CN107392225A - Plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation - Google Patents
Plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation Download PDFInfo
- Publication number
- CN107392225A CN107392225A CN201710438283.1A CN201710438283A CN107392225A CN 107392225 A CN107392225 A CN 107392225A CN 201710438283 A CN201710438283 A CN 201710438283A CN 107392225 A CN107392225 A CN 107392225A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- fourier
- rarefaction representation
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/247—Aligning, centring, orientation detection or correction of the image by affine transforms, e.g. correction due to perspective effects; Quadrilaterals, e.g. trapezoids
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention devises a kind of plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation, and its technical characteristics is:Pre-process leaf image:Each colored leaf image is converted into gray level image, and separated from background leaf image using Otsu partitioning algorithms, while is converted into two and enters to be worth image, the aperture of leaf image is eliminated with erosion algorithm;Rim detection is carried out using Canny edge detectors;Calculate the barycenter on border;Calculate Fourier descriptors;Build complete dictionary:The Fourier descriptors vector of all leaf image data sets is divided into training set and test set;Complete dictionary is made up of the Fourier descriptors vector of all training sets;Optimized by weighting rarefaction representation grader.The present invention has preferable robustness using ellipse Fourier descriptor to noise and other factors, weighting rarefaction representation grader (WSRC) is identified applied to plant species, using especially pronounced in lower-dimensional subspace, it will be apparent that improve discrimination.
Description
Technical field
It is especially a kind of sparse based on ellipse Fourier descriptor and weighting the invention belongs to image identification technical field
The plants identification method of expression.
Background technology
The plant species identification technology of application image processing and computer vision technique is to environmental protection, land management person
All it is highly important with self-employed tree cultivator.Blade is the important indication of plant species identification.Plant species knowledge is carried out based on leaf morphology
It is not a particularly significant and challenging research field.Because blade is two dimensional image, it is to carry out image procossing most
Suitable part.We can describe leaf well using form, color, texture, edge, texture, vein, tip and substrate
Piece.In substantial amounts of various species, the blade of most plants all has higher inter-species similarity and relatively low kind
Interior similarity as shown in figure 1, the difficult point of leaf image in itself is that it is variable-geometry deformation (rotation, scaling, conversion), and
Change with illumination and the change of shade.Develop related algorithm and technology, such as pattern-recognition, image procossing, digital photography
With portable computer so that automatic species identification technology is achieved.
In botany research, impeller exterior feature and edge are the grown form features of blade.Researcher proposes one newly
The morphology descriptors based on profile of grain husk, referred to as multiple dimensioned distance matrix, the geometric figure of form can be captured, this is described
Symbol is for translating, rotating, scaling and Bilateral Symmetry is all permanent constant.Sayeed et al. proposes a kind of referred to as Local map knot
For the method for structure for identifying and classifying, it applies two features of form and texture.Cerutti et al. proposes small based on the cycle
The descriptor of ripple and leaf outline shape, this method are actually to represent impeller exterior feature with the mode of vector.Zhao et al. proposes one
Kind is using the plants identification method of leaf form, and from existing different as the method for object to study simple leaf, this method can
To identify simple and compound leaf exactly.Zeng et al. proposes a kind of morphology descriptors for plant species identification,
Test result indicates that the effective correct recognition rata of the descriptor is 90% or so.LeafSnap is one more using leaf edges
Yardstick curvature model carries out the mobile applications of leaf image classification.It is obtained when identifying 184 species of Northeastern United States
Very high accuracy.Chaki et al. proposes a kind of plants identification method combined using texture and morphological feature.
Lavania et al. proposes two kinds of leaf image classification algorithms, and scalar invariant features conversion (SIFT) and principal component point is respectively adopted
(PCA) two methods are analysed, applied probability neutral net (PNN) technology is classified to standard Flavia leaf data sets.Barycenter wheel
Wide distance (CCD) is the method for an application outline shape descriptor, is successfully used to plant species identification.Fourier describes
Symbol is particularly useful to plant species identification, because it has permanent consistency to scaling, translating or rotate.Kadir [12] is proposed
A kind of method for merging Fourier descriptors and many other morphological features.As a result show, while apply Fourier descriptors
It is 88% that combination with other several morphological features, which carries out species identification accuracy rate,.
The above method shows that species identification relies primarily on the feature extracted from leaf image.However, these methods can not
The key issue overcome is the different distortion for how handling leaf characteristic and large and small across class change.Jin et al. is proposed should
With the recognition methods of leaf-teeth feature rarefaction representation method (SR), the method describes four kinds of leaf-teeth spies for plant species identification
Sign.Hsiao et al. proposes two kinds of leaf image recognition frameworks based on SR for being used for plant automatic identification, in order to simulate blade figure
Picture, they pass through the training set image study of each leaf species to an excessively complete wordbook.Each dictionary use is schemed from training
The group descriptor collection that extracts learns as in, and each descriptor is the linear combination expression by sub-fraction dictionary atom
's.
Although extract many different features, most of spies in the leaf image that we can identify from plant species
It is more sensitive to levy deformation generally to leaf image, illumination and noise, thus the pretreatment stage before feature extraction need into
Row calibration, translation, rotation and zoom operations.
The content of the invention
Be overcome the deficiencies in the prior art the mesh of the present invention, propose a kind of reasonable in design and discrimination it is high based on
The plants identification method of ellipse Fourier descriptor and weighting rarefaction representation.
The present invention solves its technical problem and takes following technical scheme to realize:
A kind of plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation, comprises the following steps:
Step 1:Pre-process leaf image:Each colored leaf image is converted into gray level image, and calculated using Otsu segmentations
Method separates leaf image from background, while is converted into two and enters to be worth image, and the aperture of leaf image is eliminated with erosion algorithm;
Step 2:Rim detection is carried out using Canny edge detectors;
Step 3:Calculate the barycenter on border;
Step 4:Calculate Fourier descriptors;
Step 5:Build complete dictionary:The Fourier descriptors vector of all leaf image data sets is divided into training set
And test set;Complete dictionary is made up of the Fourier descriptors vector of all training sets;
Step 6:Optimized by weighting rarefaction representation grader.
Further, the implementation method of the step 2 is:Each leaf image is extracted by Canny edge detectors, often
Individual leaf image is ranked up classification in starting point, is denoted as S=in the direction of the clock by the boundary representation of N number of discrete point
{(xi,yi) | i=0,1 ..., N-1 }.
Further, the method for the barycenter on the step 3 calculating border is:
The barycenter on border is expressed as (cx,cy), whereinOriginal coordinates are expressed as (cx,
cy), boundary point set representations are S'={ (xi',yi') | i=0,1 ..., N-1 }, wherein xi'=xi-cx,yi'=yi-cy。
Further, the implementation method of the step 4 is:
The ellipse Fourier conversion of leaf image boundary point set is carried out using equation below first:
Then M Fourier descriptors are extracted according to equation below:
Wherein, ZkThe component of the frequency on border is described, FDs represents Fourier descriptors, and the Fourier descriptors are border
Low frequency normalization fourier coefficient set.
The implementation method of step 5 described further is:
WSRC optimization method is constructed by equation below, calculates the projection coefficient of the complete dictionary of sample;
Subject to Ax=y
Wherein, A ∈ Rm×n(m < n) was complete dictionary, y ∈ RmIt is test sample, x ∈ RnIt is rarefaction representations of the y on A
Vector, W ∈ Rn×nIt is weighting diagonal matrix, diagonal element is w1,w2,...,wn,diIt is training sample and survey
Gap between this y of sample;
The reconstructed residual of specific sub- dictionary is calculated by equation below:
rc(y)=| | y-Dcxc||2, c=1,2 ..., C
Wherein xcIt is the projection coefficient of c-th of specific sub- dictionary, c is the quantity of classification;
Test sample y is identified by following formula:
Label (y)=min rc(y)。
The advantages and positive effects of the present invention are:
The present invention has merged ellipse Fourier descriptor (EFD) and weighting rarefaction representation grader (WSRC), using ellipse
Fourier descriptors have preferable robustness to noise and other factors, such as rotation, scaling, translation, illumination and shade, lead to
Each leaf image construction EFD is crossed, is more suitable for the identification of plant species;Weighting SR graders (WSRC) are applied to plant
Species identification, using especially pronounced in lower-dimensional subspace, it will be apparent that improve discrimination.
Brief description of the drawings
Fig. 1 is different types of leaf image pattern involved in the present invention;
Fig. 2 is the result that the present invention carries out leaf image preprocessing process;
Fig. 3 a are 30 class difference leaf images of ICL databases and corresponding boundary image used in the present invention;
Fig. 3 b are 50 Acer's monophonic maxim leaf images of ICL databases and corresponding boundary image used in the present invention;
Fig. 4 carries out plant species recognition result for the present invention by Fourier descriptors.
Embodiment
The embodiment of the present invention is further described below in conjunction with accompanying drawing.
A kind of plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation, is to be based on oval Fourier
What leaf descriptor (EFD) and weighting rarefaction representation grader (WSRC) were realized, EFD and WSRC are illustrated separately below:
Ellipse Fourier descriptor (EFD)
Fourier descriptors (FD) are an algorithms that shape is represented with border or profile.Its main thought is by making
The population frequency of shape is represented to describe boundary characteristic with one group of data.Its application is very extensive, is regarded as one kind so far
Effective description instrument.Shape description and sorting technique based on FD calculate simply, and have robustness for noise.One
As for, FD can by shape border carry out ellipse Fourier convert to obtain.
Boundary point can use complex representation (such as:For x coordinate as real part, y-coordinate is plural number).(xi,yi) represent
The coordinate of ith pixel on the closed boundary of given X-Y scheme.Redefine (xi,yi) it is plural zi=xi+j·yi
(i=0,1 ..., N-1), wherein N is the number of boundary point.It is by the ellipse Fourier transform definition of border point set afterwards:
The property of ellipse Fourier descriptor is as follows:
If Bian Huan ︰ X-Y schemes are according to distance p0=a0+jb0Enter line translation, i.e. zi'=zi+p0(i=0,1 ..., N-
1), its EFT is expressed as:
As can be seen from the above equation, conversion only influences EFT immediate component (DC).
If Suo Fang ︰ X-Y schemes zoom in and out according to distance, i.e. zi'=λ zi(i=0,1 ..., N-1), then its EFD
The factor it will zoom in and out in the same proportion:
Z'k=λ Zk(k=0,1 ..., N-1) (3)
Rotation:If X-Y scheme is centered on origin according to angleRotated, i.e.,
Then EFD will be multiplied by the identical factor:
Starting point:If the starting point on closed boundary is moved to i from 00, i.e.,Then EFD will be changed into:
In formula (1), ZkThe component of the frequency on border is described:When variable k is close to 0, ZkLow-frequency information is described, i.e., it is near
As border;High-frequency information then describes the details of shape.Z0It is immediate component, represents the centroid position on border, it is for border
Description is without what meaning.Z1The size on border is described, if all parameters are set to 0, border will turn into a circle;Other
Frequency parameter will be to Z1The circle formed is modified.In order to carry out simple, effective Boundary Recognition, we will be without using Z0With
The high fdrequency component of Fourier parameter.Z1It is generally used for normalizing fourier coefficient set so that they can keep after the scaling of border
It is constant.The low frequency normalization fourier coefficient set on border is referred to as Fourier descriptors (FDs), is expressed as:
Wherein M<N, FDs are considered as the characteristic vector corresponding with border low frequency component, for describing X-Y scheme
Border.
It can be learnt by formula (2) to formula (5), FDs has perseverance constant to rotating, scaling, translating and changing starting point
Property.Because the border of figure is binary curve, and noise at the boundary corresponds to and is dropped no high frequency coefficient, so FDs
Also there is robustness for noise and illumination to a certain extent.
Weight rarefaction representation method (WSRC)
In actual applications, the application of rarefaction representation grader (SRC) is a lot because it can use up test sample can
The reconstruction of energy.However, typical SRC have ignored similitude this problem between test sample and individual training sample.In order to
SRC performance is improved, we have proposed weighting rarefaction representation grader.In general, WSRC solves following weighting l1- models
Number minimization problem:
Wherein, A ∈ Rm×n(m < n) was complete dictionary, contained all training sample data.y∈RmIt is test specimens
This, x ∈ RnIt is that rarefaction representations of the y on A is vectorial, W ∈ Rn×nIt is weighting diagonal matrix, diagonal element is w1,w2,...,wn:
Wherein diIt is the gap between training sample and test sample y.
WSRC optimization problems can solve by weighting l1- norm minimums algorithm in formula (7).Provide test sample
Y, obtain its projection coefficient.Then, the residual error of specific sub- dictionary is defined as:
rc(y)=| | y-Dcxc||2, c=1,2 ..., C (9)
Wherein xcIt is the projection coefficient of c-th of specific sub- dictionary, c is the quantity of classification.
Then, test sample y is defined as
Label (y)=min rc(y) (10)
WSRC is SRC extension.If the in fact, weight w in WSRCi(i=1,2 ..., n) both is set to 1,
WSRC is exactly SRC.It is compared with SRC, the advantages of WSRC:It can retain test while application sparse linear representation
Similitude between the sample training data adjacent with it.Because test sample can be expressed as the linear combination of training sample,
The distance between they are advantageous to rarefaction representation, while are preferably classified.Once obtaining sparse coefficient, it just can
The classification of query image is identified with low error.
A kind of plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation, comprises the following steps:
Step 1:Pre-process leaf image:Each colored leaf image is converted into gray level image, and calculated using Otsu segmentations
Method separates leaf image from background, while is converted into two and enters to be worth image, and the aperture of leaf image is eliminated with erosion algorithm.
Step 2:Rim detection is carried out using Canny edge detectors.
Rim detection can reduce image data amount while picture structure characteristic is kept.Canny edge detectors are
The edge detection method of one standard, including 5 processes:Smoothed image, calculate gradient, non-maxima suppression, using dual threshold
Algorithm and Edge track.We extract each leaf image using Canny edge detectors, and this can be by the side of N number of discrete point
Boundary represents.We are ranked up classification in the direction of the clock in starting point, are denoted as S={ (xi,yi) | i=0,1 ..., N-
1}。
Step 3:Calculate the barycenter on border.
The barycenter on border is expressed as (cx,cy), whereinOriginal coordinates are expressed as (c by usx,
cy), boundary point set representations are S'={ (x 'i,y′i) | i=0,1 ..., N-1 }, wherein x 'i=xi-cx,y′i=yi-cy。
Step 4:Calculate Fourier descriptors.
The ellipse Fourier conversion of leaf image boundary point set can be quickly carried out from formula (1), is then carried according to formula (6)
Take M Fourier descriptors, abbreviation FDs.In fact, FDs is an one-dimensional characteristic vector, has and derive easy, standardization
The easy several advantages of easy and matching.
Step 5:Build complete dictionary.The FDs vectors of all leaf image data sets are divided into training set and test set.It is complete
Standby dictionary is made up of the FDs vectors of all training sets.
Step 6:Optimized by weighting rarefaction representation grader.
As shown in formula (7), a complete dictionary is given, constructs the optimization method of the WSRC as shown in publicity (7).So
The projection coefficient of the complete dictionary of sample is calculated according to formula (7) afterwards;The reconstructed residual of specific sub- dictionary is calculated according to formula (9);
Test sample is identified according to formula (10).
Further checking is done to the present invention below by experiment.
We are compared with Four Plants recognition methods:Using the method (SCTF) of form, color and textural characteristics,
Using the method (CCD) of barycenter profile distance, morphological feature, Fourier descriptors and multiple dimensioned distance matrix are combined
Method (SFM)) and using SR method.All experiments are used on the Intel Xeon X3430 PC with 2GB RAM
What the softwares of MATLAB 7.0 were carried out.In addition, Canny edge detection algorithms are realized by edge function, coefficient in WSRC and residual
Difference solves (http by the SRC functions in SR tool boxes://sites.google.com/site/sparsereptool).
Due to this factor of leaf color contributed when carrying out blade automatic identification it is smaller and more intractable, so
In experiment afterwards, we by each original RGB leaves image it is appropriate be cut to identical size, and be converted into gray scale
Image,
Gray=0.2989R+035870G+0.1140B (11)
Because all leaves in ICL databases are all surrounded by limited noise background, we can use Otsu
Partitioning algorithm separates gray level image and ambient noise.We are by each greyscale image transitions into binary picture, Ran Houyong
Canny edge detectors algorithm extracts boundary image.Fig. 2 is the result of leaf image preprocessing process.Fig. 3 a and Fig. 3 b are ICL
Some samples and corresponding boundary image in database, it can be found that floristics can pass through border from Fig. 3 a
Curves Recognition, from Fig. 3 b it was found from, kind inner boundary image it is very different each other.Therefore, many existing plants identification methods have
Relatively low discrimination and relatively low robustness.We intend to use whole data set testing algorithm, but when the quantity of training sample
When very big, the WSRC optimization problems solved in equation (7) then need long time.Inspired by Hu et al., we will own
Sample is divided into three subsets, is respectively labeled as subset A, B and C, and this makes it possible to carry out achievement assessment using whole data set.
Each subset includes about 70 kinds of classifications, and each class includes 30 samples.70 classifications in subset A are all from 200 classifications
Selective, the blade of wherein most can be easily discriminated.Subset B 70 classifications are from 70 classes except A classifications
Randomly select in 130 classifications beyond not, can easily be differentiated by tooth feature.Remaining 60 classification
Then form subset C.So subset B is more difficult to carry out plant species identification than subset A, subset C is that three son concentrations are most difficult to carry out
Species identification.To each subset, we with two methods are identified the folding cross-validation methods (5FCV) of Shi Yan ︰ five and stay one
Method cross validation (LOOCV).All Fourier descriptors extracted from training sample are used to construct complete dictionary, from test
All Fourier descriptors extracted in sample are used for the performance of method of testing.For each leaf image, we are equal from its border
128 points of even sampling, and build Fourier descriptors on these aspects.Important parameter in the method is to be determined by experiment
The dimension of Fourier descriptors, scope is from 50 to 120.Fig. 4 represents to describe by the different Fouriers of LOOCV method constructs
Accord with the change of discrimination of the number on to subset A.From fig. 4, it can be seen that work as Fourier descriptors number more than 80, identification
Rate tends to balance.So in following experiment, we define quantity as 80.
In SCTF, CCD, SFM, SR and the method that we are proposed, using two methods of the experiment of 5FCV and LOOCV
Independently repeated 50 times, finally draw average recognition rate.Table 1 gives the result of average recognition rate:
Table 1
It can learn that the present invention is better than other Four Plants recognition methods from table 1.Reason is as follows:(1) Fourier descriptors
To noise and other factors, such as rotation, scaling, translation, illumination and shade have preferable robustness.(2) WSRC is better than SRC,
Improved method is in lower-dimensional subspace using especially pronounced.Therefore, the present invention is feasible plant species recognition methods.
Although multiple dimensioned distance matrix (MDM) and CCD also have robustness to rotating, scaling etc., can be obtained from Fig. 3 b
To know, sample is there is also many differences in kind, so MDM the and CCD features of sample are different from each other in the kind of extraction, two methods
Discrimination it is relatively low.In SCTF methods, the form of leaf, color and texture be all used to identify species, but the color of blade exists
Different seasons is change, so when identifying subset A and B, discrimination is all very low;And identification of the method for subset C
Rate is then higher than other three kinds of methods, and this shows that the color of blade is sometimes meaningful.SR and discrimination rate of the present invention are above
Other methods, because rarefaction representation classification can intuitively find most sparse table of the test sample on complete dictionary
Show, so as to eliminate the negative effect by categorised decision.Especially in the present invention, the gap between test sample and training sample
For offsetting the influence of sparse coefficient size, and strengthen the differentiation to rarefaction representation, be so favorably improved the correct of classification
Property.So effect of the invention is better than other method.
It is emphasized that embodiment of the present invention is illustrative, rather than it is limited, therefore the present invention
It is every by those skilled in the art's technique according to the invention including the embodiment being not limited to described in embodiment
The other embodiment that scheme is drawn, also belongs to the scope of protection of the invention.
Claims (5)
- A kind of 1. plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation, it is characterised in that including following Step:Step 1:Pre-process leaf image:Each colored leaf image is converted into gray level image, and will using Otsu partitioning algorithms Leaf image is separated from background, while is converted into two and enters to be worth image, and the aperture of leaf image is eliminated with erosion algorithm;Step 2:Rim detection is carried out using Canny edge detectors;Step 3:Calculate the barycenter on border;Step 4:Calculate Fourier descriptors;Step 5:Build complete dictionary:The Fourier descriptors vector of all leaf image data sets is divided into training set and test Collection;Complete dictionary is made up of the Fourier descriptors vector of all training sets;Step 6:Optimized by weighting rarefaction representation grader.
- 2. the plants identification method according to claim 1 based on ellipse Fourier descriptor and weighting rarefaction representation, its It is characterised by:The implementation method of the step 2 is:Each leaf image, each blade figure are extracted by Canny edge detectors As the boundary representation by N number of discrete point, classification is ranked up in the direction of the clock in starting point, is denoted as S={ (xi,yi)|i =0,1 ..., N-1 }.
- 3. the plants identification method according to claim 1 based on ellipse Fourier descriptor and weighting rarefaction representation, its It is characterised by:The method that the step 3 calculates the barycenter on border is:The barycenter on border is expressed as (cx,cy), whereinOriginal coordinates are expressed as (cx,cy), border Point set is expressed as S'={ (x 'i,y′i) | i=0,1 ..., N-1 }, wherein x 'i=xi-cx,y′i=yi-cy。
- 4. the plants identification method according to claim 1 based on ellipse Fourier descriptor and weighting rarefaction representation, its It is characterised by:The implementation method of the step 4 is:The ellipse Fourier conversion of leaf image boundary point set is carried out using equation below first:<mrow> <msub> <mi>Z</mi> <mi>k</mi> </msub> <mo>=</mo> <mi>D</mi> <mi>F</mi> <mi>T</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mi>N</mi> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>N</mi> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <msub> <mi>z</mi> <mi>i</mi> </msub> <mi>exp</mi> <mrow> <mo>(</mo> <mo>-</mo> <mfrac> <mrow> <mi>j</mi> <mn>2</mn> <mi>&pi;</mi> <mi>i</mi> <mi>k</mi> </mrow> <mi>N</mi> </mfrac> <mo>)</mo> </mrow> <mo>,</mo> <mrow> <mo>(</mo> <mi>k</mi> <mo>=</mo> <mn>0</mn> <mo>,</mo> <mn>1</mn> <mo>,</mo> <mo>...</mo> <mo>,</mo> <mi>N</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>Then M Fourier descriptors are extracted according to equation below:<mrow> <mi>F</mi> <mi>D</mi> <mi>s</mi> <mo>=</mo> <mo>{</mo> <mfrac> <mrow> <mo>|</mo> <mo>|</mo> <msub> <mi>Z</mi> <mn>2</mn> </msub> <mo>|</mo> <mo>|</mo> </mrow> <mrow> <mo>|</mo> <mo>|</mo> <msub> <mi>Z</mi> <mn>1</mn> </msub> <mo>|</mo> <mo>|</mo> </mrow> </mfrac> <mo>,</mo> <mfrac> <mrow> <mo>|</mo> <mo>|</mo> <msub> <mi>Z</mi> <mn>3</mn> </msub> <mo>|</mo> <mo>|</mo> </mrow> <mrow> <mo>|</mo> <mo>|</mo> <msub> <mi>Z</mi> <mn>1</mn> </msub> <mo>|</mo> <mo>|</mo> </mrow> </mfrac> <mo>,</mo> <mo>...</mo> <mo>,</mo> <mfrac> <mrow> <mo>|</mo> <mo>|</mo> <msub> <mi>Z</mi> <mrow> <mi>M</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mo>|</mo> <mo>|</mo> </mrow> <mrow> <mo>|</mo> <mo>|</mo> <msub> <mi>Z</mi> <mn>1</mn> </msub> <mo>|</mo> <mo>|</mo> </mrow> </mfrac> <mo>}</mo> </mrow>Wherein, ZkThe component of the frequency on border is described, FDs represents Fourier descriptors, and the Fourier descriptors are the low frequency on border Normalize fourier coefficient set.
- 5. the plants identification method according to claim 1 based on ellipse Fourier descriptor and weighting rarefaction representation, its It is characterised by:The implementation method of the step 5 is:WSRC optimization method is constructed by equation below, calculates the projection coefficient of the complete dictionary of sample;<mrow> <munder> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <mi>x</mi> </munder> <mo>|</mo> <mo>|</mo> <mi>W</mi> <mi>x</mi> <mo>|</mo> <msub> <mo>|</mo> <mn>1</mn> </msub> </mrow>Subject to Ax=yWherein, A ∈ Rm×n(m < n) was complete dictionary, y ∈ RmIt is test sample, x ∈ RnIt is rarefaction representation vectors of the y on A, W∈Rn×nIt is weighting diagonal matrix, diagonal element is w1,w2,...,wn,diIt is training sample and test sample y Between gap;The reconstructed residual of specific sub- dictionary is calculated by equation below:rc(y)=| | y-Dcxc||2, c=1,2 ..., CWherein xcIt is the projection coefficient of c-th of specific sub- dictionary, c is the quantity of classification;Test sample y is identified by following formula:Label (y)=min rc(y)。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710438283.1A CN107392225A (en) | 2017-06-12 | 2017-06-12 | Plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710438283.1A CN107392225A (en) | 2017-06-12 | 2017-06-12 | Plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107392225A true CN107392225A (en) | 2017-11-24 |
Family
ID=60333303
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710438283.1A Pending CN107392225A (en) | 2017-06-12 | 2017-06-12 | Plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107392225A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112861693A (en) * | 2021-02-02 | 2021-05-28 | 东北林业大学 | Plant leaf microscopic image pore segmentation method based on deep learning |
CN117372790A (en) * | 2023-12-08 | 2024-01-09 | 浙江托普云农科技股份有限公司 | Plant leaf shape classification method, system and device |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105574475A (en) * | 2014-11-05 | 2016-05-11 | 华东师范大学 | Common vector dictionary based sparse representation classification method |
CN105631478A (en) * | 2015-12-25 | 2016-06-01 | 天津科技大学 | Plant classification method based on sparse expression dictionary learning |
-
2017
- 2017-06-12 CN CN201710438283.1A patent/CN107392225A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105574475A (en) * | 2014-11-05 | 2016-05-11 | 华东师范大学 | Common vector dictionary based sparse representation classification method |
CN105631478A (en) * | 2015-12-25 | 2016-06-01 | 天津科技大学 | Plant classification method based on sparse expression dictionary learning |
Non-Patent Citations (5)
Title |
---|
DENGSHENGZHANG等: "A comparative study of curvature scale space and fourier descriptors for shape-based image retrieval", 《JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION》 * |
ZIZHUFAN等: "Weighted sparse representation for face reconition", 《NEUROCOMPUTING》 * |
彭正初: "基于傅里叶描述子的物体形状识别的研究", 《中国优秀硕士学位论文全文数据库•信息科技辑》 * |
施耀: "基于稀疏表示的人脸识别算法研究", 《中国优秀硕士学位论文全文数据库•信息科技辑》 * |
李建斌: "基于稀疏表示的植物叶片分类识别研究", 《中国优秀硕士学位论文全文数据库•信息科技辑》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112861693A (en) * | 2021-02-02 | 2021-05-28 | 东北林业大学 | Plant leaf microscopic image pore segmentation method based on deep learning |
CN112861693B (en) * | 2021-02-02 | 2022-08-30 | 东北林业大学 | Plant leaf microscopic image pore segmentation method based on deep learning |
CN117372790A (en) * | 2023-12-08 | 2024-01-09 | 浙江托普云农科技股份有限公司 | Plant leaf shape classification method, system and device |
CN117372790B (en) * | 2023-12-08 | 2024-03-08 | 浙江托普云农科技股份有限公司 | Plant leaf shape classification method, system and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109154978B (en) | System and method for detecting plant diseases | |
CN106803247B (en) | Microangioma image identification method based on multistage screening convolutional neural network | |
Li et al. | SAR image change detection using PCANet guided by saliency detection | |
Singh et al. | Svm-bdt pnn and fourier moment technique for classification of leaf shape | |
Du et al. | Leaf shape based plant species recognition | |
WO2021003951A1 (en) | Hyperspectral image classification method based on label-constrained elastic network graph model | |
İlsever et al. | Two-dimensional change detection methods: remote sensing applications | |
Bhardwaj et al. | Recognition of plants by leaf image using moment invariant and texture analysis | |
CN109615008B (en) | Hyperspectral image classification method and system based on stack width learning | |
CN111401145B (en) | Visible light iris recognition method based on deep learning and DS evidence theory | |
CN106204651B (en) | A kind of method for tracking target based on improved judgement with generation conjunctive model | |
Shahab et al. | How salient is scene text? | |
CN110969121A (en) | High-resolution radar target recognition algorithm based on deep learning | |
Li et al. | SDBD: A hierarchical region-of-interest detection approach in large-scale remote sensing image | |
Xu et al. | A robust hierarchical detection method for scene text based on convolutional neural networks | |
CN107392225A (en) | Plants identification method based on ellipse Fourier descriptor and weighting rarefaction representation | |
CN105844299B (en) | A kind of image classification method based on bag of words | |
CN108960005B (en) | Method and system for establishing and displaying object visual label in intelligent visual Internet of things | |
Yang et al. | Rapid image detection and recognition of rice false smut based on mobile smart devices with anti-light features from cloud database | |
Elsayed et al. | Hand gesture recognition based on dimensionality reduction of histogram of oriented gradients | |
Peng et al. | Fully convolutional neural networks for tissue histopathology image classification and segmentation | |
Zhang et al. | Saliency detection via image sparse representation and color features combination | |
Bhugra et al. | Automatic quantification of stomata for high-throughput plant phenotyping | |
CN113887652B (en) | Remote sensing image weak and small target detection method based on morphology and multi-example learning | |
CN107967492A (en) | Bayes's conspicuousness detection method that a kind of combination is detected like physical property |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171124 |