CN108855988A - Walnut kernel stage division and walnut kernel grading plant based on machine vision - Google Patents

Walnut kernel stage division and walnut kernel grading plant based on machine vision Download PDF

Info

Publication number
CN108855988A
CN108855988A CN201810427065.2A CN201810427065A CN108855988A CN 108855988 A CN108855988 A CN 108855988A CN 201810427065 A CN201810427065 A CN 201810427065A CN 108855988 A CN108855988 A CN 108855988A
Authority
CN
China
Prior art keywords
walnut kernel
image
classification
target
color
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.)
Granted
Application number
CN201810427065.2A
Other languages
Chinese (zh)
Other versions
CN108855988B (en
Inventor
周军
郭俊先
张静
姜彦武
蔡建
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.)
Xinjiang Agricultural University
Original Assignee
Xinjiang Agricultural University
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 Xinjiang Agricultural University filed Critical Xinjiang Agricultural University
Priority to CN201810427065.2A priority Critical patent/CN108855988B/en
Publication of CN108855988A publication Critical patent/CN108855988A/en
Application granted granted Critical
Publication of CN108855988B publication Critical patent/CN108855988B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/34—Sorting according to other particular properties
    • B07C5/342—Sorting according to other particular properties according to optical properties, e.g. colour
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/02—Measures preceding sorting, e.g. arranging articles in a stream orientating
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/36—Sorting apparatus characterised by the means used for distribution
    • B07C5/361—Processing or control devices therefor, e.g. escort memory
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • 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
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • 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
    • G06V10/56—Extraction of image or video features relating to colour
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Abstract

The present invention relates to the classification tracer technique fields based on machine vision, are a kind of walnut kernel stage divisions and walnut kernel grading plant based on machine vision；The former includes the following steps, is successively to obtain historical sample data, image procossing, image segmentation, establishes primitive character matrix, the selection of feature, the foundation and classification tracking of model；Structure of the invention is reasonable and compact, easy to use；After acquiring walnut kernel image by machine vision, classification is tracked to walnut kernel on PC host, after walnut kernel reaches classification position, walnut kernel is blown into classification blowpit by way of air-blowing, to realize that walnut kernel is classified；Wherein, using feature b in19, K1And bin15, and using the optimal Naive Bayes Classification method of classifying quality as walnut kernel hierarchy model；It is fast to be classified processing speed, can realize automatic classification for dynamic walnut kernel on delivery platform, instead of manual grading skill, save labour.

Description

Walnut kernel stage division and walnut kernel grading plant based on machine vision
Technical field
The present invention relates to based on machine vision walnut kernel grading plant and method and technology field, be a kind of to be regarded based on machine The walnut kernel stage division and walnut kernel grading plant of feel.
Background technique
Walnut is also known as English walnut, Qiang peach etc., originates in the Central Asia, now cultivates and be distributed in extensively Yunnan Province of China, Shanxi, Xinjiang etc. 24 provinces and regions, area occupy first place in the world with yield, and the walnut yield to the end of the year 2015, Xinjiang has reached 560,000 tons or so.Core Peach kernel is full of nutrition, and containing a large amount of phosphatide, protein and vitamin, fat content is up to 60% or more, with higher Economic value and nutritive value.It is intermediate generally to pass through the works such as de- green peel, rinsing, drying to storage after the picking of walnut maturation Sequence, every procedure may all influence the quality of walnut, and may occur due to its higher fat content, during storage Go rotten, insect pest and it is rotten phenomena such as.Simultaneously as walnut shell-breaking technology is still immature at present, mechanical shell-cracking kernel-taking mode Also it be easy to cause the fragmentation of walnut kernel.Generally the quality of identification walnut kernel is mainly based on observation on the market, and benevolence clothing color is with Huang It is white to be upper, it is dark yellow be it is secondary, brown Huang is more secondary, and the critical deterioration that bellding, color are dark brown, cannot eat.State Administration of Forestry's publication 《People's Republic of China's forestry industry standard》In phase has been done to the technical requirements, the method for inspection and detected rule etc. of walnut kernel Regulation is closed, wherein the quality of walnut kernel is main other than the basic demands such as smell, moisture and various bacterium colony contents needs are up to standard It is classified according to the standard of table 1-1.
Walnut kernel quality classification standard regulation, the walnut kernel that color detection and integrity degree are greater than four points of benevolence, which is detected, to be needed Method is inspected using artificial range estimation to be classified.There is detection, sorting large labor intensity in artificial treatment, low efficiency is at high cost, simultaneously It is also easy to be influenced by the subjective factor of people, there is a situation where that artificial subjective factor causes effectiveness of classification unstable, poor accuracy etc. Shortcomings；
See to the present Research that detection of agricultural products is classified, applies to detection of agricultural products in machine vision technique both at home and abroad Technically there is more research, but detection is rested on mostly to the detection of agricultural product with mechanical vision inspection technology It is especially domestic in static agricultural product individual images, the research of agricultural product dynamic classification is still short of, related automatic classification System is less, and not yet establishes for the automatic grading system of the color of walnut kernel and integrity degree at present, the classification to walnut kernel It mainly or by the mode of manual grading skill carries out, large labor intensity, classification efficiency is lower, so carrying out certainly for walnut kernel The research of dynamic classification aspect is necessary.
Summary of the invention
The present invention provides a kind of walnut kernel stage division and walnut kernel grading plant based on machine vision, overcomes The problem of deficiency for stating the prior art can effectively solve artificial range estimation and inspect method to be classified walnut kernel, large labor intensity.
Technical solution of the present invention is first is that realized by following measures：A kind of walnut kernel classification based on machine vision Method, its step are as follows：
The first step：Obtain historical sample data：By manual grading skill, the walnut kernel sample graph of two or more grades is obtained As each 30,30 walnut kernel samples are placed on delivery platform, are obtained by the machine vision equipment on delivery platform every A walnut kernel image；
Second step：Image procossing：The walnut kernel image of acquisition is handled, the complete image of walnut kernel profile is obtained；
Third step：Image segmentation：Walnut kernel image is divided using minimum circumscribed rectangle, obtains and only contains single walnut kernel Image；
4th step：Historical sample data is handled, primitive character matrix is established：
(I), primitive character of 37 features related with walnut kernel color of image as walnut kernel color grading, tool are obtained Body includes obtaining 31 color characteristics based on hue histogram, and 31 features are bin0 to bin30 respectively；Be based on color moment 6 features are obtained, 6 features are H respectivelyμ、Hv、Hs、Sμ、SvAnd Ss；
(II), 9 features related with walnut kernel shape integrity degree are obtained, the minimum external square of walnut kernel profile is specifically included Shape length-width ratio K1, walnut kernel contour area and its minimum circumscribed circle contour area ratio K2With 7 Hu moment characteristics, 7 Hu moment characteristics It is I respectively1、I2、I3、I4、I5、I6And I7；
(III), primitive character matrix is established according to features described above, wherein one sample of each behavior, each to be classified as a spy Sign, each column are K respectively from left to right1、K2、I1、I2、I3、I4、I5、I6、I7、Hμ、Hv、Hs、Sμ、Sv、Ss、 bin0、bin1、 bin2、bin3、bin4、bin5、bin6、bin7、bin8、bin9、bin10、bin11、bin12、 bin13、bin14、bin15、 bin16、bin17、bin18、bin19、bin20、bin21、bin22、bin23、 bin24、bin25、bin26、bin27、 Bin28, bin29 and bin30；
(IV), primitive character matrix is normalized, i.e., by normalized function to by the numerical value of each column Linear Mapping is between [0,1] section；
5th step：The selection of feature：
One, based on the Feature Selection of ReliefF：Primitive character matrix is subjected to feature selecting fortune with ReliefF algorithm It calculates, removes Hu moment characteristics I3、I4、I5、I6、I7；
Two, the feature selecting of two kinds of algorithms of the information gap MID based on mRMR and mutual information MIQ：
1. using mRMR information gap MID algorithm to the eigenmatrix after the Feature Selection based on ReliefF algorithm into Row feature selecting, takes preceding 15 features of algorithms selection, and 15 features are followed successively by bin19, K1、bin15、bin20、K2、 bin13、bin16、bin12、bin18、bin14、bin17、Hv,bin11,bin21,bin15；
2. using mRMR mutual information MIQ algorithm to the eigenmatrix after the Feature Selection based on ReliefF algorithm into Row feature selecting, takes preceding 15 features of algorithms selection, and 15 features are followed successively by bin19, K1、bin15、bin16、bin13、 I2、Sv、K2, bin17, bin12, bin18, bin14, bin5, bin25 and bin20；
6th step：The foundation of model：15 obtained respectively in two criterion of mutual information difference MID and mutual information quotient MIQ In feature, preceding 3 features, i.e. bin19, K are taken respectively1And bin15, using these three features to Naive Bayes Classification Model It is trained, obtains walnut kernel hierarchy model；
7th step：Classification tracking：Classification tracking is carried out to walnut kernel to be measured according to model, it is true according to the size of output probability Determine walnut kernel rank.
Here is the further optimization and/or improvements to invention technology described above scheme one：
Above-mentioned 7th step carries out classification tracking to walnut kernel to be measured according to model, determines walnut according to the size of output probability Specific step is as follows for benevolence rank：
(1) to the detection of target：
1, it obtains walnut kernel image to be fractionated and image is handled using method identical with second step image procossing, Obtain the complete walnut kernel target image of profile；
2, it determines whether there is new walnut kernel profile and enters visual field, specific deterministic process is as follows：
A, enter the direction of visual field, the conveying direction from left to right of delivery platform, using boundary rectangle based on walnut kernel Frame chooses each walnut kernel profile in image respectively, by the most left extreme point abscissa of leftmost side walnut kernel boundary rectangle frame with it is upper The most left extreme point abscissa of leftmost side walnut kernel boundary rectangle frame compares in one frame image, if being less than, enters step b； If more than not being then new walnut kernel profile, then enter step c；
B, each walnut kernel profile in image is chosen using boundary rectangle frame respectively, obtains leftmost side walnut kernel boundary rectangle The most left extreme point abscissa of frame, judges whether the most left extreme point abscissa is less than m pixel, and m pixel is range image left edge Image range value；If being less than, it is determined with new walnut kernel profile and enters visual field, using this profile boundary rectangle as initially searching Rope frame and the hue histogram for calculating original image pixels in walnut kernel profile in search box, the two is stored, addition one A target to be tracked, enters step (two)；If more than not being then new walnut kernel profile, enter step c；
C, judging that the deletion of walnut kernel target in image marks whether is very, if true, then it is assumed that the walnut kernel is Classification is completed, the tracking target is deleted；If not true, the window's position of tracking target is calculated using method for tracking target；According to The position of the tracking window center of tracked target judges the classification position whether walnut kernel target reaches, if reaching, to control Device processed sends graded signal to the controller for being used to control walnut kernel delivery device, makes controller action, realizes point of walnut kernel Grade operation, while it is true for setting this target deleted marker to be tracked, and deletes this tracking target；If not reaching, continue to track mesh Mark then repeats to judge to track whether window center reaches classification position；
(2) classification and tracking for treating tracking target judge the grade and trace location of target to be tracked, realize target Progressive operation.
Above-mentioned (two) step treats the classification and tracking of tracking target, judges the grade and trace location of target to be tracked, Realize that specific step is as follows for target progressive operation：
(1) target following is utilized using the search box and hue histogram stored as input quantity to each target to be tracked Method calculates the window's position of tracking target；Judge whether the tracking window center of target to be tracked reaches the feature of setting and mention Fetch bit is set, if reaching feature extraction position enters (2) step；If not reaching feature extraction position, repeat to judge mesh to be tracked Whether mark reaches feature extraction position；
(2) walnut image is divided using minimum circumscribed rectangle, obtains the image for only containing single walnut kernel；
(3) judge that target to be tracked is with the presence or absence of grade in image：Extract bin19, K of each target to be tracked1With Tri- features of bin15 export each grade probability as input, using walnut kernel hierarchy model, and grade maximum probability is thus Track the affiliated rank of general objective；Grade if it exists then enters (4) step；Grade is such as not present, then terminates to track；
(4) defined area is carried out to each classification position, is judged according to the location of each tracking window center wait chase after Whether the target of track reaches classification position, if reaching, sends graded signal to being used to control walnut kernel pusher to controller The controller of structure, makes controller action, realizes the progressive operation of walnut kernel, at the same set this target deleted marker to be tracked be it is true, And delete this tracking target；If not reaching, repetition judges whether the tracking window center of target to be tracked reaches classification position.
In above-mentioned 4th step, (I) small step：37 features related with walnut kernel color of image are obtained as walnut kernel The primitive character of color grading, specifically includes and obtains 31 color characteristics based on hue histogram, and 31 features are respectively bin0…bin30；6 features are obtained with based on color moment, 6 features are the specific step of H μ, Hv, Hs, S μ, Sv and Ss respectively It is rapid as follows：
I, color space conversion；RGB color used in history image is converted to and meets human eye to color subjectivity The hsv color space of understanding；
II, color feature extracted, to by the walnut kernel image in I treated image using color histogram method and Color Moment Methods carry out color feature extracted；
(i) color histogram method extracts：The hue histogram that walnut kernel tone is distributed in history image is established, from tone The main distribution of tone of each grade walnut kernel is obtained in histogram, extracts corresponding 31 colors of the main distribution of tone Frequency modulation rate forms 31 tone characteristics of slave bin0 to the bin30 based on color histogram as tone characteristics；
(ii) color Moment Methods extract：The tone H and saturation degree S for selecting image obtain three low-order moments according to color moment, Three low-order moments be respectively first moment (mean value) H μ and S μ, second moment (variance) Hv and Sv and third moment (degree of bias) Hs and Ss；H μ, Hv, Hs, S μ, Sv, Ss composition express 6 features of distribution of color.
Specific step is as follows for above-mentioned second step image procossing：
(a) to the image of acquisition, binary conversion treatment is carried out, obtains the single channel image of the relatively sharp image of profile；
(b) denoising is carried out to by the single channel image after binary conversion treatment using morphological image method；
(c) the outermost layer profile of each walnut kernel in image is found, and the walnut kernel profile found is arranged using bubbling Name placement of the sequence method to walnut kernel profile；
(d) conveying direction from left to right based on delivery platform obtains the most left extreme point of walnut kernel outermost layer profile Coordinate, judging the extreme point, whether image left edge coincides, if being overlapped, is determined as imperfect walnut kernel profile, from profile This profile is removed in sequence；If not being overlapped, it is determined as complete walnut kernel profile, retains the walnut kernel image.
Technical solution of the present invention is second is that realized by following measures：A kind of walnut kernel of the use based on machine vision The walnut kernel grading plant of stage division and walnut kernel grading plant, including mobile platform, machine vision equipment, can be used to control Controller, delivery device and the classification blowpit of delivery device processed；Machine vision equipment, machine view are housed on a mobile platform Feel that equipment includes light box, camera and PC host；Camera is located in light box, and is electrically connected with PC host；PC host and the control Device electrical connection processed；Light box is fixedly mounted on the paralell upper end of mobile platform；The left side being equipped on light box along mobile platform direction Side entrance and right-side outlet；Edge is between left and right every fixing at least two classification blowpits on rear side of mobile platform in light box, in light The delivery device that the walnut kernel after classification can be pushed to prescribed fractionated blowpit is fixedly installed on front side of mobile platform in case； The controller is electrically connected with delivery device；The camera is for acquiring walnut kernel image；The PC host is for adopting camera The walnut kernel image collected is handled, and is classified and is tracked；The controller controls push for receiving PC host signal Mechanism realization pushes to the walnut kernel after classification in prescribed fractionated blowpit.
Here is the further optimization and/or improvements to invention technology described above scheme two：
Above-mentioned delivery device uses jet delivery device；Jet delivery device includes air compressor, solenoid valve, jet Pipe；It is fixed with air jet pipe corresponding with classification blowpit on front side of mobile platform in light box, is connected to and installs in air jet pipe upper end Having can be towards the spray head of corresponding classification blowpit；The lower end of every air jet pipe is connect with the air compressor respectively；? The solenoid valve for controlling air jet pipe on-off is fixedly installed on every air jet pipe, controller is electrically connected with each solenoid valve respectively It connects.
Above controller uses PLC controller or single-chip microcontroller, and controller electricity is connected to timer.
Above-mentioned delivery platform uses frequency conversion delivery platform, and frequency conversion delivery platform includes frequency converter, frequency converter and PC host electricity Connection.
Above-mentioned frequency converter is connect with PC host by RS485 bus communication.
Edge is corresponding with blowpit is classified between left and right every fixing four classification blowpits on rear side of mobile platform in above-mentioned light box Mobile platform on front side of be fixedly mounted there are four air jet pipe.
Structure of the invention is reasonable and compact, easy to use；After acquiring walnut kernel image by machine vision, on PC host Classification is tracked to walnut kernel, after walnut kernel reaches classification position, walnut kernel is blown into classification by way of air-blowing In blowpit, to realize that walnut kernel is classified；Wherein, using feature b in19, K1And bin15, and it is best using classifying quality Naive Bayes Classification method as walnut kernel hierarchy model；It is fast to be classified processing speed, it can be dynamic on delivery platform Walnut kernel realizes automatic classification, instead of manual grading skill, saves labour；
Except this, walnut kernel RGB image is converted into hsv color space due to using, it is more smart to the judgement of color The accuracy rate of standard, the feature of extraction is higher, improves classification quality.
Detailed description of the invention
Attached drawing 1 is the walnut kernel classification process line map in the embodiment of the present invention 1；
Attached drawing 2 is walnut kernel target detection and tracking thread figure in the embodiment of the present invention 1；
Attached drawing 3 is the classification of walnut kernel target and tracking thread figure in the embodiment of the present invention 1；
Attached drawing 4 is image procossing thread figure in the embodiment of the present invention 1；
Attached drawing 5 is that mRAR MIQ algorithm characteristics select sequential test result figure in the embodiment of the present invention 1；
Attached drawing 6 is color space model in the embodiment of the present invention 1；
Attached drawing 7 is that target tracking is classified schematic diagram in the embodiment of the present invention 1.
Attached drawing 8 is the shape feature figure of walnut kernel in the embodiment of the present invention 1；
Attached drawing 9 is feature K in the embodiment of the present invention 11Walnut kernel sample scatter plot；
Attached drawing 10 is feature K in the embodiment of the present invention 12Walnut kernel sample scatter plot；
Attached drawing 11 is the feature weight distribution map of ReliefF feature selecting in the embodiment of the present invention 1；
Attached drawing 12 is that mRAR MID algorithm characteristics select sequential test result figure in the embodiment of the present invention 1；
Attached drawing 13 is the structural schematic diagram of the embodiment of the present invention 2；
Coding in attached drawing is respectively：1 is mobile platform, and 2 be light box, and 3 be camera, and 4 be PC host, and 5 be controller, 6 It is right-side outlet for left hand inlet port, 7,8 be frequency converter, and 9 be air jet pipe, and 10 be air compressor, and 11 be solenoid valve.
Specific embodiment
The present invention is not limited by the following examples, can determine according to the technique and scheme of the present invention with actual conditions specific Embodiment.
Below with reference to embodiment 1 and attached drawing 1, to attached drawing 4 and attached drawing 6 to attached drawing 13, the invention will be further described：
Embodiment 1：As shown in Fig. 1, the walnut kernel hierarchy model construction method of machine vision classifying equipoment, step is such as Under：
The first step：As shown in Fig. 1, historical sample data is obtained, concrete operations are as follows：
By manual grading skill, five grade (such as table 2-1 walnut kernel outside level-one, second level, three-level, level Four and grade are obtained Grade) each 30 historical datas of walnut kernel sample；30 × 5 150 walnut kernel samples are sent by delivery platform to machine Visual apparatus；Wherein, need to carry out white balance correction by the camera of machine vision equipment, can initial option blue as phase Machine obtains the background colour of single image, is capable of increasing color of object and northern backcolor gap in this way, increases segmentation effect；
Second step：As shown in attached drawing 1,4, image procossing：
(a) increase the difference of background and target image by the subtraction in channel B and the channel R acquisition single channel image, it Image is handled using thresholding afterwards, completes the binary conversion treatment to image, threshold value may be set to 20 here；To image Being handled using thresholding can achieve preferable effect；Wherein, image effect when fixed threshold is 20 is best；
The above process can be separated image channel using the function that multichannel image separates, then lead under OpenCV mode Cross the functional operation acquisition for channel subtraction；
(b) denoising is carried out to by the single channel image after binary conversion treatment using morphological image method；
Due to the image after binary conversion treatment, foreground area the phenomenon that there are holes, and also due to making an uproar in background area There are white noises for sound or irrelevant part (dry walnut kernel is also easy to produce disintegrating slag)；Especially for the walnut kernel outside equal, by It is complex in its color, more hole is generated in walnut kernel profile, can generate interference to subsequent image processing operation； Therefore, it is necessary to carry out denoising by closing operation of mathematical morphology, while the hole in walnut kernel region can also be eliminated；Image Morphology is a kind of image processing method for effectively solving noise, segmentation individual picture elements and connection adjacent element.Form Learning closed operation can be used to eliminate the purely part as caused by noise, and opening operation is used to connect neighbouring region, and the two can be improved The robustness of image segmentation, although morphology opening operation or closed operation are similar with the result of corrosion and expansion, the two operations It is generally more general, it more can accurately save source images join domain；
Operation can be carried out by morphology operations function, select the binary map with hole and noise under OpenCV mode Picture selects 5 × 5 oval forming core (oval forming core keeps contour edge more smooth after tested) to do closing operation of mathematical morphology and open fortune It calculates, effectively removes lesser noise, and the hole in walnut kernel region can be eliminated.
(c) the outermost layer profile of each walnut kernel in image is found, and the walnut kernel profile found is arranged using bubbling Name placement of the sequence method to walnut kernel profile；
(d) conveying direction from left to right based on delivery platform, obtain walnut kernel outermost layer profile in the picture Most left extreme point coordinate, judging the extreme point, whether image left edge coincides, if being overlapped, is determined as imperfect walnut kernel Profile removes this profile from profile sequence；If not being overlapped, then it is determined as complete walnut kernel profile, retains the walnut kernel figure Picture；If the conveying direction from right to left based on delivery platform, in the picture most right of walnut kernel outermost layer profile is obtained Extreme point coordinate, judgment method are identical；
Due in Image Acquisition, in fact it could happen that be located at the incomplete walnut kernel profile of boundary；Especially walnut kernel profile When into viewing field of camera, if being extracted using imperfect walnut kernel profile as single walnut kernel, will cause feature extraction it is imperfect and Higher mistake divides rate；Therefore, increase mistake point rate to avoid incomplete walnut kernel profile in figure, by obtaining walnut kernel most The most left extreme point coordinate of outermost contour judges whether the method that image left edge coincides is determined image the extreme point The complete situation of middle walnut kernel profile, if being overlapped, prove the walnut kernel profile of walnut kernel in the edge of viewing field of camera, And incomplete state is presented in the picture, it should reject at this time；To improve the accuracy for guaranteeing feature extraction.
Third step：Image segmentation：
The complete walnut kernel bianry image of profile can be obtained after removing imperfect profile, using it as exposure mask, remove image After background, rectangle segmentation is realized to image using thresholding method, obtains the image for only containing single walnut kernel；
Can be under OpenCV mode, pattern mask processing, and using the function for seeking the minimum circumscribed rectangle of profile, it will Walnut kernel profile in image carries out minimum circumscribed rectangle segmentation, obtains the image for only containing single walnut kernel；
4th step：As shown in Fig. 1, historical sample data is handled, obtains the eigenmatrix of 30 × 5 samples；
I, color space conversion；RGB color used in history image is converted to and meets human eye to color subjectivity The hsv color space of understanding；
Rgb space is converted to the hsv color space for meeting human eye to color subjective understanding, and mainly in hsv color sky Between carry out color characteristic extraction, can be improved the precision of color feature extracted, thus improve walnut kernel is classified it is accurate Degree；
It is as follows that RGB color is converted into hsv color space arithmetic：
The value (r, g, b) of RGB color is first given, wherein r, g, b ∈ [0,255], if m=max (r, g, b), n= Min (r, g, b), the then value (r, g, b) for determining RGB color are transformed into the h of HSV space, s, and v value-based algorithm is：
If (h < 0) h=h+360
Here [0,360] h ∈, s ∈ [0,1], v ∈ [0,1]
The above process can carry out under OpenCV mode, i.e., carry out image using the function for the conversion that can be used for image space Spatial transformation, and the isolated function by that can be used for Color Channel carries out the separation of Color Channel；
II, color feature extracted, the extraction to color characteristic is carried out by I treated walnut kernel image；Color characteristic It is extracted using color histogram method and color Moment Methods；
(i) color histogram method extracts：Color histogram describes different color shared ratio in the picture, has figure As scaling and rotational invariance, i.e., had good robustness when inconsistent for position under walnut kernel dynamic classification；By attached drawing Shown in 6, tone is distributed on 0 ° to 360 ° of circle；It is 180 equal portions (bin) by tone equal interval quantizing, randomly selects respectively Part faint yellow benevolence, light amber benevolence and outer benevolence is waited to make hue histogram, the ordinate of hue histogram is corresponding tone zone Between frequency shared by pixel, according to hue histogram extract 0 to 30 totally 31 tone frequencies as a part of tone characteristics, group At 31 tone characteristics of slave bin0 to the bin30 based on color histogram.
(ii) color Moment Methods extract：Color moment is a kind of simple and effective color characteristic representation method.Due to image Colouring information is mainly distributed in low-order moment, so its first moment (mean value), second moment (variance) and third moment (degree of bias) foot To express color of image distribution.The mathematical formulae of three color moments is as follows：
Mean value：
Variance：
The degree of bias：
Wherein pi,jIndicate the value of i-th of channel pixel j of color image；To reduce light source variation to image characteristics extraction It influences, selects the tone H and saturation degree S of image herein, calculate separately three of them low-order moment Hμ、Hv、HsAnd Sμ、Sv、SsForm table Up to 6 features of distribution of color；
The above process can carry out under OpenCV mode, i.e., obtained using the function for the mean value and standard deviation that can acquire image The mean value and standard deviation of image, the function for reusing the degree of bias that can obtain image obtain the degree of bias of image；
III, related with walnut kernel shape integrity degree feature is obtained, it, can the characteristics of by analysis different integrity degree walnut kernel It is walnut kernel contour area and its minimum circumscribed circle area respectively to choose the difference of two characteristic present walnut kernel integrity degrees Ratio K1And the minimum extraneous rectangular aspect ratio K of profile2；
1, the ratio K of walnut kernel contour area and its minimum circumscribed circle area is obtained1：As shown in attached drawing 8, walnut kernel profile Minimum circumscribed rectangle length-width ratio is:
Wherein, l- is the length of peach kernel profile minimum circumscribed rectangle；W- is the width of peach kernel profile minimum circumscribed rectangle；
The above process can be in OpenCV mode, and the function by calculating walnut kernel profile minimum circumscribed rectangle length-width ratio is asked ?；Here 10 walnut kernel be can use, be K with ordinate1, abscissa is that walnut kernel sample obtains such as Fig. 9 feature K1Walnut kernel Sample scatterplot situation, as seen from Figure 9 feature K1The easy different integrity degree in area, therefore facilitated pair according to this feature Walnut kernel is more accurately classified.
2, the ratio K of walnut kernel contour area and its minimum circumscribed circle area is obtained2：As shown in attached drawing 8, walnut kernel profile The ratio between area and its minimum circumscribed circle contour area are:
Wherein, s- is walnut kernel contour area；πr2It is walnut kernel profile minimum circumscribed circle contour area；
The above process carries out under OpenCV mode, i.e. use can calculate walnut kernel contour area S by calculating walnut kernel wheel Profile surface Product function acquires；Using π r can be calculated2Function acquire；Here 10 walnut kernel be can use, be K with ordinate2, horizontal seat It is designated as walnut kernel sample and obtains such as 10 feature K of attached drawing2Walnut kernel sample scatterplot situation, by attached drawing 10 it can be seen that feature K2Hold The different integrity degree in easy area, therefore help more accurately to classify to walnut kernel according to this feature.
3, Hu not bending moment is obtained：When outline identification, to be remained unchanged when guaranteeing that image translation, rotation and ratio change, draw 7 features for entering to influence the geometric invariant moment of image zooming-out, are I respectively1、I2、I3、I4、I5、I6And I7；
The above process can carry out under OpenCV mode, even if obtaining the center of profile with the center moment function for calculating profile Square obtains Hu not bending moment using the function for being used to calculate its Hu not bending moment, and bending moment is not constructed Hu using second order and third central moment 7 squares；
IV, primitive character matrix is established
Primitive character matrix is established according to features described above, i.e., sample characteristics are obtained according to step I to III to each sample Data, then will obtain and be formed between sample characteristics and sample with every a line (Row) as a sample, each column (Col) are one The primitive character matrix 150 (Row) × 46 (Col) of a feature (referring to table 3-1)；
V, primitive character matrix is normalized, i.e., by normalized function to by the numerical value line of each column Property is mapped between [0,1] section；The above process can carry out under OpenCV mode, i.e., by for the function at normalization The numerical value of each column is normalized, i.e., in Linear Mapping to [0,1] section；
5th step：The selection of feature：
One, based on the Feature Selection of ReliefF：
ReliefF algorithm is a kind of feature weight algorithm, handles using the ReliefF algorithm feature that processing is tied Fruit is as shown in figure 11, it can be seen that feature 1 relevant to integrity degree comes the 23rd, and color selects in feature or exists a large amount of Redundancy feature.Wherein feature Col5-Col9 weight is negative, i.e. classifying quality and Hu moment characteristics I3To I7Non-correlation can incite somebody to action This 5 feature removals, feature is rearranged as shown in table 4-1 after removing 5 features；
The step of carrying out Feature Selection using ReliefF algorithm is as follows：
(1) initializing each feature weight is 0, and setting frequency in sampling is m；
(2) a sample R is randomly choosed from training set, and the k arest neighbors sample of R is then found out from the similar sample of R This.
(3) k nearest samples are found out from each inhomogeneity sample set of R, acquires each spy according to the following formula The weight of sign.
Wherein, Hj(j=1,2 ..., k)-indicates j-th of neighbour's sample in sample similar with R；Mj(C) (j=1,2 ..., K) j-th of neighbour's sample in-expression class C (with R inhomogeneity) sample；P (C)-is that the quantity of class C sample accounts for total number of samples amount Ratio；Class quantity where p (Class (R)) is sample R accounts for the ratio of total number of samples amount；diff(A,R1,R2) indicate sample R1 With sample R2Difference on feature A；Its calculation method is：
The weight that (2) (3) update each feature is recycled, until reaching preset frequency in sampling m.
Two, the feature selecting of two kinds of algorithm criterion of the information gap MID based on mRMR and mutual information MIQ：
MRMR (maximal correlation minimal redundancy) algorithm is one kind while guaranteeing maximum correlation, and removes redundancy feature Method；Algorithm measures the degree of correlation in character subset between feature and feature, between feature and classification using mutual information, mutually Information is estimating for two stochastic variable statistic correlations, can regard as include in a stochastic variable about another The information content of stochastic variable or a stochastic variable uncertainty of reduction due to another known stochastic variable；It is logical The two kinds of algorithms of MID, MIQ for crossing mRMR carry out feature selecting to training sample matrix, according to the degree of correlation between feature and weight Training result takes the higher feature of preceding 15 degrees of correlation as Modelling feature；
The algorithm of mRMR is as follows：
If the Joint Distribution of two stochastic variables (X, Y) is p (x, y), limit distribution is respectively p (x), p (y), then mutual trust Cease I (X；It Y is) relative entropy of Joint Distribution p (x, y) and product distribution p (x) p (y), formula is defined as：
Discrete variable is measured, makes correlation maximum and redundancy minimum by following formula：
Wherein S is characterized set；C is target category；I(xi；C) mutual information between target category c and feature i；I (xi；xj) it is characterized the mutual information between i and j；
To polytomy variable SmIt is defined as with target category c mutual information：
It can define max (D-R) or max (D/R) then to consider correlation and redundancy as a whole, the standard as evaluating characteristic Then, i.e. mutual information poor (MID) and mutual information quotient (MIQ).To the data set S for having m featurem, need from data set { S-SmIn Select so that the maximized the m+1 feature of D-R for：
So that maximized the m+1 feature of D/R is：
Feature selecting is carried out to training sample matrix using two kinds of algorithms of MID, MIQ based on mRMR, according to feature and power The training result of the degree of correlation between value takes preceding 15 features of algorithms selection；Preceding 15 features are as shown in Table 5-1；
6th step：The foundation of model is carried out by Naive Bayes Classifier and features described above；
By test sample and Naive Bayes Classifier to feature selecting, select afterwards makes Naive Bayes Classification after tested Device classification results accuracy reaches maximum K1, bin19 and bin17 3 features, utilize K1, bin19 and bin17 3 spies Sign is trained Naive Bayes Classifier, obtains walnut kernel hierarchy model；
Used here as Naive Bayes Classifier and K1, 3 features the reason of establishing model of bin19 and bin17 it is as follows：
1, three kinds of decision tree, naive Bayesian and support vector machines classifiers are selected；
2, the method choice training set of (SFS belong to heuristic search) is selected the feature after selection forward using sequence, I.e. since a feature, a feature is added every time, forms training set, and be respectively fed to be trained in three classifiers, Generate corresponding model；
3, it takes different grades of multiple walnut kernel as test sample (each 30 of each grade walnut kernel may be selected here), makes Placement test is done with still image of each model to test sample；
For the characteristic sequence of MID algorithms selection, test effect is as shown in figure 12, and wherein abscissa is the spy being added Sign, ordinate are classification accuracy, and support vector machines is classified accuracy after using 11 features and reached most as seen from the figure Big value 85.33%, decision tree accuracy after using 3 features reach maximum value 96.00%, and naive Bayesian is using 3 Accuracy reaches maximum value 97.33% after feature；
To the characteristic sequence of MIQ algorithms selection, test effect is as shown in figure 5, as seen from the figure due to 3 before selection A feature is identical, and decision tree is consistent with MID algorithm with naive Bayesian effect, but to use the supporting vector compared with multiple features Machine reaches maximum accuracy 85.33% after using 5 features；
4, the feature that removal importance is 0, leaves 3 important features, respectively：K1, bin19 and bin17.With this 3 Feature re -training decision tree, support vector machines and Naive Bayes Classifier, and pass through three of test sample after training Model is tested, and test effect is as shown in Table 6-1, can be seen that the model established by Naive Bayes Classifier by table 6-1 It is classified accuracy highest, therefore the present invention selects feature b in19, K1And bin15, and utilize the optimal simple shellfish of classifying quality This classifier of leaf is classified walnut kernel.
By test sample and Naive Bayes Classifier to feature selecting, select afterwards makes Naive Bayes Classification after tested Device classification results accuracy reaches maximum K1, bin19 and bin17 3 features；
The algorithm of Naive Bayes Classifier is as follows：
Provide s={ x1,x2,…,xmIt is an item to be sorted, each x is a characteristic attribute of s, this class that can divide It Wei not C={ y1,y2…yn, then the classifying step of naive Bayesian is：
1. calculating P (y1| s), P (y2| s) ..., P (yn|s)。
2. if P (yk| s)=max { P (y1| s), P (y2| s) ..., P (yn| s) }, then s ∈ yk。
3. critical issue therein is exactly to calculate each conditional probability in (1), just need to count using training sample at this time Calculate conditional probability.Circular is as follows：
4. counting the conditional probability estimation of each feature under each classification, i.e. P (x1|y1),P(x2|y1),…P(xm|y1)； P(x1|y2),P(x2|y2),…P(xm|y2)；…P(x1|yn),P(x2|yn),…P(xm|yn)。
5. being derived by Bayes' theorem as follows if each characteristic attribute is conditional sampling：
6. so need to only maximize molecule, and having due to each spy since denominator is constant for all categories Levying attribute is conditional sampling, then has：
It is calculated by the function of above-mentioned naive Bayesian, provides the sorting item of each rank of walnut kernel, solve this appearance Under conditions of, the probability that each classification occurs, which maximum probability, then it is assumed which classification this item to be sorted belongs to；
7th step：Classification tracking is carried out to walnut kernel to be measured according to model.
(1) as shown in Fig. 2, to the detection of target：
1, it obtains walnut kernel image to be fractionated and is used method identical with second step image procossing to carry out image Processing obtains the complete walnut kernel target image of profile；
2, it determines whether there is new walnut kernel profile and enters visual field, specific deterministic process is as follows：
A, the conveying direction from left to right based on delivery platform chooses each core in image using boundary rectangle frame respectively Peach kernel profile, by the most left extreme point abscissa (x of leftmost side walnut kernel boundary rectangle frame0) with previous frame image in leftmost side core The most left extreme point abscissa X of peach kernel boundary rectangle frame compares, if being less than (x0<X), then b is entered step；If more than (x0> X), then c is entered step；
B, each walnut kernel profile in image is chosen using boundary rectangle frame respectively, obtains the external square of leftmost side walnut kernel The most left extreme point abscissa of shape frame, judges whether the most left extreme point abscissa is less than m pixel, and m pixel is the range image left side The image range value of edge；If being less than, it is determined with new walnut kernel profile and enters visual field, using this profile boundary rectangle as initial Search box and the hue histogram for calculating original image pixels in walnut kernel profile in search box, the two is stored, addition One target to be tracked, enters step (two)；If more than not being then new walnut kernel profile, enter step c；
C, judge the deletion of walnut kernel target in image mark whether be it is true, if true, then delete the tracking target；If no Be it is true, then judge whether tracked target reaches classification position, if reaching, send graded signal to being used for controller The controller for controlling walnut kernel delivery device, makes controller action, realizes the progressive operation of walnut kernel, while setting this and waiting tracking Target deleted marker is true, and deletes this tracking target；If not reaching, continue to track target, repeat in judgement tracking window Whether the heart reaches classification position；
According to above-mentioned target following detecting step, judging whether there is the Rule of judgment that new walnut kernel profile enters visual field has Two；
Condition 1：Compare the most left extreme point abscissa (x of leftmost side walnut kernel boundary rectangle frame0) with previous frame image in most The most left extreme point abscissa X size of left side walnut kernel boundary rectangle frame；
Condition 2：The most left extreme point abscissa is less than m pixel, and m pixel is the image range value of range image left edge, m Value can be adjusted according to live service condition.
1 judgment basis of condition：By the most left extreme point abscissa (x of leftmost side walnut kernel boundary rectangle frame0) and previous frame figure The most left extreme point abscissa X of leftmost side walnut kernel boundary rectangle frame compares as in；From image, the entrance of walnut kernel Sequence is from left to right to enter, therefore the rightmost side enters earliest, and the leftmost side enters the latest, from the right side on the time Successively successively decrease to a left side；In previous frame image also so；Therefore, the walnut kernel of the leftmost side is to finally enter in previous frame image Walnut kernel, it is desirable to prove walnut kernel whether be it is new, seek to the walnut for finally entering the position of walnut kernel and previous frame Benevolence position is compared, if being newly less than the walnut kernel position that previous frame finally enters into walnut kernel position, this illustrates the new core Peach kernel enters visual field than leftmost side walnut kernel in previous frame later；If newly finally entered into walnut kernel position greater than previous frame Walnut kernel position, this illustrates that the new walnut kernel is possible to leftmost side walnut kernel in previous frame and has only moved forward some positions It sets；
2 judgment basis of condition：Condition 2 be based on condition 1 on the basis of further judgement, the reason is that classification section, institute The classification delivery device of use is likely to affect the change in location of walnut kernel in visual field, it is understood that there may be in visual field walnut kernel by Classification delivery device influence moves to left, and in next frame image, it is that the walnut kernel of the leftmost side is sentenced that PC host, which can default it, It is disconnected, the result x of judgement0<X, although meeting condition 1, actually it is not new walnut kernel profile；In order to avoid occurring The judgement of this mistake just can guarantee to certain range is newly given into the position of walnut kernel to the accurate of new profile judgement Property；Wherein, m pixel value can be selected 80, that is, judge the most left extreme point abscissa less than 80 pixels；80 pixel value, be by Multiple field test obtains, and is only limitted to use in the present embodiment 1；M pixel is adjusted with the size of acquisition image.
If in deterministic process, delivery platform is conveying direction from right to left, then most with the walnut kernel of the rightmost side in image Walnut kernel to be likely to be new is judged；Judgment step is identical as above-mentioned steps, contrary；
(2) as shown in Fig. 3, the classification and tracking for treating tracking target judge grade and the tracking position of target to be tracked It sets, realizes target progressive operation.
(1) target following is utilized using the search box and hue histogram stored as input quantity to each target to be tracked Method calculates the window's position of tracking target；Judge whether the tracking window center of target to be tracked reaches the feature of setting and mention Fetch bit is set, if reaching feature extraction position enters (2) step；If not reaching feature extraction position, repeat to judge mesh to be tracked Whether mark reaches feature extraction position；
(2) walnut image is divided using minimum circumscribed rectangle, obtains the image for only containing single walnut kernel；
(3) judge that tracked target whether there is grade in image：Extract each target to be tracked bin19, K1 and Tri- features of bin15 export each grade probability as input, using hierarchy model, and tracking thus for grade maximum probability is total The affiliated rank of target；Grade if it exists then enters (4) step；Grade is such as not present, then terminates to track；
(4) defined area is carried out to each classification position, is judged according to the location of each tracking window center wait chase after Whether the target of track reaches classification position, if reaching, sends graded signal to being used to control walnut kernel pusher to controller The controller of structure, makes controller action, realizes the progressive operation of walnut kernel, at the same set this target deleted marker to be tracked be it is true, And delete this tracking target；If not reaching, repetition judges whether the tracking window center of target to be tracked reaches classification position.
This method is based on the walnut kernel color of machine vision and the quality grading of integrity degree, referring to walnut kernel grade scale, It proposes and walnut kernel RGB image is converted into HIS color space, rgb space structure does not meet people to the master of color similarity See judgement, and in grade scale both by human eye differentiate color, so rgb space be converted to meet human eye to color be responsible for recognize The hsv color space of knowledge, and color feature extracted mainly is carried out in hsv color space, accuracy rate is higher；
The characteristics of for different brackets walnut kernel image, the primitive character of walnut kernel color and integrity degree is established, collection is passed through Middle feature selection approach is ranked up primitive character, is existed using ReliefF algorithm and the removal of mRMR algorithm to walnut kernel point Grade extracts bin19, K without influence or uncorrelated or repeated and redundant feature1With bin15 as Naive Bayes Classifier Three features most representative, that classifying type is best are used for machine learning；It extracts, analyze again to improve in this stage division Efficiency in terms of feature and training pattern；In this method, to the walnut kernel being classified using machine vision to each It is tracked by the way of showing journey into the walnut in visual field, i.e., PC host can track multiple cores in scene simultaneously more Peach kernel target is tracked, each tracing process is also responsible for the extraction of tracked walnut kernel target, classification processing etc., can be with The progressive operation of multiple walnut kernel targets is realized simultaneously, it is high-efficient.
In the present invention, for ease of description, the description of the relative positional relationship of each component is according to Figure of description 13 Butut mode is described, such as：The positional relationship of upper and lower, left and right etc. is the Butut according to Figure of description 13 Direction determines.
Embodiment 2：As shown in Fig. 13, a kind of walnut kernel stage division and walnut kernel classification based on based on machine vision The walnut kernel grading plant of device, including mobile platform 1, machine vision equipment, controller 5, delivery device and classification discharging Slot；Machine vision equipment is housed on mobile platform 1, machine vision equipment includes light box 2, camera 3 and PC host 4；Camera 3 It is electrically connected in light box 2, and with PC host 4；PC host 4 is electrically connected with the controller 5；Light box 2 is fixedly mounted on movement The paralell upper end of platform 1；The left hand inlet port 6 and right-side outlet 7 being equipped on light box 2 along 1 direction of mobile platform；In light box 2 Interior 1 rear side edge of mobile platform is between left and right every fixing at least two classification blowpits, on front side of the mobile platform 1 in light box 2 admittedly Dingan County is equipped with the delivery device that the walnut kernel after classification can be pushed to prescribed fractionated blowpit；The controller 5 and pusher Structure electrical connection.Wherein, light box 2 is the cabinet for being built-in with light source；The camera 3 is for acquiring walnut kernel image；The PC host 4, for handling the collected walnut kernel image of camera, are classified and track；The controller 5 is for receiving PC host letter Number, and control delivery device realization and push to the walnut kernel after classification in prescribed fractionated blowpit.
In use, walnut kernel is delivered to light box 2 by mobile platform 1, walnut kernel figure is obtained by camera 3 in light box 2 Picture, and by walnut kernel image transmitting to PC host 4, PC host 4 is tracked, to the grade of walnut kernel during tracking Classification judgement is carried out, after confirming grade, the tracking through camera 3 to target, when walnut kernel to be confirmed reaches respective level position, PC host 4 is sent to controller 5 executes signal, and controller 5 drives delivery device to push to walnut kernel and its rank position pair In the classification blowpit answered, the full automation classification of walnut kernel, high degree of automation, more saving manpower are realized；Simultaneously originally Decoration is the equipment based on walnut kernel stage division and walnut kernel grading plant based on machine vision, to walnut kernel classification Judge speed faster, it is more efficient.
MER-030-120U color camera can be used in camera, and image quality is high；LTS- can be used in light source in light box 2 2BR35030LED white light source；In long service life, cost is relatively low and energy conservation and environmental protection；It is cruel that Intel can be used in PC host 4CPU Farsighted tetra- core 3.4GHz of i5-4670；Image processing time is fast, and cost performance is high, improves hierarchical speed；Controller 5 can be used STC89C52RC business level single-chip microcontroller；Can switching control speed to solenoid valve it is fast, can further improve this walnut kernel point in this way The service performance of stage arrangement.
It can according to actual needs, to the core of the above-mentioned walnut kernel stage division based on machine vision and walnut kernel grading plant Peach kernel grading plant makes further optimization and/or improvements：
As shown in Fig. 13, delivery device uses jet delivery device；Jet delivery device includes air compressor 10, electricity Magnet valve 11, air jet pipe 9；It is fixed with air jet pipe 9 corresponding with classification blowpit on front side of the mobile platform 1 in light box 2, is spraying The connection of 9 upper end of tracheae is equipped with can be towards the spray head of corresponding classification blowpit；The lower end of every air jet pipe 9 respectively with institute State the connection of air compressor 10；It is fixedly installed with the solenoid valve 11 for controlling air jet pipe on-off on every air jet pipe 9, controls Device 5 processed is electrically connected with each solenoid valve 11 respectively.In use, the image that PC host 4 is acquired by camera 3 is to walnut kernel grade Judged, after confirming grade, the walnut kernel to be measured for entering visual field through 3 pairs of camera carries out continuous Image Acquisition, PC host 4 images acquired by camera 3 judges whether walnut kernel reaches respective level position, and PC host 4 is to controller 5 if arrival Send signal；PC host 4 sends signal to controller 5 and handles after controller 5 receives signal signal, determines and corresponds to The solenoid valve 11 of the grade, and it is sent to it execution signal；Solenoid valve 11 is controlled in air jet pipe 9 and is produced by air compressor 10 Raw high pressure gas is sprayed from spray head after solenoid valve 11 receives the signal for executing and opening from 9 mesohigh gas of air jet pipe, Walnut kernel is blown in the classification blowpit of corresponding grade, classification movement is completed；Using jet delivery device, it is not easy to core Peach kernel damages, and guarantees its integrality；Solenoid valve can be used MHE2-M1H-3/2G-M7 model solenoid valve, quick action, Further increase the service performance of this walnut kernel grading plant.
In order to more be accurately controlled the actuation time of solenoid valve 11, as needed, 5 dispatch from foreign news agency of controller is connected to timing Device.According to the difference of the single-chip microcontroller digit used, and the quantity of PLC internal simulation timer module used is different, passes through The mode of external electrical connections timer expands its timing function, guarantees to all solenoid valves of this walnut kernel grading plant 11 dynamic Make temporal control, blows to walnut kernel in classification blowpit and the tracking to next walnut kernel is avoided to impact；Periodically Device uses NEC D8253C-2 timer extended chip；It contains containing three mutually independent 16 bit timings/counters, meets It uses, further increases the service performance of this walnut kernel grading plant.
As shown in Fig. 13, delivery platform uses frequency conversion delivery platform, and frequency conversion delivery platform includes frequency converter 8, frequency converter 8 It is electrically connected with PC host 4.In use, controlling the start and stop and speed that frequency converter frequency realizes frequency conversion delivery platform by PC host 4 Control, sufficiently meet PC host 4 in target tracking for the controllable of the conveying speed of walnut kernel, realize the effect of classification More preferably；3G3JZ-AB004 model frequency converter can be used in frequency converter, and easily controllable, power consumption is low, and cost performance is high, further increases The service performance of this walnut kernel grading plant.
As needed, frequency converter is connect with PC host 4 by RS485 bus communication.RS485 communicates it with transmission distance The advantages that from length, transmission speed height and strong antijamming capability, improve the control efficiency of this walnut kernel grading plant.
As shown in Fig. 13, the rear side of mobile platform 1 in light box 2 is along between left and right every fixing four classification blowpits, with Air jet pipe 9 there are four being fixedly mounted is classified on front side of the corresponding mobile platform 1 of blowpit.
The above technical features constitute embodiments of the present invention, can basis with stronger adaptability and implementation result Actual needs increases and decreases non-essential technical characteristic, to meet the needs of different situations.
Table 1-1 walnut kernel specification of quality index
Table 2-1 walnut kernel grade
Grade/feature Color Integrity degree
Hoary hair road It is faint yellow Half benevolence
Bai Erlu It is faint yellow Four points of benevolence
Shallow parting Light amber Half benevolence
Shallow two tunnel Light amber Four points of benevolence
Deng outer Dark brown —
Table 3-1 feature and the row number table of comparisons
Column Col1 Col2 Col3…Col9 Col10…Col12 Col13…Col15 Col16…Col46
Feature K1 K2 I1…I7 Hμ Hv Hs Sμ Sv Ss bin0…bin30
Feature permutation after table 4-1 ReliefF Feature Selection
Column Col1 Col2 Col3 Col4 Col5…Col7 Col8…Col10 Col11…Col41
Feature K1 K2 I1 I2 Hμ Hv Hs Sμ Sv Ss bin0…bin30
Table 5-1 MID, MIQ feature selecting result
MID 30 1 26 31 2 24 27 23 29 25 28 6 22 32 16
MIQ 30 1 26 27 24 4 9 2 28 23 29 25 16 36 31
Table 6-1 decision tree, naive Bayesian and support vector machines select characteristic test effect
Classifier Decision tree SVM Naive Bayesian
Accuracy 93.33% 80.67% 94.67%

Claims (10)

1. a kind of walnut kernel stage division based on machine vision, it is characterised in that carry out in accordance with the following steps：
The first step：Obtain historical sample data：By manual grading skill, the walnut kernel sample each 30 of two or more grades is obtained It is a, 30 walnut kernel samples are placed on delivery platform, each walnut is obtained by the machine vision equipment on delivery platform Benevolence image；
Second step：Image procossing：The walnut kernel image of acquisition is handled, the complete image of walnut kernel profile is obtained；
Third step：Image segmentation：Walnut kernel image is divided using minimum circumscribed rectangle, obtains the figure for only containing single walnut kernel Picture；
4th step：Historical sample data is handled, primitive character matrix is established：
(I), primitive character of 37 features related with walnut kernel color of image as walnut kernel color grading is obtained, it is specific to wrap It includes and 31 color characteristics is obtained based on hue histogram, 31 features are bin0 to bin30 respectively；6 are obtained with based on color moment A feature, 6 features are H respectivelyμ、Hv、Hs、Sμ、SvAnd Ss；
(II), 9 features related with walnut kernel shape integrity degree are obtained, it is long to specifically include walnut kernel profile minimum circumscribed rectangle Width ratio K1, walnut kernel contour area and its minimum circumscribed circle contour area ratio K2With 7 Hu moment characteristics, 7 Hu moment characteristics are respectively I1、I2、I3、I4、I5、I6And I7；
(III), primitive character matrix is established according to features described above, wherein one sample of each behavior, each to be classified as a feature, Each column are K respectively from left to right1、K2、I1、I2、I3、I4、I5、I6、I7、Hμ、Hv、Hs、Sμ、Sv、Ss、bin0、bin1、bin2、 bin3、bin4、bin5、bin6、bin7、bin8、bin9、bin10、bin11、bin12、bin13、bin14、bin15、bin16、 bin17、bin18、bin19、bin20、bin21、bin22、bin23、bin24、bin25、bin26、bin27、bin28、bin29 And bin30；
(IV), primitive character matrix is normalized, i.e., by normalized function to by the numerical linear of each column It is mapped between [0,1] section；
5th step：The selection of feature：
One, based on the Feature Selection of ReliefF：Primitive character matrix is subjected to feature selecting operation with ReliefF algorithm, is gone Except Hu moment characteristics I3、I4、I5、I6、I7；
Two, the feature selecting of two kinds of algorithms of the information gap MID based on mRMR and mutual information MIQ：
1. being carried out using the information gap MID algorithm of mRMR to the eigenmatrix after the Feature Selection based on ReliefF algorithm special Sign selection, takes preceding 15 features of algorithms selection, 15 features are followed successively by bin19, K1、bin15、bin20、K2、bin13、 bin16、bin12、bin18、bin14、bin17、Hv,bin11,bin21,bin15；
2. being carried out using the mutual information MIQ algorithm of mRMR to the eigenmatrix after the Feature Selection based on ReliefF algorithm special Sign selection, takes preceding 15 features of algorithms selection, 15 features are followed successively by bin19, K1、bin15、bin16、bin13、I2、Sv、 K2, bin17, bin12, bin18, bin14, bin5, bin25 and bin20；
6th step：The foundation of model：In 15 features that two criterion of mutual information difference MID and mutual information quotient MIQ obtain respectively In, preceding 3 features, i.e. bin19, K are taken respectively1And bin15, Naive Bayes Classification Model is instructed using these three features Practice, obtains walnut kernel hierarchy model；
7th step：Classification tracking：Classification tracking is carried out to walnut kernel to be measured according to model, core is determined according to the size of output probability Peach kernel rank.
2. the walnut kernel stage division according to claim 1 based on machine vision, it is characterised in that the 7th step is according to mould Type carries out classification tracking to walnut kernel to be measured, and determining walnut kernel rank according to the size of output probability, specific step is as follows：
(1) to the detection of target：
1, it obtains walnut kernel image to be fractionated and image is handled using method identical with second step image procossing, obtain The complete walnut kernel target image of profile；
2, it determines whether there is new walnut kernel profile and enters visual field, specific deterministic process is as follows：
A, enter the direction of visual field, the conveying direction from left to right of delivery platform, using boundary rectangle frame point based on walnut kernel Not Xuan Qu each walnut kernel profile in image, by the most left extreme point abscissa of leftmost side walnut kernel boundary rectangle frame and previous frame figure The most left extreme point abscissa of leftmost side walnut kernel boundary rectangle frame compares as in, if being less than, enters step b；If more than, It is not then new walnut kernel profile, then enters step c；
B, each walnut kernel profile in image is chosen using boundary rectangle frame respectively, obtains leftmost side walnut kernel boundary rectangle frame most Left extreme point abscissa, judges whether the most left extreme point abscissa is less than m pixel, and m pixel is the figure of range image left edge As value range；If being less than, it is determined with new walnut kernel profile and enters visual field, simultaneously using this profile boundary rectangle as initial ranging frame The hue histogram for calculating original image pixels in walnut kernel profile in search box, the two is stored, and adds one wait chase after Track target enters step (two)；If more than not being then new walnut kernel profile, enter step c；
C, judging that the deletion of walnut kernel target in image marks whether is very, if true, then it is assumed that the walnut kernel is to have completed Classification, deletes the tracking target；If not true, the window's position of tracking target is calculated using method for tracking target；According to being chased after The position of the tracking window center of track target judges the classification position whether walnut kernel target reaches, if reaching, to controller Graded signal is sent to the controller for being used to control walnut kernel delivery device, makes controller action, realizes the classification behaviour of walnut kernel Make, while it is true for setting this target deleted marker to be tracked, and deletes this tracking target；If not reaching, continue to track target, then Repetition judges to track whether window center reaches classification position；
(2) classification and tracking for treating tracking target judge the grade and trace location of target to be tracked, realize target classification Operation.
3. the walnut kernel stage division according to claim 2 based on machine vision, it is characterised in that (two) step is treated The classification and tracking for tracking target judge the grade and trace location of target to be tracked, realize the specific step of target progressive operation It is rapid as follows：
(1) method for tracking target is utilized using the search box and hue histogram stored as input quantity to each target to be tracked Calculate the window's position of tracking target；Judge whether the tracking window center of target to be tracked reaches the feature extraction position of setting It sets, if reaching feature extraction position enters (2) step；If not reaching feature extraction position, repeat to judge that target to be tracked is No arrival feature extraction position；
(2) walnut image is divided using minimum circumscribed rectangle, obtains the image for only containing single walnut kernel；
(3) judge that target to be tracked is with the presence or absence of grade in image：Extract bin19, K of each target to be tracked1And bin15 Three features export each grade probability as input, using walnut kernel hierarchy model, and tracking thus for grade maximum probability is total The affiliated rank of target；Grade if it exists then enters (4) step；Grade is such as not present, then terminates to track；
(4) defined area is carried out to each classification position, is judged according to the location of each tracking window center to be tracked Whether target reaches classification position, if reaching, sends graded signal to being used to control walnut kernel delivery device to controller Controller makes controller action, realizes the progressive operation of walnut kernel, while it is true for setting this target deleted marker to be tracked, and is deleted Except this tracks target；If not reaching, repetition judges whether the tracking window center of target to be tracked reaches classification position.
4. the walnut kernel stage division according to claim 1 or 2 or 3 based on machine vision, it is characterised in that the 4th step (I) in, primitive character of 37 features related with walnut kernel color of image as walnut kernel color grading, respectively base are obtained 31 color characteristic bin0 to bin30 are obtained in hue histogram；6 feature H μ, Hv, Hs, S μ, Sv are obtained with based on color moment And Ss, specific step is as follows：
I, color space conversion；RGB color used in history image is converted to and meets human eye to color subjective understanding Hsv color space；
II, color feature extracted uses color histogram method and color to by the walnut kernel image in I treated image Moment Methods carry out color feature extracted；
(i) color histogram method extracts：The hue histogram that walnut kernel tone is distributed in history image is established, from tone histogram The main distribution of tone of each grade walnut kernel is obtained in figure, extracts the corresponding 31 tone frequencies of the main distribution of tone As tone characteristics, 31 tone characteristics of slave bin0 to the bin30 based on color histogram are formed；
(ii) color Moment Methods extract：The tone H and saturation degree S of selection image, foundation color moment three low-order moments of acquisition, three Low-order moment is first moment (mean value) H μ and S μ, second moment (variance) Hv and Sv and third moment (degree of bias) Hs and Ss respectively；It is described H μ, Hv, Hs, S μ, Sv, Ss composition express 6 features of distribution of color.
5. the walnut kernel stage division according to claim 1 or 2 or 3 or 4 based on machine vision, it is characterised in that second Walking image procossing, specific step is as follows：
(a) to the image of acquisition, binary conversion treatment is carried out, obtains the single channel image of the relatively sharp image of profile；
(b) denoising is carried out to by the single channel image after binary conversion treatment using morphological image method；
(c) the outermost layer profile of each walnut kernel in image is found, and bubble sort method is used to the walnut kernel profile found To the name placement of walnut kernel profile；
(d) conveying direction from left to right based on delivery platform obtains the most left extreme point coordinate of walnut kernel outermost layer profile, Judging the extreme point, whether image left edge coincides, if being overlapped, is determined as imperfect walnut kernel profile, from profile sequence Remove this profile；If not being overlapped, it is determined as complete walnut kernel profile, retains the walnut kernel image.
6. a kind of using according to claim 1 to the walnut of the walnut kernel stage division described in 5 any one based on machine vision Benevolence grading plant, it is characterised in that including mobile platform, machine vision equipment, the controller that can be used to control delivery device, push away Send mechanism and classification blowpit；Machine vision equipment is housed on a mobile platform, machine vision equipment includes light box, camera and PC Host；Camera is located in light box, and is electrically connected with PC host；PC host is electrically connected with the controller；Light box is fixedly mounted on The paralell upper end of mobile platform；The left hand inlet port and right-side outlet being equipped on light box along mobile platform direction；In light box Mobile platform on rear side of along between left and right every fixing at least two classification blowpits, be fixedly mounted on front side of the mobile platform in light box There is the delivery device that the walnut kernel after classification can be pushed to prescribed fractionated blowpit；The controller is electrically connected with delivery device It connects；The camera is for acquiring walnut kernel image；The PC host is used to handle the collected walnut kernel image of camera, Classification and tracking；The controller controls delivery device realization and pushes away the walnut kernel after classification for receiving PC host signal It send to prescribed fractionated blowpit.
7. walnut kernel grading plant according to claim 6, it is characterised in that delivery device uses jet delivery device；Spray Gas delivery device includes air compressor, solenoid valve, air jet pipe；It is fixed on front side of mobile platform in light box and classification discharging The corresponding air jet pipe of slot, air jet pipe upper end connection be equipped with can towards it is corresponding classification blowpit spray head；Every jet The lower end of pipe is connect with the air compressor respectively；It is fixedly installed on every air jet pipe for controlling air jet pipe on-off Solenoid valve, controller are electrically connected with each solenoid valve respectively.
8. walnut kernel grading plant according to claim 6 or 7, it is characterised in that controller uses PLC controller or list Piece machine, controller electricity are connected to timer.
9. walnut kernel grading plant according to claim 8, it is characterised in that delivery platform uses frequency conversion delivery platform, becomes Frequency delivery platform includes frequency converter, and frequency converter is electrically connected with PC host, and frequency converter is connected with PC host by RS485 bus communication It connects.
10. walnut kernel grading plant according to claim 8, it is characterised in that along left and right on rear side of the mobile platform in light box Four classification blowpits are fixed at interval, and there are four air jet pipes for fixed installation on front side of mobile platform corresponding with classification blowpit.
CN201810427065.2A 2018-05-07 2018-05-07 Walnut kernel grading method and walnut kernel grading device based on machine vision Active CN108855988B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810427065.2A CN108855988B (en) 2018-05-07 2018-05-07 Walnut kernel grading method and walnut kernel grading device based on machine vision

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810427065.2A CN108855988B (en) 2018-05-07 2018-05-07 Walnut kernel grading method and walnut kernel grading device based on machine vision

Publications (2)

Publication Number Publication Date
CN108855988A true CN108855988A (en) 2018-11-23
CN108855988B CN108855988B (en) 2023-09-08

Family

ID=64327494

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810427065.2A Active CN108855988B (en) 2018-05-07 2018-05-07 Walnut kernel grading method and walnut kernel grading device based on machine vision

Country Status (1)

Country Link
CN (1) CN108855988B (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109727231A (en) * 2018-12-17 2019-05-07 江苏鸿祺生物科技有限公司 A kind of appearance random defect detecting spheroid shape soft capsule for identification from Determination method
CN110030927A (en) * 2019-03-27 2019-07-19 中北大学 Walnut detection device
CN110363229A (en) * 2019-06-27 2019-10-22 岭南师范学院 A kind of characteristics of human body's parameter selection method combined based on improvement RReliefF and mRMR
CN110759195A (en) * 2019-11-06 2020-02-07 广东博智林机器人有限公司 Leveling control method, device and equipment
CN111398837A (en) * 2020-04-01 2020-07-10 重庆大学 Vehicle battery health state estimation method based on data driving
CN111968063A (en) * 2020-09-07 2020-11-20 北京凌云光技术集团有限责任公司 Morphological image filtering device and method
CN112132792A (en) * 2020-09-04 2020-12-25 浙江大学 Hyperspectrum and deep learning-based pixel level detection method for endogenous foreign matters in pecans
CN112362673A (en) * 2020-11-17 2021-02-12 清华大学天津高端装备研究院洛阳先进制造产业研发基地 Visual detection method and system for dumplings
TWI768932B (en) * 2021-05-27 2022-06-21 倍利科技股份有限公司 Image color adjustment method
CN114887898A (en) * 2022-04-06 2022-08-12 厦门陆海环保股份有限公司 Air valve control method and system of pneumatic sorting equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020139729A1 (en) * 2001-04-03 2002-10-03 Pearson Thomas C. Acoustical apparatus and method for sorting objects
CN106180001A (en) * 2016-09-21 2016-12-07 福建农林大学 Class spherical nut Non-Destructive Testing classification conveyer device and control method thereof
CN208407761U (en) * 2018-05-07 2019-01-22 新疆农业大学 Walnut kernel grading plant

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020139729A1 (en) * 2001-04-03 2002-10-03 Pearson Thomas C. Acoustical apparatus and method for sorting objects
CN106180001A (en) * 2016-09-21 2016-12-07 福建农林大学 Class spherical nut Non-Destructive Testing classification conveyer device and control method thereof
CN208407761U (en) * 2018-05-07 2019-01-22 新疆农业大学 Walnut kernel grading plant

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
郭政;史建新;周军;蔡建;: "机器视觉下核桃自动分级系统下位机设计" *

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109727231A (en) * 2018-12-17 2019-05-07 江苏鸿祺生物科技有限公司 A kind of appearance random defect detecting spheroid shape soft capsule for identification from Determination method
CN110030927A (en) * 2019-03-27 2019-07-19 中北大学 Walnut detection device
CN110363229A (en) * 2019-06-27 2019-10-22 岭南师范学院 A kind of characteristics of human body's parameter selection method combined based on improvement RReliefF and mRMR
CN110363229B (en) * 2019-06-27 2021-07-27 岭南师范学院 Human body characteristic parameter selection method based on combination of improved RReliefF and mRMR
CN110759195A (en) * 2019-11-06 2020-02-07 广东博智林机器人有限公司 Leveling control method, device and equipment
CN111398837A (en) * 2020-04-01 2020-07-10 重庆大学 Vehicle battery health state estimation method based on data driving
CN112132792A (en) * 2020-09-04 2020-12-25 浙江大学 Hyperspectrum and deep learning-based pixel level detection method for endogenous foreign matters in pecans
CN112132792B (en) * 2020-09-04 2022-05-10 浙江大学 Hyperspectrum and deep learning-based pixel-level detection method for endogenous foreign matters in pecans
CN111968063A (en) * 2020-09-07 2020-11-20 北京凌云光技术集团有限责任公司 Morphological image filtering device and method
CN111968063B (en) * 2020-09-07 2024-01-26 凌云光技术股份有限公司 Morphological image filtering device and method
CN112362673A (en) * 2020-11-17 2021-02-12 清华大学天津高端装备研究院洛阳先进制造产业研发基地 Visual detection method and system for dumplings
TWI768932B (en) * 2021-05-27 2022-06-21 倍利科技股份有限公司 Image color adjustment method
CN114887898A (en) * 2022-04-06 2022-08-12 厦门陆海环保股份有限公司 Air valve control method and system of pneumatic sorting equipment
CN114887898B (en) * 2022-04-06 2024-02-13 厦门陆海环保股份有限公司 Pneumatic sorting equipment air valve control method and system

Also Published As

Publication number Publication date
CN108855988B (en) 2023-09-08

Similar Documents

Publication Publication Date Title
CN108855988A (en) Walnut kernel stage division and walnut kernel grading plant based on machine vision
Zhou et al. Wheat ears counting in field conditions based on multi-feature optimization and TWSVM
CN105718945B (en) Apple picking robot night image recognition method based on watershed and neural network
López-García et al. Automatic detection of skin defects in citrus fruits using a multivariate image analysis approach
Huang Detection and classification of areca nuts with machine vision
CN109272500B (en) Fabric classification method based on adaptive convolutional neural network
CN110837768B (en) Online detection and identification method for rare animal protection
CN106934386B (en) A kind of natural scene character detecting method and system based on from heuristic strategies
CN101398898B (en) Plant leaf identification method based on manifold learning
Tamilselvi et al. Unsupervised machine learning for clustering the infected leaves based on the leaf-colours
CN108319973A (en) Citrusfruit detection method on a kind of tree
CN107230188B (en) Method for eliminating video motion shadow
CN108537239A (en) A kind of method of saliency target detection
CN107230203A (en) Casting defect recognition methods based on human eye vision attention mechanism
CN110610506B (en) Image processing technology-based agaricus blazei murill fruiting body growth parameter detection method
CN105825233B (en) A kind of pedestrian detection method based on on-line study random fern classifier
CN109271932A (en) Pedestrian based on color-match recognition methods again
CN102208099A (en) Illumination-change-resistant crop color image segmentation method
CN105138975B (en) A kind of area of skin color of human body dividing method based on degree of depth conviction network
CN109344842A (en) A kind of pedestrian&#39;s recognition methods again based on semantic region expression
CN108846404A (en) A kind of image significance detection method and device based on the sequence of related constraint figure
Pandey et al. Non-destructive quality grading of mango (Mangifera Indica L) based on CIELab colour model and size
CN108710862A (en) A kind of high-resolution remote sensing image Clean water withdraw method
CN102184404A (en) Method and device for acquiring palm region in palm image
CN106778897B (en) Plant species recognition methods twice based on COS distance and center profile distance

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
GR01 Patent grant
GR01 Patent grant