CN105260730A - Machine learning-based contact-type imaging microfluid cell counter and image processing method thereof - Google Patents
Machine learning-based contact-type imaging microfluid cell counter and image processing method thereof Download PDFInfo
- Publication number
- CN105260730A CN105260730A CN201510824763.2A CN201510824763A CN105260730A CN 105260730 A CN105260730 A CN 105260730A CN 201510824763 A CN201510824763 A CN 201510824763A CN 105260730 A CN105260730 A CN 105260730A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- sensor chip
- cell
- image sensor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/698—Matching; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention provides a machine learning-based single-frame super-resolution contact-type imaging microfluid cell counter, which comprises a microfluid pipeline 101, an image sensor chip 102, and a single-frame super-resolution image processing module 103 based on machine learning. The CMOS image sensor chip 102 which can be massively produced and is low in price is adopted to realize lens-free contact-type imaging, an expensive optical lens can be prevented from being used, and the cost and the size are reduced. The single-frame super-resolution image processing algorithm based on machine learning comprises a step of offline training and a step of online detection, thereby improving the resolution of a grabbed image and improving recognition and counting accuracy of a cell 104. Meanwhile, compared with multi-frame super-resolution algorithm, the single-frame algorithm does not need to limit a fluid speed to grab multiple reference images with sub pixel displacement, burden of image storage does not exist, the detection flux of the system can be improved, and cell morphology recognition and cell number counting at a high fluid speed can be ensured.
Description
Technical field
The present invention relates to the single frames super-resolution contact imaging microfluidic cell counter based on machine learning, belong to micro-fluidic Imaging biological medical detection technology.
Background technology
At present, the flow cytometry that automatically can record one or more cell quantities and concentration in solution is all widely used in biological study and clinical diagnosis, and such as, in HIV test-and-treat, it can detect the T lymphocyte quantity of CD4+ and CD8+.These application require that flow cytometer has high flux and pin-point accuracy.The principle of traditional flow cytometer makes cell aggregation become very narrow linear fashion quickly through laser beam, and by detecting direct scattering simultaneously, the fluorescence signal of lateral scattering and transmitting judges the information such as quantity, size of cell.But the Systems for optical inspection meticulous due to complexity is expensive, traditional flow cytometer is unfavorable for that bedside detects.In addition, this method counted by optical scatter signals does not inherently possess image Detection Information intuitively.In recent years, along with the development based on laboratory technique on micro-fluidic sheet, the flow cytometer of miniaturization becomes possibility.By micro-fluidic chip being integrated on complementary metal oxide semiconductor (CMOS) (CMOS) image sensor chip, can realize based on micro-fluidic without camera lens contact imaging system.When there being light source irradiation chip surface, this system directly can obtain the shadow image of cell to be measured projection.According to the imaging of these contacts, the Detection Information such as the quantity of cell next can be obtained.
But owing to adopting the contact imaging not having camera lens to amplify, the cytological map that image sensor chip obtains is in fact all low resolution, lost the detailed information of cellular morphology.(SuT-W, SeoS, ErlingerA, andOzcanA, " High-throughputlensfreeimagingandcharacterizationofahete rogeneouscellsolutiononachip; " BiotechnologyandBioengineering, vol.102, pp.856 – 868,2009) achieve a LUCAS system, cell to be measured directly easily drops on the glass on image sensor chip surface by it, judges cell category by the intensity distributions analyzing shadow image.Owing to not having continuous print cell to flow, therefore, the detection flux of LUCAS system is very low.(ZhengG, LeeSALee, YangS, YangC-H, " Sub-pixelresolvingoptofluidicmicroscopeforon-chipcellima ging, " LabonaChip, vol.10, pp.3125 – 3129,2010; LeeSA, LeitaoR, ZhengG, YangS, RodriguezA, YangC-H, " Colorcapablesub-pixelresolvingoptofluidicmicroscopeandit sapplicationtobloodcellimagingformalariadiagnosis; " PLOSONE, vol.6, pp.e26127,2011) achieve one adopt multiframe super-resolution image Processing Algorithm without lens imaging system, the reference frame image of this system by capturing 40 to 100 continuously, and therefrom recover the high-resolution image of a frame.Although this system improves the resolution of imaging, owing to needing to capture a series of cytological map with Displacement, it can only be instilled solution and be flowed by capillarity, and therefore the detection flux of system is still not high, and the storage space needed is also very large.
Based on above Problems existing, we propose the present invention, a kind of single frames super-resolution contact imaging microfluidic cell counter based on machine learning and image processing method thereof.
Summary of the invention
Goal of the invention: the flow cytometry for live biometric medical diagnosis again to realize low cost, miniaturization with high flux, pin-point accuracy simultaneously, the present invention proposes a kind of single frames super-resolution contact imaging microfluidic cell counter based on machine learning.By adopting without the imaging of camera lens contact, the use of expensive optical lens can be removed, and the cmos image sensor chip adopting mass producible cheap, reduce cost and volume.By adopting the single frames super-resolution image Processing Algorithm based on machine learning, the accuracy of systems axiol-ogy can be improved.Meanwhile, relative to the super-resolution algorithms of multiframe, single frames algorithm does not need limited flow rate to capture the reference picture that multiple have Displacement, therefore can not bring the burden that image stores yet, and can improve the detection flux of system.
Content of the present invention can be divided into two parts: (1) for low resolution cell image gather without the micro-fluidic imaging system of camera lens contact; (2) based on the single frames super-resolution image process that machine learning realizes.
Based on the single frames super-resolution contact imaging microfluidic cell counter of machine learning, comprise: Micro-flow pipe (101) also comprises image sensor chip (102) with Micro-flow pipe (101) composition diagram as collecting part, light source (105) irradiates cell to be measured (104) sample liquid Micro-flow pipe (101) from image sensor chip (102) and Micro-flow pipe (101) top, image sensor chip (102) continues the image under the projection of seizure cell, single frames super-resolution image processing module (103) captures the processing section of image as image sensor chip (102).
The view data that image sensor chip (102) captures is transferred on external computer by the interface on P.e.c., can carry out next step image procossing.Time shutter of image sensor chip (102), area-of-interest and the number of image frames that will capture can send signal by computer and control.
The cmos image sensor chip that described image sensor chip (102) can adopt cmos image sensor chip or CCD chip or customize.
Described single frames super-resolution image processing module (103), directly can also be become with image acquisition segment set by Design of Digital Integrated Circuit mode in the image sensor chip (102) of customization.The view data that image sensor chip (102) captures is transferred on external computer by the interface on P.e.c., can carry out next step image procossing.Time shutter of image sensor chip (102), area-of-interest and the number of image frames that will capture can send signal by computer and control.
Single frames super-resolution image processing module (103) algorithm is through the description of digital programmable language, adopt the instrument of Design of Digital Integrated Circuit, coded description can be carried out the copying checking of front end, formal verification, logic synthesis, and the placement-and-routing of rear end, parasitic parameter extraction, sequential power consumption analysis, after satisfying condition, GDS layout file corresponding with image sensor chip (102) the image acquisition part of customization for the GDS layout file of now single frames super-resolution image processing module (103) correspondence generation can be stitched together, form a complete chip layout, the GDS file of this overall domain is delivered to integrated circuit processing factory and carries out flow, namely the cmos image sensor chip manufacturing of customization is completed.
Because the initial cell image collected has lower resolution, therefore need to carry out superresolution processing to improve resolution, for follow-up cell type identification and cell quantity statistics are prepared.For contact imaging microfluidic cell counter, the present invention adopts the machine learning method based on extreme learning machine to carry out the superresolution processing of single-frame images.
Superresolution processing based on extreme learning machine can be divided into two steps: off-line training and on-line checkingi.In off-line training, train model with the characteristic component obtained from the extraction radio-frequency component in high-definition picture and corresponding low-resolution image; In on-line checkingi process, obtain the corresponding radio-frequency component of low-resolution image according to training pattern, low-resolution image and radio-frequency component combine and obtain corresponding high-definition picture.
Superresolution processing training process based on extreme learning machine is as follows:
1) initialization t Im × n low-resolution image LR and t IM × N (HR) high-definition picture HR;
2) target HF row vector T is produced;
3) input low-resolution image LR is made to become IM × N size from size Im × n by bicubic interpolation;
4) t radio-frequency component image IM × N (HF)=IM × N (HR) – IM × N (LR_int) is obtained;
5) row vector T is produced based on IM × N (HF);
6) eigenmatrix X is produced;
7) picture element density distribution is extracted by 3 × 3 block of pixels from t IM × N (LR_int);
8) weight vectors between L hidden place to output point is produced;
9) produce matrix A, B at random, use Sigmoid type function G;
10)T＝βH(X)＝βG(AX+B)
11) C is adjustment parameter;
First, the high-definition image of a series of cell under difference rotation form is obtained by high resolution microscope.Obtain corresponding low-resolution image by bicubic down-sampling afterwards, then the size of this low-resolution image is changed to the size of original high resolution image by the mode of interpolation.Then, high-frequency component is subtracted each other by low-resolution image after interpolation and high-definition picture and is obtained, and as one of training objective---row vector T.Afterwards, the low-resolution image of interpolation obtains image pixel intensities pattern after 3 × 3 block of pixels longitudinally process, and it single order second order comprised on 9 pixel intensity values, 5 horizontal and vertical directions is led.Obtain eigenmatrix X thus.Finally, matrix X and T is used to training ELM-SR model.
12) flow process of on-line checkingi is as follows: input low-resolution image LRIm × n for checking (LR ');
13) eigenmatrix X ' is produced;
14) transformation of Im × n (LR ') to IM × N (LR_int ') is realized by bicubic interpolation;
15) from IM × N (LR_int '), pixel intensity distribution is extracted by 3 × 3 block of pixels;
16) radio-frequency component IM × N (HF ') is calculated;
17) by being added with the low-resolution image after interpolation by the radio-frequency component recovered, final high resolution output is obtained.
First, by the mode of interpolation, make the size of the low-resolution image of input be consistent with the adjusted size of high score rate image.Afterwards by identical processing mode, from the low-resolution image after interpolation, obtain eigenmatrix X '.Then, X ' is input in training pattern, obtains corresponding radio-frequency component HF '.Finally, by being added with the low-resolution image after interpolation by the radio-frequency component recovered, final high-definition picture is obtained.
The course of work is as follows:
Micro-flow pipe 101 is placed on image sensor chip 102 top, makes that cell is sustainable to be flow through from this passage.The height of Micro-flow pipe selects the diameter a little more than cell to be measured, make flow cell as far as possible close to imageing sensor surface thus obtain higher picture contrast; The chip area that the selection of length takes imageing sensor as far as possible makes larger cell imaging region; Width can be selected as far as possible wide to increase the blocking that systems axiol-ogy flux also reduces cell simultaneously.The material of Micro-flow pipe can select PDMS dimethyl silicone polymer.When carrying out cell detection, utilize syringe pump cell 104 sample liquid to be measured to be continued in injection channel, and control its flowing velocity.Peak Flow Rate needs to match with the frame rate of imageing sensor.In order to capture cell image, adopting light source 105, such as standard white light-emitting diode, irradiating from image sensor chip 102 top the cell 104 flowed sample liquid.Therefore, image sensor chip 102 can continue the image catching flow cell, and its Pixel Dimensions can change according to selecting of image sensor chip 102, usually takes as far as possible little Pixel Dimensions to improve the resolution of the original image captured.Afterwards, the single frames superresolution processing module based on machine learning carries out SUPERRESOLUTION PROCESSING FOR ACOUSTIC to the cell original image captured, and obtains high resolution graphics, then carries out identification and the counting of cell category according to the feature of image.
Beneficial effect
This invention solves resolution restriction and detects flux restriction two problems, can realize the online image recognition based on superresolution processing, ensures the cellular morphology identification under high flow rate and cell quantity statistics.Compared with traditional cell counter, this system can realize the error probability lower than 8% when adding up cell absolute quantity, meanwhile, can realize the cell recognition rate of 0.10 covariance when detecting many cells mixed liquor.
Compared with other super resolution algorithm, the extreme learning machine that this invention adopts has obvious advantage: because the weight coefficient between input point and hidden place is random generation, without the need to the repetitive exercise process of complexity, especially, when inputting huge number, training speed and training effectiveness are significantly improved.
Accompanying drawing illustrates:
Fig. 1. be single frames super-resolution contact imaging microfluidic cell counter sectional view; 101 is Micro-flow pipe, and 102 is image sensor chip, and 103 is the single frames super resolution image processing module based on machine learning, and 104 is cell to be measured, and 105 is light source.
Fig. 2. the off-line training for the single frames superresolution processing based on extreme learning machine and the process flow diagram at X-ray inspection X.
Embodiment
Based on the single frames super-resolution contact imaging microfluidic cell counter of machine learning, comprising: Micro-flow pipe 101, image sensor chip 102, based on the single frames super-resolution image processing module 103 that machine learning realizes, light source 105, as shown in Figure 1.Light source 105 projects downwards above Micro-flow pipe 101 with image sensor chip 102, the initial cell image collected due to image sensor chip 102 has lower resolution, therefore need to carry out superresolution processing to improve resolution, for follow-up cell type identification and cell quantity statistics are prepared.For contact imaging microfluidic cell counter, the raw image data that imageing sensor captures is transferred on external computer by the interface on printed circuit board (PCB), can carry out next step image procossing.Time shutter of image sensor chip 102, area-of-interest and the number of image frames that will capture can send signal by computer and control.In addition, also can by design cycle on the image sensor chip 102 of customization the hardware implementing of super-resolution image processing module by digital integrated circuit.
The present invention adopts the machine learning method based on extreme learning machine to carry out the superresolution processing of single-frame images, can be divided into two steps: off-line training and on-line checkingi.As shown in Figure 2, in off-line training, first adopt high-power microscope to obtain the high-definition picture of a series of sample cell as training input, train model with the characteristic component obtained in the radio-frequency component extracted from high-definition picture and corresponding low-resolution image; In on-line checkingi process, calculate the corresponding radio-frequency component of low-resolution image according to the training pattern obtained in off-line training, low-resolution image and high fdrequency component combine and obtain corresponding high-definition picture.
Principle of work after the system integration is good and engineering as follows.First, testing sample cell is prepared; Utilize syringe pump to continue in injection channel by cell sample liquid to be measured afterwards, control its flowing velocity; With standard white light-emitting diode as light source, irradiate from imageing sensor top the cell flowed sample liquid; Then, cmos image sensor can continue the image catching cell.Finally, improve by the single frames super-resolution image process realized based on machine learning the resolution capturing cell image, carry out next step cell recognition and counting.Compared with other super resolution algorithm, the extreme learning machine that this invention adopts has obvious advantage: because the weight coefficient between input point and hidden place is random generation, without the need to the repetitive exercise process of complexity, especially, when inputting huge number, training speed and training effectiveness are significantly improved.
According to foregoing summary, the cmos image sensor (AptinaMT9M032) that have employed one piece of gray scale makes detection chip, the pixel size of chip is 2.2 μm × 2.2 μm, detecting pixel region area is 3.24mm (H) × 2.41mm (V), and array size is 1472 (H) × 1096 (V).Imageing sensor and Micro-flow pipe integrated before, with oxygen plasma machine clean surface.In order to use chip detection area to greatest extent, Micro-flow pipe adopts direction diagonally, and length is 4.6mm, duct width 500 μm, height 30 μm.Cmos image sensor chip soldering is on 5.6cm × 5.6cm printed circuit board (PCB) of one piece of low cost.Communication between sensor chip and computer adopts USB chip.Have selected the imaging area interested of 640 × 480.
Through experiment test, designed micro-fluidic single frames super-resolution contact imaging microfluidic cell counter can obtain the miscount rate being less than 8%.Meanwhile, the cell recognition rate of 0.10 covariance can be realized when detecting RBC and HepG2 mixed liquor.
Claims (4)
1. based on the contact imaging microfluidic cell counter of machine learning, comprise Micro-flow pipe (101), it is characterized in that, also comprise image sensor chip (102) with Micro-flow pipe (101) composition diagram as collecting part, Micro-flow pipe (101) is directly integrated in the top of image sensor chip (102), light source (105) irradiates cell to be measured (104) sample liquid Micro-flow pipe (101) from image sensor chip (102) and Micro-flow pipe (101) top, when cell to be measured (104) continues to flow to delivery outlet from the input port of Micro-flow pipe (101), image sensor chip (102) continues the imaged image under the projection of seizure cell, single frames super-resolution image processing module (103) captures the processing section of image as image sensor chip (102).
2. counter as claimed in claim 1, it is characterized in that, described image sensor chip (102) can adopt ready-made commercial cmos image sensor chip or ccd image sensor chip, or the cmos image sensor chip of customization.
3. counter as claimed in claim 1, it is characterized in that, described single frames super-resolution image processing module (103), directly can also be become with image acquisition segment set by Design of Digital Integrated Circuit mode in the image sensor chip (102) of customization.
4. the image processing method of cell counter as claimed in claim 1, it is characterized in that, the resolution capturing cell image is improved by single frames super-resolution image processing module (103), carry out next step cell recognition and counting, be divided into two steps and off-line training and on-line checkingi to carry out superresolution processing, step is as follows:
Step one, off-line training, comprising:
1) initialization t Im × n low-resolution image LR and t IM × N (HR) high-definition picture HR;
2) target HF row vector T is produced;
3) input low-resolution image LR is made to become IM × N size from size Im × n by bicubic interpolation;
4) t radio-frequency component image IM × N (HF)=IM × N (HR) – IM × N (LR_int) is obtained;
5) row vector T is produced based on IM × N (HF);
6) eigenmatrix X is produced;
7) picture element density distribution is extracted by 3 × 3 block of pixels from t IM × N (LR_int);
8) weight vectors between L hidden place to output point is produced;
9) produce matrix A, B at random, use Sigmoid type function G;
10)T＝βH(X)＝βG(AX+B)；
11) C is adjustment parameter;
Step 2, on-line checkingi
12) the low-resolution image LRIm × n (LR ') of input for checking;
13) eigenmatrix X ' is produced;
14) transformation of Im × n (LR ') to IM × N (LR_int ') is realized by bicubic interpolation;
15) from IM × N (LR_int '), pixel intensity distribution is extracted by 3 × 3 block of pixels;
16) radio-frequency component IM × N (HF ') is calculated;
17) by being added with the low-resolution image after interpolation by the radio-frequency component recovered, final high resolution output is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510824763.2A CN105260730A (en) | 2015-11-24 | 2015-11-24 | Machine learning-based contact-type imaging microfluid cell counter and image processing method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510824763.2A CN105260730A (en) | 2015-11-24 | 2015-11-24 | Machine learning-based contact-type imaging microfluid cell counter and image processing method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105260730A true CN105260730A (en) | 2016-01-20 |
Family
ID=55100411
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510824763.2A Pending CN105260730A (en) | 2015-11-24 | 2015-11-24 | Machine learning-based contact-type imaging microfluid cell counter and image processing method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105260730A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107623827A (en) * | 2017-08-15 | 2018-01-23 | 上海集成电路研发中心有限公司 | A kind of intelligent CMOS image sensor chip and its manufacture method |
| WO2019071918A1 (en) * | 2017-10-11 | 2019-04-18 | 南京大学 | Platelet projection imaging-based detection device and method therefor |
| WO2019084517A1 (en) * | 2017-10-26 | 2019-05-02 | Essenlix Corporation | System and methods of image-based assay using crof and machine learning |
| CN111896456A (en) * | 2020-06-19 | 2020-11-06 | 北京凡谱科技有限公司 | Single cell analysis method based on micro-fluidic and hyperspectral imaging |
| CN114505105A (en) * | 2022-01-13 | 2022-05-17 | 电子科技大学 | Micro-fluidic chip based on memory calculation |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103606136A (en) * | 2013-12-04 | 2014-02-26 | 西安电子科技大学 | Video super resolution method based on keyframes and non-local restriction |
| CN103871041A (en) * | 2014-03-21 | 2014-06-18 | 上海交通大学 | Image super-resolution reconstruction method based on cognitive regularization parameters |
| CN204116189U (en) * | 2014-03-13 | 2015-01-21 | 严媚 | The imaging of intelligent optical miniflow and detection system |
-
2015
- 2015-11-24 CN CN201510824763.2A patent/CN105260730A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103606136A (en) * | 2013-12-04 | 2014-02-26 | 西安电子科技大学 | Video super resolution method based on keyframes and non-local restriction |
| CN204116189U (en) * | 2014-03-13 | 2015-01-21 | 严媚 | The imaging of intelligent optical miniflow and detection system |
| CN103871041A (en) * | 2014-03-21 | 2014-06-18 | 上海交通大学 | Image super-resolution reconstruction method based on cognitive regularization parameters |
Non-Patent Citations (2)
| Title |
|---|
| 刘娜等: ""基于多层卷积神经网络学习的单帧图像超分辨率重建方法"", 《基于多层卷积神经网络学习的单帧图像超分辨率重建方法》 * |
| 田青等: ""单帧图像的超分辨率重建技术"", 《中国体视觉与图像分析》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107623827A (en) * | 2017-08-15 | 2018-01-23 | 上海集成电路研发中心有限公司 | A kind of intelligent CMOS image sensor chip and its manufacture method |
| CN107623827B (en) * | 2017-08-15 | 2020-06-09 | 上海集成电路研发中心有限公司 | Intelligent CMOS image sensor chip and manufacturing method thereof |
| WO2019071918A1 (en) * | 2017-10-11 | 2019-04-18 | 南京大学 | Platelet projection imaging-based detection device and method therefor |
| WO2019084517A1 (en) * | 2017-10-26 | 2019-05-02 | Essenlix Corporation | System and methods of image-based assay using crof and machine learning |
| CN111896456A (en) * | 2020-06-19 | 2020-11-06 | 北京凡谱科技有限公司 | Single cell analysis method based on micro-fluidic and hyperspectral imaging |
| CN114505105A (en) * | 2022-01-13 | 2022-05-17 | 电子科技大学 | Micro-fluidic chip based on memory calculation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105260730A (en) | Machine learning-based contact-type imaging microfluid cell counter and image processing method thereof | |
| Jia et al. | Automatic polyp recognition in colonoscopy images using deep learning and two-stage pyramidal feature prediction | |
| Ahmed et al. | Survey and performance analysis of deep learning based object detection in challenging environments | |
| Huang et al. | A contact-imaging based microfluidic cytometer with machine-learning for single-frame super-resolution processing | |
| Gu et al. | Simultaneous vision-based shape and motion analysis of cells fast-flowing in a microchannel | |
| JP5146462B2 (en) | Blood fluidity measurement system | |
| CN101796391B (en) | Blood examination apparatus | |
| CN103984979A (en) | Lens-diffraction-imaging-free automatic algae detection and counting device and method | |
| Li et al. | One-stage disease detection method for maize leaf based on multi-scale feature fusion | |
| CN110728666B (en) | Typing method and system for chronic nasosinusitis based on digital pathological slide | |
| CN110598658A (en) | Convolutional network identification method for sow lactation behaviors | |
| Zhu et al. | Single object tracking in satellite videos: Deep Siamese network incorporating an interframe difference centroid inertia motion model | |
| CN109967143A (en) | A kind of cell size detection method based on micro-fluidic microscopic system | |
| Ma et al. | NPU RGB+ D dataset and a feature-enhanced LSTM-DGCN method for action recognition of basketball players | |
| Xie et al. | Tea-YOLOv8s: A tea bud detection model based on deep learning and computer vision | |
| Qiao et al. | A counting method of red jujube based on improved YOLOv5s | |
| Cong et al. | Citrus tree crown segmentation of orchard spraying robot based on RGB-D image and improved mask R-CNN | |
| US11592657B2 (en) | Method and system for identifying objects in a blood sample | |
| Sha et al. | Development of an accurate and automated quality inspection system for solder joints on aviation plugs using fine-tuned YOLOv5 models | |
| Li et al. | YOLOv5-KCB: A new method for individual pig detection using optimized K-means, CA attention mechanism and a bi-directional feature pyramid network | |
| Xu et al. | Robust 3D hand detection from a single RGB-D image in unconstrained environments | |
| Lee et al. | A hardware accelerated system for high throughput cellular image analysis | |
| KR20180105439A (en) | Flow cytometry apparatus and method based on image | |
| Yang et al. | An Improved Mask R-CNN Micro-Crack Detection Model for the Surface of Metal Structural Parts | |
| Huang et al. | Low illumination soybean plant reconstruction and trait perception |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160120 |