CN106599902B - A kind of sugar crystal Classification and Identification and crystalline quality control method based on image - Google Patents

A kind of sugar crystal Classification and Identification and crystalline quality control method based on image Download PDF

Info

Publication number
CN106599902B
CN106599902B CN201610971626.6A CN201610971626A CN106599902B CN 106599902 B CN106599902 B CN 106599902B CN 201610971626 A CN201610971626 A CN 201610971626A CN 106599902 B CN106599902 B CN 106599902B
Authority
CN
China
Prior art keywords
image
classification
sugar crystal
sample
identification
Prior art date
Application number
CN201610971626.6A
Other languages
Chinese (zh)
Other versions
CN106599902A (en
Inventor
杨丽明
张攀
董超
Original Assignee
浙江大学
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 浙江大学 filed Critical 浙江大学
Priority to CN201610971626.6A priority Critical patent/CN106599902B/en
Publication of CN106599902A publication Critical patent/CN106599902A/en
Application granted granted Critical
Publication of CN106599902B publication Critical patent/CN106599902B/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
    • G06K9/6227Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
    • G06K9/6256Obtaining sets of training patterns; Bootstrap methods, e.g. bagging, boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6267Classification techniques
    • G06K9/6268Classification techniques relating to the classification paradigm, e.g. parametric or non-parametric approaches
    • G06K9/6269Classification techniques relating to the classification paradigm, e.g. parametric or non-parametric approaches based on the distance between the decision surface and training patterns lying on the boundary of the class cluster, e.g. support vector machines
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6267Classification techniques
    • G06K9/6279Classification techniques relating to the number of classes
    • G06K9/628Multiple classes
    • G06K9/6281Piecewise classification, i.e. whereby each classification requires several discriminant rules
    • G06K9/6282Tree-organised sequential classifiers

Abstract

The invention discloses a kind of sugar crystal Classification and Identification and crystalline quality control method based on image, belong to image recognition sorting technique and industrial control system field.Step: 1) under the collaboration to electro-optical device, the original sample image of sugar crystal is obtained from micro-photographing apparatus, original sample image is pre-processed using Preprocessing Technique, and pretreated image is split, obtains the region of interest of sugar crystal sample image;2) it is directed to region of interest, the characteristic parameter of every width sample image is therefrom extracted according to given identification feature and multi-categorizer in parallel is trained using it;3) the setting control period obtains image to be classified as unit of a control period within the period, is classified after pretreatment using trained multi-categorizer in parallel, the crystalline state information of sugar crystal is calculated according to classification results;4) executive device movement is controlled according to crystalline state information.This method can realize the automation of cane sugar boiling and crystal process on the whole.

Description

A kind of sugar crystal Classification and Identification and crystalline quality control method based on image

Technical field

The present invention relates to image recognition sorting technique and industrial control system, in particular to sugar crystal image multi-categorizer, The method and apparatus classified using multi-categorizer to sugar crystal image, and sugared crystalline is carried out using identification classification results Measure the control method of control.

Background technique

Industrially, sugar crystallization process needs operator constantly to observe crystal grain situation in tank, corresponding to take Operation.Majority sugar refinery is all that sampling probe taking-up massecuite sample is observed under lamp or hand is twisted with feeling and judged, Bu Nengzhi at present It connects and observes crystal grain situation of change in tank, bring certain difficulty to the accuracy of operation.To solve this problem, Ren Menkai It sends out, have developed and sugared microexamination mirror is boiled based on video technique, by the microscope on tank skin, can clearly observe tank The situation of change of interior crystal finds abnormal grain growth in time, to take corresponding operation.In fact, this is still relied on It works in human eye, only labour is transferred to eye-observation and boils sugared show from the crystal quality situation for directly distinguishing that probe samples On the image of micro mirror acquisition, and have not been changed artificial conventional operation mode.It is aobvious to sugar is boiled in order to realize the automation of sugar boiling process The sugar crystal image automatic identification classification tool that micro- peephole acquires out has very important significance.

However, whole clearing, when acquiring image, light is from crystal back on the one hand since sugar crystal surface is more smooth It injects, passes through crystal, just very high in the brightness that the smoother plane transmission of crystal goes out, other parts are then darker, acquire out Image is one, and there are the images of critical noisy;On the other hand, various crystal are mixed in the crystal to conform to quality requirements, it is such as poly- Brilliant, viscous brilliant, pseudo- crystalline substance etc., so that the type of sample greatly increases.Therefore, both at home and abroad at present to the identification of sugar crystal classification still without Effective method.

The automatic recognition classification of sugar crystal is an important ring for the automation of sugar boiling process, it is obtained the result is that after The decision-making foundation of continuous control system, cannot accurately identify classification results, just cannot achieve the automation of sugar boiling process.

Summary of the invention

It is an object of the invention to solve problems of the prior art, and provide a kind of sugar crystal based on image point Class identification and crystalline quality control method.

Sugar crystal Classification and Identification and crystalline quality control method based on image, comprise the steps of:

1) under the collaboration to electro-optical device, the original sample image of sugar crystal is obtained from micro-photographing apparatus, utilizes figure As preconditioning technique pre-processes original sample image, and pretreated image is split, obtains sugar crystal sample The region of interest of this image;

2) it is directed to region of interest, characteristic parameter and the utilization of every width sample image are therefrom extracted according to given identification feature It is trained multi-categorizer in parallel;

3) the setting control period obtains image to be classified as unit of a control period within the period, sharp after pretreatment Classified with trained multi-categorizer in parallel, the crystalline state information of sugar crystal is calculated according to classification results;

4) executive device movement is controlled according to crystalline state information, carry out adding water or adds honey.

In above-mentioned technical proposal, following implementation is can be used in each step:

The training of parallel connection multi-categorizer comprises the steps of in step 2):

2.1) correction, identification are accepted or rejected for the characteristic selection sugar crystal image recognition feature of sugar crystal and in fit procedure Feature includes: single crystal grain size, crystal grain number, crystalline areas and image area ratio, transparency and shape;

2.2) result of regulation identification classification only has two classes, and one kind is the sugar crystal image to conform to quality requirements, is defined as Clear, another kind of is the sugar crystal image for not meeting quality requirement, is defined as unclear;

2.3) it is directed to region of interest, according to the sugar crystal identification feature of setting, feature extraction goes out each of each sample image Kind quantitative description parameter, the sample data set 1 used as training decision tree classifier;By stochastical sampling to sample data set 1 progress double sampling generates different data sample set, the sample data set 2 used as training SVM classifier;Two kinds points Class device carries out stand-alone training and test based on the different views of same data, and two groups of different sample datas are respectively divided into training Data and test data, training data are used to carry out classifier to the model trained for training classifier, test data Assessment is service check;

2.4) simple subprogram traverses entire training dataset, classifies to the sugar crystal data set of known class: fighting to the finish For plan Tree Classifier, before carrying out the identification classification of the last one feature, according to identification feature, it is unsatisfactory for currently identifying special Sugar crystal data when sign are labeled with the label of unclear, exit the training process of epicycle;Otherwise it is instructed into the classification of next round Practice, is performed until the last one feature;By the training that last is taken turns, can obtain finally as a result, all sugar crystal image Data are all assigned to two classifications, and one kind is to may determine that up-to-standard clear classification, and one kind is to may determine that quality not Qualified unclear classification;

2.5) above two single classifier identification classification results obtained are based on, integrate two kinds points by combined method Class is as a result, obtain a final classification device;The combined method uses following algorithm:

It is as follows to define error rate:

Wherein, xjFor j-th of sample that training data is concentrated, Ci(xj) it is i-th of classifier to sample xjClassification knot Fruit, i=1 or 2;I is indicator, if parameter A is true, otherwise I (A)=1 is 0;wjThe jth concentrated for training data The weight of a sample;yjFor target classification result;

Calculate decision tree classifier importance beThe importance of SVM classifier is

By parameter alpha12For updating training sample weight, weight update mechanism are as follows:

Wherein ZjFor standardizing factor, for ensuring w1 (j+1)+w2 (j+1)=1;

By w1 (j+1)And w2 (j+1)The weight of decision tree classifier and SVM classifier, is used for when respectively as next subseries Calculate the final classification result of current sample.

The crystalline state information that sugar crystal is calculated in step 3) comprises the steps of:

3.1) the setting control period, as unit of a control period, within the control period under the collaboration to electro-optical device, The original image to be classified of specified quantity is obtained from micro-photographing apparatus;

3.2) to every original image to be classified, pretreatment and image segmentation are carried out, according to given identification feature from The middle characteristic parameter for extracting each image is input to trained multi-categorizer in parallel and classifies, up-to-standard figure As stamping clear label, image off quality stamps unclear label;

3.3) to tagged identification classification results, it is automatically converted into sugar crystal image file using IDL program, This document is identified image classification file;

3.4) batch processing is carried out to image classification file, obtains the sugar crystal figure for stamping clear/unclear label respectively The quantity of picture, according to quantity obtain current sugar crystal locating for crystalline state information.

Above-mentioned batch processing process are as follows:

3.4.1) by the image file category having been classified distribute distinguished with classification naming entitled clear, In the file of unclear;

3.4.2 the quantity of image file in each file) is counted;

3.4.3 all image files in the file of clear, unclear in this control period) are deleted, so as under The use in one control period;

3.4.4 the autoexec) is periodically executed, is i.e. after the timing of control cycle length arrives, executes primary batch File is handled, using statistical result as the information source of this control decision.

It is as follows according to sugar crystal crystalline state information control executive device action step in step 4):

4.1) the data block DATA BLOCK in PLC host computer reads in each classification quantity in this control period manually, makees For the foundation of subsequent control movement, quantity of classifying is as analog input data;

4.2) PLC timer is reset;

4.3) analog input data are compared in PLC host computer with given threshold, obtain corresponding switch letter Number: if the quantity of unclear classification is greater than a certain threshold value N1, PLC charging motor obtains enabling signal, and motor drives feed pump It opens, material enters tank, boils tank charging with the control of Ratio control method, while opened loop control boils level in tank;If clear classification Sugar crystal amount of images be greater than a certain threshold value N2, PLC discharging machine enabling signal, motor drives discharging pump to open, and material goes out Tank, while opened loop control boils level in tank and makes to boil tank stopping crystallization;

4.4) PLC wait next time timing arrive, repeat above-mentioned 4.1) -4.3) step.

The present invention has the advantages that as follows:

1, it to decision tree classifier is trained by sample set and SVM support vector machine classifier is effectively combined, obtains To classifying quality preferably multi-categorizer in parallel is identified, carried out automatically to by boiling the collected sugar crystal image of sugared microexamination mirror Analysis and classification.

2, the control system based on PLC controller and device are using classification results as the work of crystalline quality control system Foundation realizes the automation of cane sugar boiling and crystal process on the whole.

Detailed description of the invention

Fig. 1 is sugar crystal image recognition categorizing system and crystalline quality control method work flow diagram;

Fig. 2 is the training flow chart of decision tree and SVM classifier;

Fig. 3 is classification results figure of the decision tree classifier under sugar crystal single features data;

Fig. 4 is decision tree and SVM parallel connection Combination of Multiple Classifiers schematic diagram;

Fig. 5 is to carry out classification work flow diagram to sugar crystal image to be identified using multi-categorizer in parallel;

Fig. 6 is the control system work flow diagram based on PLC controller;

Specific embodiment

The present invention is further elaborated in the following with reference to the drawings and specific embodiments, so that those skilled in the art can be more Essence of the invention is understood well.

Due to the interference of the critical noisy and various mixed crystals of image, discovery utilizes single point based on decision-tree model Class device, it is difficult to carry out accurately identification classification.The purpose of the present invention is study a kind of more accurate sugar crystal image recognition point Class system and crystalline quality control method.Using decision tree and SVM support vector machines parallel connection multi-categorizer, in this control period The sugar crystal image of acquisition is automatically analyzed and is identified classification, as unit of a control period, quantitatively calculates our institutes The sugar crystal quality information needed, and the work foundation of later crystallization quality control system is utilized this information as, it is then real The automation of existing cane sugar boiling and crystal process.In addition, this sugared image analysis and automatic recognition system improve sugared crystalline quality judgement Objectivity, while the work load for boiling sugared operator is also mitigated, improve the accuracy of analysis.

, it is specified that the identification classification results of every 20min are converged with 20min for a control period in the present embodiment Always, as the information source of each control decision.

As shown in Figure 1, sugar crystal image recognition classification proposed by the present invention and crystalline quality control method, by microimaging Device 1, to electro-optical device 2, image recognition multi-categorizer 3, PLC upper computer software 4, the control system based on PLC controller and dress It sets 5, PLC timer 6 and computer 7 forms.Wherein, image recognition multi-categorizer 3, PLC upper computer software 4, based on PLC control The control system of device and the communication and information processing of the parts such as device 5 and PLC timer 6 are relied on computer 7 and are completed.

A variety of sugar crystal image recognition features proposed by the present invention, they include: single crystal grain size, crystal Numbers of particles, crystalline areas and image area ratio, transparency and shape.

As shown in Fig. 2, being located in advance using Preprocessing Technique to sample image from sugar crystal original image 11 Reason 12, including gray scale adjustment, median filtering, image enhancement etc., and 13 are split to pretreated image, obtain sugar crystal The region of interest of image.

For region of interest, according to the sugar crystal identification feature set in step before, feature extraction 14 goes out each sample The various quantitative description parameters of image, the sample data set 1 15 used as training decision tree classifier.

Data obtained in the previous step are generated by 16 double sampling of random sampling technique by random sampling technique different Data sample set, as the sample data set 2 17 that uses of training SVM classifier.

Two kinds of classifiers carry out stand-alone training and test based on the different views of same data.By two groups of different sample numbers According to respectively training data 19 and test data 18 is divided into, the former is used for for training classifier 21, the latter to the mould trained It is service check that type, which carries out classifier evaluation 20,.

The result of regulation identification classification only has two classes, and one kind is the image to conform to quality requirements, is defined as clear, another Class is the image for not meeting quality requirement, is defined as unclear.

Classifier 1 is based on decision-tree model, can be used for handling continuous characteristic as under predictive variable and target variable Classification, obtain regression tree.It can be with using the classification that the machine learning software with tree visualization tool carries out under single features Obtain structure as shown in Figure 3.

As shown in figure 3, the node 31 on the left side indicates data, that is, unclear classification of non-mass qualification, it is purely only a kind of Classification, but the node 32 on the right is still mixed with clear and unclear two classification, needs further progress training.At this time after Continuous addition feature is to increase the tree node in upper figure, until obtaining satisfied result, that is to say, that until leaf node is that only have Until a kind of pure node of classification, this classification or it is that may determine that up-to-standard clear classification or is that may determine that matter Measure underproof unclear classification.

Simple subprogram traverses entire training dataset 1, classifies to the sugar crystal data set 1 of known class.To decision For Tree Classifier, before carrying out the identification classification of the last one feature, according to identification feature, each sugar crystal data (image) or the label for being labeled with unclear exits the training process of epicycle or enters the classification based training of next round, and one Straight row is to a last feature.By last take turns training, can obtain finally as a result, all sugar crystal image datas all Two classifications are assigned to, one kind is to may determine that up-to-standard clear classification, and one kind may determine that off quality Unclear classification.

Training process when decision is exactly the process for increasing node and forming tree, is obtained after training similar to binary tree Structural model, this is also the purpose of decision tree classifier training.To the decision tree classifier trained test set testing classification Performance, it is ensured that classifying quality.

Classifier 2 is based on SVM support vector machines.The SVM has the learning method of supervision for one on statistical concepts, For carrying out classification and regression analysis.SVM is a kind of basis for having solid theory, novel small-sample learning method simultaneously. The theoretical basis of SVM is the VC dimension theory of structural risk minimization principle and basic statistical study theory.

The principle of SVM is that the point in lower dimensional space is mapped in higher dimensional space, them is made to become linear separability.Make again Classification boundaries are judged with the principle of linear partition, and this principle can be realized by kernel function.The kernel function has 4 Kind:

(1) linear kernel function

(2) Polynomial kernel function

(3) Radial basis kernel function

(4) Sigmoid kernel function

The selection of kernel function has apparent influence for nicety of grading, and different kernel functions has different adaptability, such as What suitable kernel function of selection is one important problem when carrying out svm classifier.Firstly the need of kernel function is determined, then selection is closed Practical problem is transformed into higher dimensional space using kernel function by suitable parameter.For example, when selecting Radial basis kernel function, C and γ this Two parameters, value are related to the height of nicety of grading.Under normal circumstances, C and γ value common method is cross validation Grid data service.Herein using Radial basis kernel function as SVM kernel function.

Simple subprogram traverses entire training dataset 2, carries out identification classification to the data set 2 of known class.To svm classifier For device, trained process is exactly to obtain rule according to the sugar crystal training data of these known class, establishes disaggregated model Process.It final all sugar crystal training datas or is categorized into up-to-standard clear classification or is categorized into off quality Unclear classification.

To the SVM classifier trained test set testing classification performance, it is ensured that classifying quality.

A kind of decision-making foundation that representative result combined method is final as decision-making device 43 is selected, to classifier 1 Decision tree classifier 41 and classifier 2SVM classifier 42 identify that classification results merge and recombination obtains a kind of more points of parallel connection Class device 44, as shown in Figure 4.

The comprehensive performance of multi-categorizer is built upon on the basis of the difference between multiple classifiers, and classifier otherness can To be generated by the way that different parameters is arranged.Herein, different node-classification feature selectings can be used for decision tree; And SVM then can choose the parameter etc. of different kernel function or kernel function.

Based on above two single classifier identification classification results obtained, because more than image interference factor and mixing each Kind improper crystal, therefore obtained classification accuracy rate does not reach requirement, therefore use a kind of combined method to integrate two kinds of classification There is the final classification device of more preferable classification capacity to get more accurate to classifying for clear and unclear as a result, obtaining one Classifier.This decision-making technique direct intervention classification results.

The combined method uses following algorithm.In the algorithm, the importance of two classifier classification results relies on In its classification error rate.The definition of error rate is

Wherein, xjFor j-th of sample that training data is concentrated, Ci(xj) it is i-th of classifier to sample xjClassification knot Fruit, i=1 or 2;I is indicator, if parameter A is true, otherwise I (A)=1 is 0;wjThe jth concentrated for training data The weight of a sample;yjFor target classification result.Concept based on error rate, then the importance of decision tree classifier beThe importance of SVM classifier is

It will be apparent that the more high then importance of error rate is lower, then this classifier gets over " right to speak " of final classification result It is low.

Parameter alpha12For updating training sample weight, the weight update mechanism of the algorithm is determined by following formula

Wherein ZjFor standardizing factor, for ensuring w1 (j+1)+w2 (j+1)=1.

By w1 (j+1)And w2 (j+1)The weight of decision tree classifier and SVM classifier, is used for when respectively as next subseries Calculate the final classification result of current sample.That is, the training result of single classifier is to the important of final decision result Degree is by weights influence, to a certain sample data, if it identifies that classification results are consistent with known class, in final decision In weight it is just high, i.e., influence assembled classifier model result a possibility that with regard to big;Conversely, will pay for, this point is reduced Class device directly affects " right to speak " to next sample data in weight, the measure for being promoted and being punished by this weight, Can effectively integrated decision-making Tree Classifier and SVM classifier classification advantage, promoted sorter model precision so that obtain group Merge connection classifier to significantly improve the classification accuracy of unfiled image data.

As shown in figure 5, the image 51 of every classification to be identified will also be pre-processed by being similar to model training process And segmentation, the characteristic parameter of each image is therefrom extracted according to given identification feature, be input to this system decision tree and SVM parallel connection multi-categorizer 52 is classified, the result is that or stamping the up-to-standard clear label of symbol or stamping symbol Unclear label off quality.

To tagged identification classification results 53, sugar crystal image file 54 is automatically converted into using IDL program. Image file at this time is different from original image file, from the data file transition for having played label, therefore it is It is identified image classification file 55.

The autoexec .bat 56 write under Windows operating system, function is as follows:

By the image file category having been classified distribution 57 with i.e. difference entitled clear, the unclear's of classification naming In file.

Count the quantity of image file in 58 each files.

All image files in the file of clear, unclear in 59 control periods are deleted, so as to next Control the use in period.

As shown in fig. 6, every PLC 20min timing executes an autoexec to after 61, using statistical result as The information source of this control decision.PLC circulation step:

Data block DATA BLOCK in upper computer software reads in each classification quantity 62 in this control period manually, as The foundation of subsequent control movement, it is notable that classification quantity is analog data.

PLC timer restarts timing.

Analog input data are compared with given threshold in PLC host computer, obtain corresponding switching signal.

If the quantity of unclear classification is greater than a certain threshold value N1 63, PLC charging motor obtains enabling signal 64, motor Feed pump 65 is driven to open, material enters tank.To guarantee crystalline quality, it is necessary to assure between each charging (green molasses, methylene and water) Proportionate relationship can be controlled with Ratio control method and boil tank charging 66, while opened loop control boils level 67 in tank.Otherwise compare The sugar crystal amount of images of clear classification.

If the sugar crystal amount of images of clear classification be greater than a certain threshold value N2 69, PLC discharging machine enabling signal 70, Motor drives discharging pump to open 71, and material goes out tank, while opened loop control boils level 72 in tank.It must make to boil tank stopping crystallization simultaneously 73。

It feeds in the process flow of cane sugar boiling and crystal or only or only discharges, the two operations will not deposit simultaneously , in this way it is necessary that charging and discharging can not carry out with being without end, level height arrival upper limit value 68 and lower limit value 74 When can motor automatic stop be operated, to ensure equipment safety.As for the assignment problem of upper limit or lower limit, according to Specifically boil tank situation and industrial requirements, limit value data and current level data, DB block be written, convenient for PLC ipc monitor, Assignment.

Timing arrives next time for PLC waiting, repeats PLC circulation step.

Above-mentioned embodiment is only a preferred solution of the present invention, so it is not intended to limiting the invention.Have The those of ordinary skill for closing technical field can also make various changes without departing from the spirit and scope of the present invention Change and modification.Therefore all mode technical solutions obtained for taking equivalent substitution or equivalent transformation, all fall within guarantor of the invention It protects in range.

Claims (4)

1. a kind of sugar crystal Classification and Identification and crystalline quality control method based on image, which is characterized in that by following steps group At:
1) under the collaboration to electro-optical device, the original sample image of sugar crystal is obtained from micro-photographing apparatus, it is pre- using image Processing technique pre-processes original sample image, and is split to pretreated image, obtains sugar crystal sample graph The region of interest of picture;
2) it is directed to region of interest, the characteristic parameter of every width sample image is therefrom extracted according to given identification feature and utilizes its right Multi-categorizer in parallel is trained;
3) the setting control period obtains image to be classified as unit of a control period within the period, and instruction is utilized after pretreatment The multi-categorizer in parallel perfected is classified, and the crystalline state information of sugar crystal is calculated according to classification results;
4) executive device movement is controlled according to crystalline state information, carry out adding water or adds honey;
The training of parallel connection multi-categorizer comprises the steps of in step 2):
2.1) correction, identification feature are accepted or rejected for the characteristic selection sugar crystal image recognition feature of sugar crystal and in fit procedure It include: single crystal grain size, crystal grain number, crystalline areas and image area ratio, transparency and shape;
2.2) result of regulation identification classification only has two classes, and one kind is the sugar crystal image to conform to quality requirements, is defined as Clear, another kind of is the sugar crystal image for not meeting quality requirement, is defined as unclear;
2.3) it is directed to region of interest, according to the sugar crystal identification feature of setting, feature extraction goes out the various fixed of each sample image Characterising parameter is measured, the sample data set 1 used as training decision tree classifier;By stochastical sampling to sample data set 1 into Row double sampling generates different data sample set, the sample data set 2 used as training SVM classifier;Two kinds of classification Device carries out stand-alone training and test based on the different views of same data, and two groups of different sample datas are respectively divided into trained number According to and test data, for training classifier, test data is used to carry out classifier to the model trained commenting training data Estimate i.e. service check;
2.4) simple subprogram traverses entire training dataset, classifies to the sugar crystal data set of known class: to decision tree For classifier, before carrying out the identification classification of the last one feature, according to identification feature, when being unsatisfactory for current identification feature Sugar crystal data be labeled with the label of unclear, exit the training process of epicycle;Otherwise enter the classification based training of next round, It is performed until the last one feature;By the training that last is taken turns, can obtain finally as a result, all sugar crystal image data All be assigned to two classifications, one kind is to may determine that up-to-standard clear classification, one kind be may determine that it is off quality Unclear classification;
2.5) above two single classifier identification classification results obtained are based on, two kinds of classification knots are integrated by combined method Fruit obtains a final classification device;The combined method uses following algorithm:
It is as follows to define error rate:
Wherein, xjFor j-th of sample that training data is concentrated, Ci(xj) it is i-th of classifier to sample xjClassification results, i=1 Or 2;I is indicator, if parameter A is true, otherwise I (A)=1 is 0;wjJ-th sample concentrated for training data Weight;yjFor target classification result;
Calculate decision tree classifier importance beThe importance of SVM classifier isIts Middle ε1And ε2The respectively classification error rate of decision tree classifier and SVM classifier;
By parameter alpha12For updating training sample weight, weight update mechanism are as follows:
Wherein ZjFor standardizing factor, for ensuring w1 (j+1)+w2 (j+1)=1;
By w1 (j+1)And w2 (j+1)The weight of decision tree classifier and SVM classifier when respectively as next subseries, for calculating The final classification result of current sample.
2. the sugar crystal Classification and Identification and crystalline quality control method, feature according to claim 1 based on image exists In the crystalline state information for calculating sugar crystal in step 3) comprises the steps of:
3.1) setting control the period, as unit of a control period, control the period in electro-optical device collaboration under, from show The original image to be classified of specified quantity is obtained in micro- photographic device;
3.2) to every original image to be classified, pretreatment and image segmentation is carried out, is therefrom mentioned according to given identification feature The characteristic parameter for taking each image is input to trained multi-categorizer in parallel and classifies, and up-to-standard image is beaten Upper clear label, image off quality stamp unclear label;
3.3) to tagged identification classification results, sugar crystal image file, this article are automatically converted into using IDL program Part is identified image classification file;
3.4) batch processing is carried out to image classification file, obtains the sugar crystal image for stamping clear/unclear label respectively Quantity, according to quantity obtain current sugar crystal locating for crystalline state information.
3. the sugar crystal Classification and Identification and crystalline quality control method, feature according to claim 2 based on image exists In the batch processing process are as follows:
3.4.1) the image file category having been classified is distributed with i.e. difference entitled clear, the unclear's of classification naming In file;
3.4.2 the quantity of image file in each file) is counted;
3.4.3 all image files in the file of clear, unclear in this control period are deleted) so as to next Control the use in period;
3.4.4 it) periodically executes the autoexec and executes a batch processing after the timing of i.e. control cycle length arrives File, using statistical result as the information source of this control decision.
4. the sugar crystal Classification and Identification and crystalline quality control method, feature according to claim 1 based on image exists In as follows according to sugar crystal crystalline state information control executive device action step in step 4):
4.1) the data block DATA BLOCK in PLC host computer reads in each classification quantity in this control period manually, as rear The foundation of continuous control action, quantity of classifying is as analog input data;
4.2) PLC timer is reset;
4.3) analog input data are compared in PLC host computer with given threshold, obtain corresponding switching signal: such as The quantity of fruit unclear classification is greater than a certain threshold value N1, and PLC charging motor obtains enabling signal, and motor drives feed pump to open, Material enters tank, boils tank charging with the control of Ratio control method, while opened loop control boils level in tank;If the sugar of clear classification Crystal amount of images is greater than a certain threshold value N2, PLC discharging machine enabling signal, and motor drives discharging pump to open, and material goes out tank, together When opened loop control boil in tank level and make to boil tank and stop crystallization;
4.4) PLC wait next time timing arrive, repeat above-mentioned 4.1) -4.3) step.
CN201610971626.6A 2016-11-04 2016-11-04 A kind of sugar crystal Classification and Identification and crystalline quality control method based on image CN106599902B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610971626.6A CN106599902B (en) 2016-11-04 2016-11-04 A kind of sugar crystal Classification and Identification and crystalline quality control method based on image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610971626.6A CN106599902B (en) 2016-11-04 2016-11-04 A kind of sugar crystal Classification and Identification and crystalline quality control method based on image

Publications (2)

Publication Number Publication Date
CN106599902A CN106599902A (en) 2017-04-26
CN106599902B true CN106599902B (en) 2019-09-24

Family

ID=58590887

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610971626.6A CN106599902B (en) 2016-11-04 2016-11-04 A kind of sugar crystal Classification and Identification and crystalline quality control method based on image

Country Status (1)

Country Link
CN (1) CN106599902B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102855494A (en) * 2012-08-20 2013-01-02 中国测绘科学研究院 Method and device for extracting water body of satellite remote sensing image
CN103018247A (en) * 2012-11-29 2013-04-03 天津理工大学 Urine cell image classification and identification system and method
CN104599276A (en) * 2015-01-27 2015-05-06 广西大学 Sugar crystal false three-dimensional microscopic image processing method and device for sugar boiling process
CN105338236A (en) * 2014-07-25 2016-02-17 诺基亚技术有限公司 Method and apparatus for detecting object in image and electronic device
CN105378453A (en) * 2012-12-19 2016-03-02 皇家飞利浦有限公司 System and method for classification of particles in a fluid sample

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102855494A (en) * 2012-08-20 2013-01-02 中国测绘科学研究院 Method and device for extracting water body of satellite remote sensing image
CN103018247A (en) * 2012-11-29 2013-04-03 天津理工大学 Urine cell image classification and identification system and method
CN105378453A (en) * 2012-12-19 2016-03-02 皇家飞利浦有限公司 System and method for classification of particles in a fluid sample
CN105338236A (en) * 2014-07-25 2016-02-17 诺基亚技术有限公司 Method and apparatus for detecting object in image and electronic device
CN104599276A (en) * 2015-01-27 2015-05-06 广西大学 Sugar crystal false three-dimensional microscopic image processing method and device for sugar boiling process

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于图像的蛋白质晶体分类方法研究;杨曦;《中国优秀博硕士学位论文全文数据库(硕士)-基础科技辑》;20070615(第06期);摘要,第6,11,15,20,44页 *

Also Published As

Publication number Publication date
CN106599902A (en) 2017-04-26

Similar Documents

Publication Publication Date Title
Jantzen et al. Pap-smear benchmark data for pattern classification
US6092059A (en) Automatic classifier for real time inspection and classification
US6246785B1 (en) Automated, microscope-assisted examination process of tissue or bodily fluid samples
AU699884B2 (en) Method and apparatus for continuously monitoring and forecasting slide and specimen preparation for a biological specimen population
US4965725A (en) Neural network based automated cytological specimen classification system and method
US5544650A (en) Automated specimen classification system and method
Leemans et al. AE—automation and emerging technologies: On-line fruit grading according to their external quality using machine vision
US4338024A (en) Flow analyzer and system for analysis of fluids with particles
US5933519A (en) Cytological slide scoring apparatus
Shahin et al. A machine vision system for grading lentils
US5436978A (en) Method and an apparatus for differentiating a sample of biological cells
US20040126008A1 (en) Analyte recognition for urinalysis diagnostic system
US7254266B2 (en) Method and apparatus for generating special-purpose image analysis algorithms
Naim et al. Automated morphological classification of APM galaxies by supervised artificial neural networks
Davis et al. Real-time observation of taxa-specific plankton distributions: an optical sampling method
ES2123475T3 (en) Method for integrating an automated system to a laboratory.
US20020031255A1 (en) Multi-neural net imaging apparatus and method
EP3388833A1 (en) Method and apparatus for automated whole blood sample analyses from microscopy images
WO1981003080A1 (en) Method and apparatus for measuring mean cell volume of red blood cells
US7739064B1 (en) Inline clustered defect reduction
EP0562506A2 (en) Method and device for sorting bulk material
DE3503475C2 (en)
EP0745243B9 (en) Automated cytological specimen classification methods
CN104392292B (en) Graphical re-inspection user setup interface
US20070016507A1 (en) System for analysis and prediction of financial and statistical data

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