CN106991441A - Merge the plant specimen sorting technique and system of multiple dimensioned direction textural characteristics - Google Patents

Merge the plant specimen sorting technique and system of multiple dimensioned direction textural characteristics Download PDF

Info

Publication number
CN106991441A
CN106991441A CN201710200865.6A CN201710200865A CN106991441A CN 106991441 A CN106991441 A CN 106991441A CN 201710200865 A CN201710200865 A CN 201710200865A CN 106991441 A CN106991441 A CN 106991441A
Authority
CN
China
Prior art keywords
image
plant
textural characteristics
plant specimen
specimen
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
Application number
CN201710200865.6A
Other languages
Chinese (zh)
Inventor
侯北平
王周敏
穆清萍
黄�俊
董霏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Lover Health Science and Technology Development Co Ltd
Zhejiang University of Science and Technology ZUST
Original Assignee
Zhejiang Lover Health Science and Technology Development Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang Lover Health Science and Technology Development Co Ltd filed Critical Zhejiang Lover Health Science and Technology Development Co Ltd
Priority to CN201710200865.6A priority Critical patent/CN106991441A/en
Publication of CN106991441A publication Critical patent/CN106991441A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • 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
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a kind of plant specimen sorting technique and system for merging multiple dimensioned direction textural characteristics, including：Herborization sample image；Image preprocessing；Extract area-of-interest；Area-of-interest is split by connected domain, the plant leaf blade set split；Calculate individual blade geometric properties；The main direction of blade texture is determined by the amplitude histogram of gradients of individual blade；Gaussian filtering is carried out to single blade using the Gaussian kernel of different scale, and calculates the textural characteristics on main direction；Using eigenmatrix, individual blade is classified with SVM algorithm, the vaned results set E of plant leaf blade specimen institute is obtained；Using the most classification results as final plant specimen to be detected of set E occurrence numbers.This plant specimen sorting technique, detection efficiency and nicety of grading are high, the factors such as multiple dimensioned, otherness have been taken into full account, while greatly reducing the labor intensity of testing staff.

Description

Merge the plant specimen sorting technique and system of multiple dimensioned direction textural characteristics
Technical field
The invention belongs to computer pattern recognition field, and in particular to a kind of multiple dimensioned direction textural characteristics of fusion Plant specimen sorting technique and system.
Background technology
Plant is that species is the most various in the world, the widest a kind of life form of distribution, with human lives and ring Border has very close relationship.Meanwhile, plant is the highly important part of global biosphere, in human being's production, life Irreplaceable effect is all played in various fields living.With the development of modern civilization, the destruction of ecological environment, plant species Class is constantly reduced.Therefore plant classification and identification are for protecting the diversity of plant, and maintaining ecological balance is most important.
Compared to the root of plant, stem, for the organ such as branch, the blade of plant can be more prone to extract feature, and Time-to-live is longer.Plant leaf blade is a very effective index for studying different plant species morphological differences simultaneously, therefore The Classification and Identification of plant leaf blade be for plant classification the most directly and effective manner.However, due to plant leaf blade mark There is the situation such as a large amount of words and legend annotation in incomplete blade, breakage and sample during collection, preservation etc. in this, this A little factors turn into a technical barrier of the classification to plant leaf blade.
The classification and identification of plant leaf blade are general manually visually to be led using offline under existence conditions and technical background Observation and classification in sight.Observed by the people with specialty background knowledge and determine different classes of plant, although certain The demand of classification is met in degree, but there is Railway Project in artificial detection：One is people in point of the long-time for plant leaf blade It is easy to mistake occur in class process；Two be for similar plant leaf blade, it is necessary to people spends a lot of time is observed and recognized, is imitated Rate is low；Three be that floristics is various, it is necessary to consume substantial amounts of man power and material, is wasted to a certain extent to professional people Just use.
It can be seen that, simple artificial detection classification is wasted time and energy and efficiency is low.Publication No. CN103617417B middle promulgated by the State Council Bright patent " Automatic plant identification method and system " disclose it is a kind of built by cloud data, feature extraction, classification differentiate meter Calculate, final to judge which species plant belongs to, this method and system can realize the classification of plant, but it is several only to only used shape What feature, the factors such as direction, yardstick, texture is not considered, it is impossible to suitable for leaf blade specimen environment-identification.
Classify current limitation for plant leaf blade specimen, the present invention takes into full account that floristics is various, same plant Having differences property of blade (such as size, yardstick) practical factor, realizes the classification of leaf blade specimen.
The content of the invention
For the above-mentioned technological deficiency and deficiency present in prior art, the invention provides the multiple dimensioned direction of one kind fusion The plant specimen sorting technique and system of texture and geometric properties, detection efficiency and nicety of grading are high, while greatly reducing inspection The labor intensity and cost of survey personnel.
A kind of plant specimen sorting technique for merging multiple dimensioned direction texture and geometric properties, including：
(1) plant leaf blade specimen is collected in advance, and different plant leaf blade specimens are made a distinction, and plant leaf blade specimen is returned For several typical categories, the image of the individual blade of quantity is set to collection of specimens of all categories, Sample Storehouse is set up；
(2) the plant specimen image of plant specimen to be detected is gathered；
(3) image preprocessing is carried out to the plant specimen image collected；
(4) to plant specimen image zooming-out area-of-interest (ROI, the Region of after pretreatment Interesting)；
(5) image segmentation is carried out for area-of-interest, obtains including the image collection of multiple single blade images；
(6) the geometric properties matrix of a leaf image in image collection is calculated；
(7) the textural characteristics matrix of current vane image is calculated；
(8) obtained geometric properties matrix and the eigenmatrix of textural characteristics matrix computations current vane image is utilized；
(9) eigenmatrix of the eigenmatrix of current vane image and the image of Sample Storehouse is utilized into support vector cassification Algorithm (SVM, Support Vector Machine) is classified, and obtains the identification classification results of current vane image；
(10) repeat step (6)-(9), classification is identified to leaf image all in image collection, obtains working as planting in advance The most result of occurrence number is used as final plant leaf blade specimen in the classification results set of thing sample, selection sort results set Classification results.
Preferably, in step (4), the step of extracting area-of-interest includes：
(4-1) extracts the connected domain in the plant specimen image after pretreatment according to eight directional chain-codes；
(4-2) is screened according to length and width, the area of the minimum enclosed rectangle of connected domain, is rejected in plant specimen image Garbage (word description such as that may be present, numbering, legend) in addition to branches and leaves；
(4-3) carries out morphological operation to the connected domain extracted：Opening operation, eliminates the branch part in plant specimen, Obtain area-of-interest.
Preferably, in step (5), leaf image point is carried out to area-of-interest using the minimum enclosed rectangle of connected domain Cut.
Preferably, in step (6), the geometric properties matrix is obtained by following methods：Calculate current vane image Rectangular degree, circularity, the geometric properties of three invariable rotaries of convexity, and then obtain geometric properties matrix.
Preferably, in step (7), the method for calculating the textural characteristics matrix of current vane image is as follows：
(7-1) carries out gaussian filtering to current vane image；
(7-2) is directed in the leaf image after gaussian filtering each pixel, calculates the gradient information of each pixel；
(7-3) determines bins bar numbers, using the gradient magnitude of pixel as weight, calculates the blade figure after gaussian filtering Amplitude-gradient orientation histogram of picture；
Peak value highest bins in amplitude-gradient orientation histogram as the main direction of plant leaf blade, is denoted as by (7-4) φ；
(7-5) is calculated under different scale respectively, the textural characteristics of main direction φ local binary patterns, is finally given and is worked as The textural characteristics matrix of frontal lobe picture.
Preferably, in step (7-1), selection (2k+1) × (2k+1) templates, wherein k=1,2,3, using using two dimension Space Gauss discrete function calculates Filtering Template, and then current vane image is filtered using obtained Filtering Template.
Preferably, in step (7-5), wherein：
Main direction φ circular operator is：
The texture eigenvalue of the local binary patterns of each pixel is calculated by following formula and obtained in leaf image：
Wherein：xc、ycIt is the transverse and longitudinal coordinate of central pixel point, xp、ypIt is the transverse and longitudinal coordinate of sampled point；P is sampled point Number, p=0 ..., P-1, R are sample radius, gcCentered on pixel gray value, gpFor the gray value of sampled point, it can use double Linear interpolation is calculated and obtained.
Present invention also offers a kind of plant specimen categorizing system for merging multiple dimensioned direction textural characteristics, its feature exists In, including：
Image acquisition units, the plant specimen image for gathering plant specimen to be measured；
Graphics processing unit, for being pre-processed to described plant specimen image, region of interesting extraction, image point Cut, feature extraction, and calculate classification results；Specific method is：
(i) image for gathering image acquisition units carries out image gray processing, binaryzation；
(ii) ROI extractions are carried out to the plant specimen image after binaryzation using morphological operation；
(iii) image segmentation is carried out to ROI region, obtains the set of plant leaf blade；
(iv) its texture and geometric properties is extracted for each blade, finally gives eigenmatrix；
(v) all blade classification results are calculated with sorting algorithm, and will appear from the most conduct of number of times and finally recognize knot Really；
Testing result display unit, for showing classification results.
Preferably, described image collecting unit includes industrial camera, camera lens and lighting source；Described image procossing Unit is industrial control computer, and the classification of plant specimen is realized by sorting algorithm and software programming.Described image processing is single Human interface software can also be included in first, the to be sorted plant specimen image and classification knot described for showing in real time Really, record sort historical data, and the operational order of user is received to software progress parameter setting.
The present invention, using the texture and geometric properties method for merging multiple dimensioned direction, is realized by high-speed industrial video camera Plant leaf blade specimen Fast Classification, has the advantages that multiple dimensioned directionality, real-time, precision are high；Ensureing the premise of measurement accuracy Under corresponding testing cost is greatly reduced.
Brief description of the drawings
Fig. 1 is the step schematic flow sheet of the plant specimen sorting technique of the present invention.
Fig. 2 is 3 × 3 regions of piece image (z is gray value).
Fig. 3 is that bilinear interpolation calculates schematic diagram.
Embodiment
In order to more specifically describe the present invention, below in conjunction with the accompanying drawings and embodiment is to technical scheme It is described in detail.
As shown in figure 1, a kind of plant specimen sorting technique for merging multiple dimensioned direction texture and geometric properties, including it is as follows Step：
(1) plant leaf blade is collected in advance, and different plant leaf blade specimens are manually distinguished, and plant leaf blade specimen is returned For several typical categories, Sample Storehouse is used as to plant leaf blade of all categories collection certain amount image；In this step, on an image Only comprising a plant leaf blade, every kind of some images of classification herborization；
(2) plant specimen image to be detected is gathered in real time；Plant specimen image generally comprise some blades, one or many Individual limb and also include some label character information etc. sometimes；
(3) image preprocessing is carried out to the plant specimen image collected：Gray processing, binaryzation：
Image gray processing：
F (i, j)=0.3R (i, j)+0.59G (i, j)+0.11B (i, j) (1)
Image binaryzation：
Wherein f (i, j) and g (i, j) represent the corresponding gray level image of plant specimen image and bianry image respectively, R (i, J), G (i, j), B (i, j) represent three kinds of images after the separation of plant specimen image RGB channel, T1Represent the binaryzation threshold of setting Value, i, j is the coordinate of a certain pixel on plant specimen image respectively；
(4) image after being directed to image preprocessing well carries out the extraction of area-of-interest：
I. connected domain profile is extracted according to eight directional chain-codes, according to the length and width of the minimum enclosed rectangle of connected domain, area Screened, garbage (word description such as that may be present, volume in plant specimen image are rejected using formula (6) and formula (7) Number, legend etc.)：
Dmax=max [dis (Cm,Cn)] (3)
Dmin=max [dis (Cm,Cn)] (4)
Ad=Dmax×Dmin (5)
{Dmax or Dmin or Ad＜ { T2min or T3min or T4min} (6)
{Dmaxor Dmin or Ad＞ { T2maxor T3maxor T4max} (7)
Wherein CmAnd CnIt is arbitrary two pixels of column or row on connected domain profile, DmaxFor the height value of connected domain (or length value), DminFor the width value of connected domain, AdFor the area value of connected domain boundary rectangle；T2min、T3min、T4min、T2max、 T3max、T4maxThe threshold value of garbage connected domain is respectively screened, the leaf structure feature that can be screened according to actual needs carries out true It is fixed, during judgement：Length (the D of connected domainmax) or width (Dmin) or area (Ad) any numerical value be less than given threshold (such as Dmax Less than T2min, or DminLess than T3min) or more than given threshold (such as DmaxMore than T2max, or DminMore than T3max), It is judged as garbage；Max [] is to take max function；dis(Cm,Cn) represent to solve pixel CmAnd CnBetween abscissa (it is directed to Dmin) or ordinate (be directed to Dmax) distance function；
II. morphological operation is carried out to the connected domain extracted：Opening operation, eliminates the branch part in plant specimen, obtains To area-of-interest：
(5) leaf image segmentation is carried out to area-of-interest using the minimum enclosed rectangle of connected domain, separated from artwork, Obtain the set S={ S of leaf image group1,S2,S3…Sz…Sn, wherein n is the quantity of leaf image, SzFor z-th of plant The leaf image of thing blade；
(6) monolithic leaf image S is calculatedz(z=1,2,3...n) rectangular degree (Rectangularity), circularity (Circularity), the geometric properties of convexity (Convexity) three invariable rotaries：
Rectangulatity=A/Ad (8)
Convexity=A/Ac (10)
Wherein rmaxBe the center of gravity of connected domain to the ultimate range of all profiles, A refers to connected domain area；AcRefer to connection The convex closure area in domain；
Geometry is obtained by rectangular degree (Rectangularity), circularity (Circularity), convexity (Convexity) special Levy matrix N.
(7) monolithic leaf image S is calculatedz(z=1,2,3...n) multiple dimensioned direction textural characteristics matrix：
I. (2k+1) × (2k+1) templates are chosen, wherein k=1,2,3, it is utilized respectively formula (11) and calculates k=1,2,3 Filtering Template, is then filtered using obtained Filtering Template to leaf image：
Gaussian filtering formwork calculation formula is as follows：
Usually,
σ=0.3 ((k-1) × 0.5-1)+0.8 (12)
Wherein σ refers to the Gaussian kernel of gaussian filtering, and k is Filtering Template radius size, and (i, j) is template internal coordinate label；
II. each pixel in the leaf image after gaussian filtering (k=1) is directed to, each pixel is calculated Gradient information.
For function f (x, y), f is two-dimentional column vector in the definition of coordinate (x, y) place gradient：
VectorRange value (length) be expressed as M (x, y), gradient direction is θ (x, y), i.e.,：
gx=(z7+2z8+z9)-(z1+2z2+z3) (16)
gy=(z3+2z6+z9)-(z1+2z4+z7) (17)
Wherein z1~z9Mark is shown in Fig. 2, wherein gxIt is the gradient of X-direction, gyIt is the gradient of Y-direction, z1~z9Belonging to referring to The pixel value of label respective pixel；
III. bins bar number b are determined, using the gradient magnitude of the pixel in leaf image as weight w, are calculated Amplitude-gradient orientation histogram of gaussian filtering (k=1) leaf image afterwards；
IV. using peak value highest bins in amplitude-gradient orientation histogram as plant leaf blade main direction φ：
Wherein：I=0,1,2...b；
V. the local binary patterns of plant leaf blade main direction φ (i.e. during k=1,2,3) under different scale are calculated respectively The textural characteristics matrix M of (LBP, Local Binary Patterns)1,M2,M3：
Main direction φ circular operator：
xc、ycIt is the transverse and longitudinal coordinate of central pixel point, xp、ypIt is the transverse and longitudinal coordinate of sampled point；
The LBP textural characteristics of any pixel point can be expressed as in leaf image：
Wherein P is sampled point number, and p=0 ..., P-1, R are sample radius, gcCentered on pixel gray value, gpFor The gray value of sampled point；
By traveling through whole leaf image, the textural characteristics data of all pixels point are obtained, k=1 is respectively obtained, 2, 3 corresponding eigenmatrix M1,M2,M3；
Wherein gpAdopt and try to achieve with the following method：
X dimension linear interpolation calculations：
Y dimension linear interpolation calculations：
That is bilinear interpolation formula, as shown in Figure 3：
(8) eigenmatrix of monolithic leaf image, including multiple dimensioned direction textural characteristics and geometric properties are calculated：
C=[ε1×M1 ε2×M2 ε3×M3 ε4×N] (27)
Wherein M1,M2,M3For the multiple dimensioned direction textural characteristics matrix of monolithic leaf image, N is monolithic plant leaf blade The geometric properties matrix of image, ε1,ε2,ε3,ε4It is characterized matrix coefficient；
(9) classification results of monolithic plant leaf blade are determined using svm classifier algorithm, repeat step (6)~(9) are treated The results set E of individual blade after all segmentations of plant leaf blade specimen of detection；
(10) occur most results in statistics set E, and be used as the classification of final plant leaf blade specimen to be detected As a result, if occurring multiple results in set E, plant leaf blade specimen to be detected is subjected to artificial discrimination processing；
Realize the classification to multiple plant leaf blade specimens in repeat step (2)~(10).
It is special with geometry that the plant specimen classification results that the detection method of present embodiment is drawn merge multiple dimensioned direction texture Levy, it is credible with higher reliability.
This plant specimen categorizing system includes image acquisition units, graphics processing unit and classification results display unit.
Image acquisition units, for gathering plant specimen image to be measured in real time；Industrial camera, camera lens and illumination can be used Light source is realized.Industrial camera uses Daheng MER-125-30UC colour 1/3Sony ICX445 ccd video cameras, full frame exposure Scan mode, resolution ratio is 1292 (H) × 964 (V), and frame per second is up to 30 frames/second, and output interface is Mini USB2.0 interfaces, mirror Head bayonet socket is C mouthfuls, compact, it is easy to install, and can meet detection in real time and require；Camera lens from Japan Computar 8mm without Distort camera lens.Light source selects 92 × 20 × 20mm LDL2-80X16SW strip sources, and light source controller selects Daheng CC-ST- 1024 model C C series LED light source controllers.
Graphics processing unit, for being pre-processed to described plant leaf blade specimen image, ROI extract, segmentation, feature Extract, and calculate classification results；Graphics processing unit hardware can use industrial control computer, and core is the meter write Calculation machine software.Specific works mode is：
(i) image for gathering image acquisition units carries out image gray processing, binaryzation；
(ii) ROI extractions are carried out to the plant specimen image after binaryzation using morphological operation；
(iii) image segmentation is carried out to ROI region, obtains the set of plant leaf blade；
(iv) its texture and geometric properties is extracted for each blade；
(v) all blade classification results are calculated with sorting algorithm, and will appear from the most conduct of number of times and finally recognize knot Really；
Described graphics processing unit realizes the classification of plant leaf blade specimen by software programming and above-mentioned sorting algorithm. Human interface software can also be included in described image processing unit, for showing described plant leaf blade specimen to be sorted in real time Image and classification results, record sort historical data, and the operational order of user is received to software progress parameter setting.It is actual During shown by display screen.Industrial control computer connects industrial camera and light source controller by USB data line.Work Industry control computer is using grinding magnificent industrial control computer, and the machine uses Intel dual core processors, dominant frequency 3.0GHz, internal memory 2G, Hard disk 160G, 19 cun of liquid crystal displays meet the requirement of industry spot adverse circumstances.
Testing result display unit, actually may be selected the display being connected with industrial control computer, coordinates what is write Plant leaf blade specimen categorizing system software, can intuitively display image processing unit classification results.

Claims (9)

1. a kind of plant specimen sorting technique for merging multiple dimensioned direction textural characteristics, it is characterised in that including：
(1) plant leaf blade specimen is collected in advance, and different plant leaf blade specimens are made a distinction, and plant leaf blade specimen is classified as several Typical categories are planted, the image of the individual blade of quantity is set to collection of specimens of all categories, Sample Storehouse is set up；
(2) the plant specimen image of plant specimen to be detected is gathered；
(3) image preprocessing is carried out to the plant specimen image collected；
(4) to the plant specimen image zooming-out area-of-interest after pretreatment；
(5) image segmentation is carried out for area-of-interest, obtains including the image collection of multiple single blade images；
(6) the geometric properties matrix of a leaf image in image collection is calculated；
(7) the textural characteristics matrix of current vane image is calculated；
(8) obtained geometric properties matrix and the eigenmatrix of textural characteristics matrix computations current vane image is utilized；
(9) eigenmatrix of the eigenmatrix of current vane image and the image of Sample Storehouse is utilized into support vector cassification algorithm Classification, obtains the identification classification results of current vane image；
(10) repeat step (6)-(9), classification is identified to leaf image all in image collection, obtains current plant mark The most result of occurrence number dividing as final plant leaf blade specimen in this classification results set, selection sort results set Class result.
2. the plant specimen sorting technique of the multiple dimensioned direction textural characteristics of fusion according to claim 1, it is characterised in that In step (4), the step of extracting area-of-interest includes：
(4-1) extracts the connected domain in the plant specimen image after pretreatment according to eight directional chain-codes；
(4-2) is screened according to length and width, the area of the minimum enclosed rectangle of connected domain, rejects delimbing in plant specimen image Garbage beyond leaf；
(4-3) carries out morphological operation to the connected domain extracted：Opening operation, eliminates the branch part in plant specimen, obtains Area-of-interest.
3. the plant specimen sorting technique of the multiple dimensioned direction textural characteristics of fusion according to claim 1, it is characterised in that In step (5), leaf image segmentation is carried out to area-of-interest using the minimum enclosed rectangle of connected domain.
4. the plant specimen sorting technique of the multiple dimensioned direction textural characteristics of fusion according to claim 1, it is characterised in that In step (6), the geometric properties matrix is obtained by following methods：The rectangular degree of calculating current vane image, circularity, convexity three The geometric properties of individual invariable rotary, and then obtain geometric properties matrix.
5. the plant specimen sorting technique of the multiple dimensioned direction textural characteristics of fusion according to claim 1, it is characterised in that In step (7), the method for calculating the textural characteristics matrix of current vane image is as follows：
(7-1) carries out gaussian filtering to current vane image；
(7-2) is directed in the leaf image after gaussian filtering each pixel, calculates the gradient information of each pixel；
(7-3) determines bins bar numbers, using the gradient magnitude of pixel as weight, calculates the leaf image after gaussian filtering Amplitude-gradient orientation histogram；
Peak value highest bins in amplitude-gradient orientation histogram as the main direction of plant leaf blade, is denoted as φ by (7-4)；
(7-5) is calculated under different scale respectively, and the textural characteristics of main direction φ local binary patterns finally give and work as frontal lobe The textural characteristics matrix of picture.
6. the plant specimen sorting technique of the multiple dimensioned direction textural characteristics of fusion according to claim 5, it is characterised in that In step (7-1), selection (2k+1) × (2k+1) templates, wherein k=1,2,3, using using two-dimensional space Gauss discrete function Filtering Template is calculated, then current vane image is filtered using obtained Filtering Template.
7. the plant specimen sorting technique of the multiple dimensioned direction textural characteristics of fusion according to claim 5, step (7-5) In, wherein:
Main direction φ circular operator is：
x p = x c + R cos ( 2 π p P + φ )
y p = y c - R s i n ( 2 π p P + φ )
The texture eigenvalue of the local binary patterns of each pixel is calculated by following formula and obtained in leaf image：
LBP P , R ( x p , y p ) = Σ p = 0 p - 1 s ( g p - g c ) 2 p
s ( x ) = 1 , x &GreaterEqual; 0 0 , x < 0
Wherein：xc、ycIt is the transverse and longitudinal coordinate of central pixel point, xp、ypIt is the transverse and longitudinal coordinate of sampled point；P is sampled point number, p= 0 ..., P-1, R be sample radius, gcCentered on pixel gray value, gpFor the gray value of sampled point.
8. a kind of plant specimen categorizing system for merging multiple dimensioned direction textural characteristics, it is characterised in that including：
Image acquisition units, the plant specimen image for gathering plant specimen to be measured；
Graphics processing unit, for being pre-processed to described plant specimen image, region of interesting extraction, image segmentation, Feature extraction, and calculate classification results；
Testing result display unit, for showing classification results.
9. the plant specimen categorizing system of the multiple dimensioned direction textural characteristics of fusion according to claim 8, it is characterised in that Described image collecting unit includes industrial camera, camera lens and lighting source；Described image processing unit is computer；The inspection It is the display being connected with computer to survey result display unit.
CN201710200865.6A 2017-03-30 2017-03-30 Merge the plant specimen sorting technique and system of multiple dimensioned direction textural characteristics Pending CN106991441A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710200865.6A CN106991441A (en) 2017-03-30 2017-03-30 Merge the plant specimen sorting technique and system of multiple dimensioned direction textural characteristics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710200865.6A CN106991441A (en) 2017-03-30 2017-03-30 Merge the plant specimen sorting technique and system of multiple dimensioned direction textural characteristics

Publications (1)

Publication Number Publication Date
CN106991441A true CN106991441A (en) 2017-07-28

Family

ID=59411940

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710200865.6A Pending CN106991441A (en) 2017-03-30 2017-03-30 Merge the plant specimen sorting technique and system of multiple dimensioned direction textural characteristics

Country Status (1)

Country Link
CN (1) CN106991441A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109712135A (en) * 2018-12-28 2019-05-03 南京林业大学 A kind of fruit defects recognition methods, device and its storage medium
CN110120054A (en) * 2019-05-15 2019-08-13 深圳铂石空间科技有限公司 Automatic counting method and device, medium, electronic equipment based on image procossing
CN113610187A (en) * 2021-08-19 2021-11-05 江西应用技术职业学院 Wood texture extraction and classification method based on image technology

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103295238A (en) * 2013-06-03 2013-09-11 南京信息工程大学 ROI (region of interest) motion detection based real-time video positioning method for Android platform
CN105404853A (en) * 2015-10-29 2016-03-16 江南大学 Content-based plant leaf online recognition system
CN106295661A (en) * 2016-08-15 2017-01-04 北京林业大学 The plant species identification method of leaf image multiple features fusion and device

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103295238A (en) * 2013-06-03 2013-09-11 南京信息工程大学 ROI (region of interest) motion detection based real-time video positioning method for Android platform
CN105404853A (en) * 2015-10-29 2016-03-16 江南大学 Content-based plant leaf online recognition system
CN106295661A (en) * 2016-08-15 2017-01-04 北京林业大学 The plant species identification method of leaf image multiple features fusion and device

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
REZA DAVARZANI 等: "Robust Image Description with Weighted and Adaptive Local Binary Pattern Features", 《2014 22ND INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION》 *
张宁 等: "基于图像分析的植物叶片识别技术综述", 《计算机应用研究》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109712135A (en) * 2018-12-28 2019-05-03 南京林业大学 A kind of fruit defects recognition methods, device and its storage medium
CN110120054A (en) * 2019-05-15 2019-08-13 深圳铂石空间科技有限公司 Automatic counting method and device, medium, electronic equipment based on image procossing
CN113610187A (en) * 2021-08-19 2021-11-05 江西应用技术职业学院 Wood texture extraction and classification method based on image technology
CN113610187B (en) * 2021-08-19 2022-02-01 江西应用技术职业学院 Wood texture extraction and classification method based on image technology

Similar Documents

Publication Publication Date Title
CN107016405B (en) A kind of pest image classification method based on classification prediction convolutional neural networks
Ke et al. A review of methods for automatic individual tree-crown detection and delineation from passive remote sensing
CN109636772A (en) The defect inspection method on the irregular shape intermetallic composite coating surface based on deep learning
CN106056118A (en) Recognition and counting method for cells
CN108355987A (en) A kind of screen printing of battery quality determining method based on piecemeal template matches
CN107230203A (en) Casting defect recognition methods based on human eye vision attention mechanism
CN112232263B (en) Tomato identification method based on deep learning
CN108090906A (en) A kind of uterine neck image processing method and device based on region nomination
CN104680130A (en) Chinese character recognition method for identification cards
Lu et al. Detection of urban expansion in an urban-rural landscape with multitemporal QuickBird images
CN102324032A (en) Texture feature extraction method for gray level co-occurrence matrix in polar coordinate system
CN107292314A (en) A kind of lepidopterous insects species automatic identification method based on CNN
Iovan et al. Detection, characterization, and modeling vegetation in urban areas from high-resolution aerial imagery
CN112634243A (en) Image classification and recognition system based on deep learning under strong interference factors
CN106991441A (en) Merge the plant specimen sorting technique and system of multiple dimensioned direction textural characteristics
CN107256398A (en) The milk cow individual discrimination method of feature based fusion
CN107240112A (en) Individual X Angular Point Extracting Methods under a kind of complex scene
Zhang et al. A pointer meter recognition algorithm based on deep learning
CN107657220A (en) A kind of leukorrhea mould automatic testing method based on HOG features and SVM
CN114140665A (en) Dense small target detection method based on improved YOLOv5
CN110110667A (en) A kind of processing method, system and the associated component of diatom image
CN113435460A (en) Method for identifying brilliant particle limestone image
Pourakpour et al. Automated mitosis detection based on combination of effective textural and morphological features from breast cancer histology slide images
US11488401B2 (en) Prostate cancer tissue image classification with deep learning
He et al. Progress of machine vision in the detection of cancer cells in histopathology

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: 20170728