CN103226088B - Particulate counting method - Google Patents
Particulate counting method Download PDFInfo
- Publication number
- CN103226088B CN103226088B CN201310119542.6A CN201310119542A CN103226088B CN 103226088 B CN103226088 B CN 103226088B CN 201310119542 A CN201310119542 A CN 201310119542A CN 103226088 B CN103226088 B CN 103226088B
- Authority
- CN
- China
- Prior art keywords
- particle
- counting
- image
- identification
- machine learning
- 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.)
- Active
Links
Landscapes
- Image Processing (AREA)
Abstract
The present invention discloses a particulate counting method and device thereof, wherein the method includes the steps of collecting the particulate image through camera device, identifying and counting the particulate with intelligent image processing and machine learning technique, placing a transparent tray in photographic area, which above, left, right and back sides are respectively provided with a group of material lighting source with adjustable angle and light intensity, placing camera device communicating with the computer on the transparent tray, calling the device driver of camera device via interface, using the keyboard and mouse to take photos for particles while managing and processing image data with counting module, rapidly pre-identifying the image photographed, screening out unsatisfactory photos to complete accurate counting. The invention has advantages of rapid processing, strong consistency of results and batch processing of modern image processing, so as to count particles efficiently, rapidly and accurately.
Description
Technical field
The invention belongs to areas of information technology, relate to a kind of particle counting method and device thereof.
Background technology
So-called mass of 1000 kernel is with a gram weight for 1,000 grain (seed) represented, and it is the index embodying grain size and turgor, and being the content of inspection grain quality and crop species test, is also the important evaluation index in wine brewing process.Mensuration at present for mass of 1000 kernel index mainly realizes based on the mode of artificial counting, and namely number is got 1,000 grain (seed) and obtained grain weight by weighing mode.Adopt the method for this artificial counting, due to easily tired after the long intensive work of human eye, count accuracy is difficult to ensure, and counting efficiency is lower.In order to improve counting efficiency, occurred the technique and method that various employing recognition technology counts grain at present, such as following patent discloses the method or equipment that adopt recognition technology to count grain (particle).
Application number is " 02126213.6 ", the patent of invention that name is called " adopting the method and apparatus of scattered light histogram fast distinguishing particles " discloses unique method and the equipment that rapid identification is suspended in the microscopic particles of protozoa and other microorganisms and so in liquid or gas, the intense light source that the method comprises employing laser and so on irradiates the particle that will detect, by one group of light sensors scattered light around detection zone, convert the light detected to electric signal, at least one occurrence frequency/probability histogram is adopted to compare the signal of derivation, thus qualitative and/or quantitative judge is carried out to the microscopic particles existed.This patented method is mainly used in the microscopic particles identifying the protozoan and other microorganism and so on be suspended in liquid or gas, for the suspended particulates identification under steric environment, there is certain advantage, but the deficiency that the method exists as the identification of the non-suspended particle of plane is: (1) utilizes light sensing equipment to have high requirements for equipment itself and experimental situation, and apparatus expensive is unfavorable for industrial applications; (2) utilize light scattering technique generally to need by multiple angle shot to reach the location of particle for the particle recognition in stereo-picture, and for the particulate identification of plane picture, the method is also not suitable for; (3) the method only relates to the identification of particulate (particulate group), and does not carry out segmentation consideration to heap superimposition adhesion situation.
Application number is called for the patent of invention of " 201110267255.0 " just discloses name the automatic division method of graininess object " in a kind of digital picture ", the method is for digital picture, especially the feature such as the gray scale of graininess object, structure distribution and geometric configuration in micro-image, first applying automatic threshold method is separated object and background; Then calculate its gradient vector field, in gradient vector field, search for key point, desirable key point all has corresponding gradient vector to distribute at 8 neighborhoods, and its Grad is zero, and the key point of acquisition is as the center of each graininess object; Then define a new effective energy function based on gray scale and locus in order to calculated direction gradient, replaced traditional shade of gray; Finally apply the border that active contour model searches graininess object.This patented method still proposes high requirement for the environment and image capturing environment that identify object, namely the method is only applicable to the particle identification in MIcrosope image, the general background of MIcrosope image is single, and there is fluorescence labeling as identification target, therefore there is notable difference with the particle identification of non-MIcrosope image that collects under common shooting environmental, and whether be applicable to this application scenarios and there is certain query.
The patent of invention that application number is " 95102156.7 " discloses name and is called " cereal grain color sorting apparatus ", the present invention, by both detecting the foreign matter different from grain with removing color in visible-range with a sorter, also identifies the foreign matter identical or transparent with grain with removing color near infrared range.This patent only relates to physical color identification method for sorting, and the counting for grain does not propose solution, and namely the method has some superiority for the Quality Detection of grain, but identifies not practical use for the counting of grain.In addition, the method needs applicable near infrared gear to detect, and the equipment cost related to during the cost of this equipment makes to detect and human cost increase greatly.Therefore, the method is inapplicable for grain counting and the mensuration of mass of 1000 kernel.
Application number is called " granular material sorting classifying method of view-based access control model identification " for the patent of invention of " 200810052381.2 " discloses name, and it comprises and obtains sample image, image feature information pre-service, design category device, sets up characteristic information experts database and material sorting to be selected is carried out with characteristic information experts database mating, carrying out sorting classifying.The method is applicable to the sizing screening of granule materials, and the enumeration problem therefore for granule materials does not provide solution.Meanwhile, the method relates to relatively loaded down with trivial details flow process, defines the scope of application and the application scenarios of system to a certain extent.
In sum, although current existing technology improves accuracy to grain count and counting efficiency, achieve Aulomatizeted Detect, but these technology need specialty and expensive equipment (such as laser or near infrared) usually, or it is higher to image quality requirements, complicated structure simultaneously, requires higher to operating personnel.Be not suitable for light-weighted solution in site work.
Summary of the invention
The present invention is directed to the deficiency existed in existing particle identification method of counting, provide a kind of particle counting method and device thereof.
The present invention is achieved by following technical solution.
A kind of particle counting method, camera arrangement is utilized to carry out particle image acquisition, intelligent image process and machine learning techniques is used to realize counting the identification of particle, while taking pictures, image data management and process is carried out by counting module, image pre-identification is fast carried out when having taken pictures, filter out undesirable photo, complete accurate counting, its Main process steps is as follows:
(1) particle image acquisition, by the device drives of interface interchange camera, uses keyboard and mouse to handle and takes pictures;
(2) the particle image of collection is converted into RGB numerical matrix, regulates the light and shade of image, strengthen contrast;
(3) K-Means clustering algorithm automatic screening effective coverage is utilized;
(4) limb recognition technology identification independently particle is utilized;
(5) modeling is carried out to the individual particles identified, obtain the statistical model of the profile shape characteristic about particle, size characteristic;
(6) estimate the adhesion block of suspection, carry out simulation according to individual particles model and fill, estimate the number comprised, and export adhesion rule and cutting effect figure;
(7) the cutting effect figure exported manually is passed judgment on, the precision of identification is calculated in conjunction with artificial counting result, the method of machine learning is used to learn identification and the counting of the particle in this situation to the mode ancillary computer systems that there is adhesion and stacking region employing artificial counting, if identify, the accuracy rate of counting is greater than 95%, complete the counting to particle, and export count results, otherwise, the result of artificial cognition is replaced the adhesion block estimated, re-training model, until machine learning system model reaches stable.
In above-mentioned steps (1), particle image acquisition comprises the steps:
A, by aerosol sample take out be positioned on transparent panel or blank;
B, tiling particle, make particle be covered with the viewfinder range of camera installation equably;
C, collection image;
D, repeat b and step c twice to reach the random object of sampling, obtain n*3 and open image, n is sample size.
As another object of the present invention, provide a kind of particle counting device, comprise transparent pallet and be placed in the material lighting source of transparent pallet surrounding, camera installation is placed with above transparent pallet, camera installation is connected with computing machine by data line, by the device drives of interface interchange camera installation, use keyboard and mouse to handle and take pictures.
The angle of described material lighting source can regulate.
Described material lighting source is the adjustable White LED light source of light intensity.
Described material lighting source comprises one group and is placed in left side lamp on the left of transparent pallet, is placed in the starboard light on the right side of transparent pallet and is placed in the back of the body lamp of transparent tray back.
The invention has the beneficial effects as follows:
Compared with prior art, the present invention uses intelligent image process and machine learning techniques to realize counting the quick and precisely identification of particle, by particle (grain) image that obtains using particle image acquisition as handling object, first utilize image recognition to count particle (grain) overlay area is chosen automatically, and the grain of some non-complex regions (there is not adhesion and stacking region) is directly counted.For there is adhesion and stacking region, carry out Intelligent Recognition and counting by the method for machine learning.
The present invention only need adopt the camera arrangement of a simple and easy installation of energy and common family expenses camera to take pictures to a collection of particle, then image file is imported computing machine, uses Modling model to process image.Adopt image recognition technology accurately to identify independently particle, utilize statistical model identify for dissimilar particle (Chinese sorghum, paddy rice etc.) and estimate the particle of adhesion, finally counted accurately.And use the mode of artificial counting to detect result, and according to the result re-training model revised, the precision of continuous lift scheme.The camera arrangement that the present invention adopts is installed and is disposed easily, can be quick installed at multiple environment.Not high to the requirement of place light source and photographic equipment yet, can the photographic intelligence of quick obtaining particle.By with the statistical model of manual intervention and machine learning method, detection and training can be combined, add the precision of computer recognizing.The production environment of particle counting can be applied to easily, replace the mode of artificial counting, increase work efficiency significantly.The present invention has given full play to that the processing speed of modern image handle is fast, result consistance strong, the advantage of batch processing, thus realizes efficiently, quickly and accurately counting particle.Whole mensuration process manually relies on low, shows good repeatability and high efficiency, and camera arrangement is not high to the quality requirements of image simultaneously, extensively can promote the use of in particle counting field.
Accompanying drawing explanation
Fig. 1 is process flow diagram of the present invention;
Fig. 2 is counting assembly structural representation in the present invention.
In figure: the transparent pallet of 1-, lamp on the left of 2-, 3-carries on the back lamp, 4-starboard light, 5-camera installation, 6-computing machine, 7-data line, 8-particle.
Embodiment
Technical scheme of the present invention is further described below in conjunction with embodiment, but described in claimed scope is not limited to.
As shown in Figure 1, a kind of particle counting method, camera arrangement is utilized to carry out particle image acquisition, intelligent image process and machine learning techniques is used to realize counting the identification of particle, while taking pictures, carry out image data management and process by counting module, carrying out image pre-identification fast when having taken pictures, filtering out undesirable photo, complete accurate counting, its Main process steps is as follows:
(1) particle image acquisition, by the device drives of interface interchange camera, uses keyboard and mouse to handle and takes pictures;
(2) the particle image of collection is converted into RGB numerical matrix, regulates the light and shade of image, strengthen contrast;
(3) K-Means clustering algorithm automatic screening effective coverage is utilized;
K-Means algorithm described in this step is input cluster number k, and comprises the database of n data object, exports k the cluster meeting variance minimum sandards.K-means algorithm accepts input quantity k; Then n data object is divided into k cluster to make obtained cluster meet, the object similarity in same cluster is higher; And object similarity in different cluster is less.Cluster similarity be utilize the average of object in each cluster obtain center object (center of attraction) coming and carry out calculating.
(4) limb recognition technology identification independently particle is utilized;
Limb recognition technology described in this step is extracted the circumference of particle (grain), and obtaining intermediate treatment object particle (grain) region (namely not comprising background whiteboard area), its algorithm comprises the Image outline identification based on wavelet transformation, the Image outline identification based on the robert factor, the Image outline identification based on the laplace operator factor, the outline identification based on gaussian filtering.
(5) modeling is carried out to the individual particles identified, obtain the statistical model of the profile shape characteristic about particle, size characteristic;
(6) estimate the adhesion block of suspection, carry out simulation according to individual particles model and fill, estimate the number comprised, and export adhesion rule and cutting effect figure;
(7) the cutting effect figure exported manually is passed judgment on, the precision of identification is calculated in conjunction with artificial counting result, the method of machine learning is used to learn identification and the counting of the particle in this situation to the mode ancillary computer systems that there is adhesion and stacking region employing artificial counting, if identify, the accuracy rate of counting is greater than 95%, complete the counting to particle, and export count results, otherwise, the result of artificial cognition is replaced the adhesion block estimated, re-training model, until machine learning system model reaches stable.
In above-mentioned steps (1), particle image acquisition comprises the steps:
A, by aerosol sample take out be positioned on transparent panel or blank;
B, tiling particle, make particle be covered with the viewfinder range of camera installation equably;
C, collection image;
D, repeat b and step c twice to reach the random object of sampling, obtain n*3 and open image, n is sample size.
Counting module described in the present invention can realize the management processing to view data, allows user to be gathered by visualization interface management and obtains image, and realize counting the identification of particle (grain) image by image recognition and machine learning method.Machine Learning Theory mainly design and analysis some allow computing machine can the algorithm of " study " automatically.Machine learning algorithm is that class automatic analysis from data obtains rule, and the algorithm that assimilated equations is predicted unknown data.Because relate to a large amount of statistical theories in learning algorithm, machine learning and statistical inference student's federation are particularly close, are also referred to as Statistical Learning Theory.Therefore, the machine learning model of the method for the invention by training, can for there is adhesion and stacking particle (grain) region identifies intelligently and counts.
As shown in Figure 2, for realizing above-mentioned particle counting method, the invention provides a kind of particle counting device, comprise transparent pallet 1 and be placed in the material lighting source of transparent pallet 1 surrounding, camera installation 5 is placed with above transparent pallet 1, camera installation 5 is connected with computing machine 6 by data line 7, by the device drives of interface interchange camera installation 5, uses keyboard and mouse to handle and takes pictures.
The angle of described material lighting source can regulate.
Described material lighting source is the adjustable White LED light source of light intensity.
Described material lighting source comprises one group and is placed in left side lamp 2 on the left of transparent pallet 1, is placed in the starboard light 4 on the right side of transparent pallet 1 and is placed in the back of the body lamp 3 at transparent pallet 1 back side.
The present invention institute in use, a transparent pallet 1 is placed at photograph zone level, above transparent pallet 1, left side, right side and the back side respectively arrange one group of angle and the adjustable material lighting source of light intensity, material lighting source is respectively and is placed in left side lamp 2 on the left of transparent pallet 1, be placed in the starboard light 4 on the right side of transparent pallet 1 and be placed in the back of the body lamp 3 at transparent pallet 1 back side, and is high brightness White LED light source.Camera installation 5 is positioned over transparent pallet 1 top.When needs are taken pictures, particle 8 is interspersed among on transparent pallet 1, and random applied particle 8.Camera installation 5 is connected to computing machine 6 by data line 7, by the device drives of interface interchange camera installation 5, uses keyboard and mouse to handle and take pictures.While taking pictures, carry out image data management and process by counting module, carrying out image pre-identification fast when having taken pictures, filtering out undesirable photo.According to each testing result training pattern and the parameter of adjustment model, the precision of continuous lift scheme.
If the present invention does not possess above device in use, can also replace in the following manner: particle (grain) sample is laid on the flat board of white (or Transparent color), about 50 centimeters erection image capture devices above slab normal, ensure that light source and image capture device focal length stablize the collection and counting that can complete particle image, structure is simple, easy to operate.
Claims (2)
1. a particle counting method, it is characterized in that: utilize camera arrangement to carry out particle image acquisition, intelligent image process and machine learning techniques is used to realize counting the identification of particle, while taking pictures, image data management and process is carried out by counting module, image pre-identification is fast carried out when having taken pictures, filter out undesirable photo, complete accurate counting, its Main process steps is as follows:
(1) particle image acquisition, by the device drives of interface interchange camera, uses keyboard and mouse to handle and takes pictures;
(2) the particle image of collection is converted into RGB numerical matrix, regulates the light and shade of image, strengthen contrast;
(3) K-Means clustering algorithm automatic screening effective coverage is utilized;
(4) limb recognition technology identification independently particle is utilized;
(5) modeling is carried out to the individual particles identified, obtain the statistical model of the profile shape characteristic about particle, size characteristic;
(6) estimate the adhesion block of suspection, carry out simulation according to individual particles model and fill, estimate the number comprised, and export adhesion rule and cutting effect figure;
(7) the cutting effect figure exported manually is passed judgment on, the precision of identification is calculated in conjunction with artificial counting result, the method of machine learning is used to learn identification and the counting of the particle in this situation to the mode ancillary computer systems that there is adhesion and stacking region employing artificial counting, if identify, the accuracy rate of counting is greater than 95%, complete the counting to particle, and export count results, otherwise, the result of artificial cognition is replaced the adhesion block estimated, re-training model, until machine learning system model reaches stable.
2. a kind of particle counting method according to claim 1, is characterized in that: in described step (1), particle image acquisition comprises the steps:
A, by aerosol sample take out be positioned on transparent panel or blank;
B, tiling particle, make particle be covered with the viewfinder range of camera installation equably;
C, collection image;
D, repeat b and step c twice to reach the random object of sampling, obtain n*3 and open image, n is sample size.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310119542.6A CN103226088B (en) | 2013-04-08 | 2013-04-08 | Particulate counting method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310119542.6A CN103226088B (en) | 2013-04-08 | 2013-04-08 | Particulate counting method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103226088A CN103226088A (en) | 2013-07-31 |
CN103226088B true CN103226088B (en) | 2015-04-22 |
Family
ID=48836616
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310119542.6A Active CN103226088B (en) | 2013-04-08 | 2013-04-08 | Particulate counting method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103226088B (en) |
Families Citing this family (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104034638B (en) * | 2014-06-26 | 2016-08-17 | 芜湖哈特机器人产业技术研究院有限公司 | The diamond wire online quality detecting method of granule based on machine vision |
CN105405145B (en) * | 2015-11-17 | 2019-07-16 | 四川飞拓信息技术有限公司 | A kind of granule number grain method based on image Segmentation Technology |
CN106295705B (en) * | 2016-08-17 | 2019-04-05 | 广东三三智能科技有限公司 | A kind of more color materials under movement background screen number system |
CN110178012B (en) * | 2016-12-16 | 2022-04-05 | 艾珀尔有限公司 | Classification analysis method, classification analysis device, and recording medium for classification analysis |
CN106898010A (en) * | 2017-03-01 | 2017-06-27 | 上海市农业科学院 | Particle copies the method and device planted |
CA3003118A1 (en) | 2017-04-28 | 2018-10-28 | The Royal Institution For The Advancement Of Learning/Mcgill University | Apparatus for analyzing a sample of granular material |
EP3623793B1 (en) * | 2017-05-07 | 2022-09-07 | Aipore Inc. | Identification classification analysis method and classification analysis device |
CN107545563B (en) * | 2017-07-13 | 2023-07-14 | 上海造币有限公司 | Strip punching counting system and counting method |
CN109345505A (en) * | 2018-08-09 | 2019-02-15 | 北京木业邦科技有限公司 | A kind of material checks equipment |
WO2020072031A1 (en) * | 2018-10-01 | 2020-04-09 | Hewlett-Packard Development Company, L.P. | Batch particle sorting |
CN109598328B (en) * | 2018-11-21 | 2023-09-12 | 山东农业大学 | Distributed thousand grain counting method, system, device and terminal |
CN109655385B (en) * | 2019-02-19 | 2024-01-23 | 西北农林科技大学 | Cereal phenotype integration detection device |
CN109948531B (en) * | 2019-03-19 | 2023-04-18 | 广东省农业科学院农业生物基因研究中心 | Method for improving crop seed germination rate by applying deep learning convolutional neural network |
CN109919248B (en) * | 2019-03-19 | 2023-04-18 | 广东省农业科学院农业生物基因研究中心 | Rice seed testing device based on convolutional neural network |
CN109934297B (en) * | 2019-03-19 | 2023-04-18 | 广东省农业科学院农业生物基因研究中心 | Rice seed test method based on deep learning convolutional neural network |
CN111272626A (en) * | 2020-01-19 | 2020-06-12 | 上海化工研究院有限公司 | Method for testing interception efficiency of non-woven filter material for liquid-solid separation |
CN111353977A (en) * | 2020-02-26 | 2020-06-30 | 杭州电子科技大学 | Small particle object counting system and method based on deep learning |
CN112760831A (en) * | 2020-12-30 | 2021-05-07 | 西安标准工业股份有限公司 | Intelligent piece counting method and system based on sewing equipment |
CN116242777B (en) * | 2023-03-24 | 2023-11-28 | 江苏乾禧环保科技有限公司 | Multi-stage screening quality judging system for ceramsite preparation based on machine vision |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100578192C (en) * | 2005-08-15 | 2010-01-06 | 中国农业大学 | Apparatus for measuring spectrum of fog drop, and image processing device |
CN101281112A (en) * | 2008-04-30 | 2008-10-08 | 浙江理工大学 | Image type automatic analysis method for mesh adhesion rice corn |
CN101441721B (en) * | 2008-11-28 | 2010-11-17 | 江苏大学 | Device and method for counting overlapped circular particulate matter |
CN102692364B (en) * | 2012-06-25 | 2014-05-28 | 上海理工大学 | Blurring image processing-based dynamic grain measuring device and method |
CN203177944U (en) * | 2013-04-08 | 2013-09-04 | 贵州茅台酒股份有限公司 | Particle counting apparatus |
-
2013
- 2013-04-08 CN CN201310119542.6A patent/CN103226088B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN103226088A (en) | 2013-07-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103226088B (en) | Particulate counting method | |
CN104599275B (en) | The RGB-D scene understanding methods of imparametrization based on probability graph model | |
Rachmadi et al. | Vehicle color recognition using convolutional neural network | |
CN106682633B (en) | The classifying identification method of stool examination image visible component based on machine vision | |
CN105809146B (en) | A kind of image scene recognition methods and device | |
CN103984939B (en) | A kind of sample visible component sorting technique and system | |
CN108171184A (en) | Method for distinguishing is known based on Siamese networks again for pedestrian | |
CN109952614A (en) | The categorizing system and method for biomone | |
CN108269250A (en) | Method and apparatus based on convolutional neural networks assessment quality of human face image | |
CN109684906B (en) | Method for detecting red fat bark beetles based on deep learning | |
CN110276386A (en) | A kind of apple grading method and system based on machine vision | |
CN203177944U (en) | Particle counting apparatus | |
CN108229561A (en) | Particle product defect detection method based on deep learning | |
CN103020639A (en) | Method for automatically identifying and counting white blood cells | |
US10769432B2 (en) | Automated parameterization image pattern recognition method | |
CN103345617A (en) | Method and system for recognizing traditional Chinese medicine | |
CN104021375A (en) | Model identification method based on machine learning | |
CN101441721A (en) | Device and method for counting overlapped circular particulate matter | |
CN103177266A (en) | Intelligent stock pest identification system | |
CN103440035A (en) | Gesture recognition system in three-dimensional space and recognition method thereof | |
CN109815945A (en) | A kind of respiratory tract inspection result interpreting system and method based on image recognition | |
CN107330478A (en) | A kind of cherry tomato sorting technique, system and cherry tomato on-line sorting system | |
CN107103608A (en) | A kind of conspicuousness detection method based on region candidate samples selection | |
CN110263608A (en) | Electronic component automatic identifying method based on image feature space variable threshold value metric | |
CN106228136A (en) | Panorama streetscape method for secret protection based on converging channels feature |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |