takes the image of flowers to be identified, and flower chart picture to be identified is extracted characteristic Z ^k, wherein k=1...m represents m the feature that every pictures extracts respectively;

According to Z ^k, by flowers classification judgement formula identification flowers classification, and from flowers popular science knowledge storehouse, access this flowers classification corresponding character explanation;

Wherein flowers classification judgement formula is:

J wherein ^*Expression utilizes linear expression and the minimum error values of test data of the training data of certain j classification, selects by minimum, and j is and identifies the flowers classification that obtains.

2. the flowers classification discrimination method based on the sparse sign of local restriction according to claim 1 is characterized in that in the step (3-4), η reduces along with the increase of iterations t,

3. the flowers classification discrimination method based on the sparse sign of local restriction according to claim 1 is characterized in that, the described χ of step (3-5) ²The expression formula of kernel function is exp (χ ²(x, x)/μ), χ wherein ²Be card side's distance of symmetrical expression, μ is the χ of current training dataset ²The average of distance.

4. the flowers classification discrimination method based on the sparse sign of local restriction according to claim 1, it is characterized in that, the described title of utilizing S flowers classification commonly used of existing plants and flowers wikipedia definition of step (1), utilize network search engines to search for flowers character introduction and the picture of each flowers classification correspondence, be specially: the title of utilizing 313 herbage flower classifications commonly used of existing plants and flowers wikipedia definition, utilize network search engines to search for flowers character introduction and the picture of each flowers classification correspondence, each flowers classification is downloaded 100 pictures.

5. the flowers classification discrimination method based on the sparse sign of local restriction according to claim 1 is characterized in that, the feature that every pictures extracts comprises color histogram, SIFT feature, HOG feature, BOW feature, SSIM feature, GB feature.

Images