CN106204642B - A kind of cell tracker method based on deep neural network - Google Patents
A kind of cell tracker method based on deep neural network Download PDFInfo
- Publication number
- CN106204642B CN106204642B CN201610511744.9A CN201610511744A CN106204642B CN 106204642 B CN106204642 B CN 106204642B CN 201610511744 A CN201610511744 A CN 201610511744A CN 106204642 B CN106204642 B CN 106204642B
- Authority
- CN
- China
- Prior art keywords
- cell
- picture
- frame picture
- queue
- similarity
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
The cell tracker method based on deep neural network that the invention discloses a kind of, is related to technical field of image processing, solves the problems, such as cell tracker in continuous microscopic cell sequence of pictures.The present invention includes collecting cell image data, determines cell data, extracts cell characteristic using convolutional neural networks, obtains the multitask observation model of an initialization；The first picture and cell position of given microscope photograph sequence, obtain the candidate cell position in the first picture near cell position in next frame picture；Positive sample and negative sample, the multitask observation model of training initialization are up-sampled in the first picture, and candidate cell position is predicted, obtains the prediction similarity probability value of cell；It predicts cell position of the maximum candidate cell position of similarity probability value as next frame picture, maximum predicted similarity probability value and corresponding cell position is saved in a model modification strategy, carry out threshold value and be compared.The present invention is used for the tracking of cell.
Description
Technical field
A kind of cell tracker method based on deep neural network is related to biomedical, machine view for the tracking of cell
The technical fields such as feel, image procossing, artificial intelligence.
Background technique
Vision tracking is an important subject of computer vision field.The purpose of visual pursuit be tracking it is single or
The track of multiple objects has been widely used for many practical visual tasks, such as video monitoring, automated driving system and biology
Cell lineage analysis.Application program is tracked as a typical vision, the target of cell tracker is in order to directly from microgram
As tracking cell in sequence.By the result of tracking cell, we can study cell behavior, further construct cell lineage and
Analyze cytomorphology.Therefore, cell is vital with automatic tracking method.
The challenge of tracking cell can be summarized as four classes.First challenge is the problem on deformation of cell.The deformation packet of cell
Include elongation, expansion and contraction.Traditional visual tracking method is easier the target without significant change in shape of processing rigid body.
However the nonrigid shape of cell makes tracing task have more challenge, and because over time, their chop and change shapes
Shape.The challenge of second class is the behavior about cell.Such as cell migration has complicated motor pattern.Complicated cell fortune
Dynamic model formula increases difficulty to tracing task.Third challenges the living environment from cell.Many of cell liquid particle,
In include dead cell, bacterium and other organic substances.Tracking cell method has to differentiate between cell and other particles.The last one
Challenge is that the cell image resolution ratio under microscope is low, and the cell relative volume in image is smaller.
With the development of deep learning, characteristic learning method has been applied successfully to computer vision field.At us
Work in, convolutional neural networks be used to study robustness cell characteristic so that improve nonrigid cell tracking know
Not rate.Convolutional neural networks are widely used in visual task: in traditional image classification problem, convolutional neural networks are for big
It has been proved to improve accuracy of identification on scale data collection, in many machine learning methods, it is current state-of-the-art feature
One of learning model method.
In actual visual problem, deep learning method encounters overfitting problem sometimes.Since cell object is non-just
The characteristics of property, tracking cell method requires good Generalization Capability.Multi-task learning is usually by big PROBLEM DECOMPOSITION at independence
Task, it is a kind of method of machine learning, association from all independent tasks study to shared cell characteristic.It is this
Learning process provides a kind of preferable generalization.It has been applied successfully to many fields, including machine translation, and voice is known
Not, target identification etc..
Summary of the invention
The present invention provides a kind of cell tracker method based on deep neural network in view of the above shortcomings, solves existing
There are in technology cytomorphosis, the behavior of cell, the living environment of cell and microscope in continuous microscopic cell sequence of pictures
Under the low problem for causing cell tracker ineffective of cell image resolution ratio.
To achieve the goals above, the technical solution adopted by the present invention are as follows:
A kind of cell tracker method based on deep neural network, which comprises the steps of:
(1) cell image data is collected, cell data are cut out, cell characteristic is extracted using convolutional neural networks, obtains
The multitask observation model of one initialization；
(2) the first frame picture of microscope photograph sequence, and the cell position in given first frame picture are given, grain is utilized
Sub- filtering method obtains multiple candidate cell positions in present frame picture near cell position in next frame picture, wherein under
One frame picture refers to next frame picture in sequence；
(3) positive sample and negative sample are up-sampled in first frame picture, the training one on the multitask observation model of initialization
A binary classifier predicts candidate cell position trained multitask observation model, obtains the prediction phase of cell
Like degree probability value；
(4) cell position of the prediction maximum candidate cell position of similarity probability value as next frame picture is chosen, it will
Maximum predicted similarity probability value and corresponding cell position are saved in a model modification strategy, a threshold value are arranged, such as
The updated maximum predicted similarity probability value of fruit is less than threshold value, goes to the tracking that step (2) enter next frame picture；Otherwise
Go to the training of step (3) resampling.
Further, include the following steps: in the step (1)
(11) cell characteristic extract task: by construction one three classification task come define cell characteristic extract task, three
A classification includes cell-free classification, cell class in division, ordinary cells classification, wherein cell-free classification includes dead cell
And other organic matters；
(12) data acquisition:, will be right from the situation comprising three category features in step (11) is obtained in original microscope photograph
The Pictures location answered intercepts out, is then cut out to the picture slice unification of all interceptions to 32 × 32 size
Cell data；
(13) convolutional neural networks training: obtained cell data are input to convolutional neural networks learning characteristic and extracted and are appointed
Business, obtains the multitask observation model of initialization.Further, the step (2) includes the following steps:
(21) the first frame picture of microscope photograph sequence, and the cell position in given first frame picture are given, will be given
Fixed cell position is input in a particle filter model；
(22) particle number that setting particle filter model generates, candidate particle position is obtained by particle filter algorithm
It sets, that is, obtains multiple candidate cell positions.
Further, the step (3) specifically comprises the following steps:
(31) sample positive negative sample: 1 to 2 pixels of cell position intercept N number of positive sample on first frame picture, 50
M negative sample of position acquisition other than pixel；
(32) on the basis of step (1) obtains the multitask observation model of initialization, current classification task is relearned,
Obtain trained multitask observation model；
(33) the candidate particle position that trained multitask observation model is used to generate particle filter is predicted,
Provide the similarity probability value of each candidate cell position.
Further, the model modification strategy in the step (4) specifically comprises the following steps:
(41) first to the queue initialization in model: N number of from 1 to 2 pixel interceptions of cell position on first frame picture
Positive sample carries out model initialization, and the similarity probability of N number of positive sample of initialization is 1, is then decayed by similarity regular
Formula decays to similarity, guarantee queue head cell similarity decay it is slow, the attenuation cell of rear of queue it is fast；
In above formula, βkIt is the similarity of cell, wherein (1, Z) k ∈, Z are the numbers of cell,It is a constant, for controlling
Make the speed of decaying；(42) having new cell picture to enter queue every time will decay more to the similarity of cell in queue
Newly, then to be judged by one: βnew>βminIt, should if new picture is also smaller than the smallest similarity in queue
New cell picture discarding is added without, and otherwise new cell picture is added to queue head and is updated；
(43) if there is new picture is added in queue, the smallest cell sample of similarity is searched in the queue from queue
In delete, do not change the sequence of other samples of queue.
Further, the basic structure of convolutional neural networks can be indicated with following formula in the step (13):
Wherein, P and Q defines the size of convolution kernel, and p, q indicate line number and columns, and for defining pixel position, f is
Convolution kernel function, l ∈ (1, L) indicate the current convolutional neural networks number of plies,The feature of l layers of i row j column is defined, k is fixed
The justice parameter of convolution kernel, b are corresponding bias functions.
The pondization operation of convolutional neural networks can be indicated with following formula:
a(l)=f (β(l)·down(a(l-1))+b(l-1),
In above-mentioned formula, down illustrates the down-sampling operation of convolutional neural networks, and β is relevant parameter.
Compared with the prior art, the advantages of the present invention are as follows:
One, the feature extraction of cell is that a crucial pretreatment operation can be very by screening to cell picture
The feature extraction effect of good increase cell, in concrete operations, the cell picture of arrangement by normalizing to 32 × 32 again
Size, convolutional neural networks are designed to two layers of convolution, the study that two layers of down-sampling layer can be best to cell characteristic；
Two, during tracking, cell can be moved, deformation, and constantly weight is needed in traditional cell tracker method
New training, and we maintain a sample queue in the present invention, the Sample preservation in queue its similarity probability value, every time
After new sample enters queue, the smallest sample of similarity will be rejected from queue, and the positive sample of this method sampling is in cell
It is more preferable during mobile, cells deformation；
Three, by the method for deep learning to cell carry out feature extraction, can be good to multitask observation model into
Row initialization well, increases the robustness of model, improves the effect of tracking；
Four, in a continuous microcytoscope sequence of pictures, pass through the position of cell in calibration first frame picture, energy
The position of all picture cells in enough accurately tracking sequences.
Detailed description of the invention
Fig. 1 is multitask observation model figure in the present invention；
Fig. 2 is cell recognition flow chart in the present invention；
Fig. 3 is the exemplary diagram of model modification sample maintenance in the present invention；
Fig. 4 is cell tracker effect picture in the present invention.
Specific embodiment
The present invention is further illustrated with reference to the accompanying drawings and examples.
Referring to Fig. 1, a kind of kernel model of the cell tracker method based on deep neural network is the sight of a multitask
Survey model.The observation model of multitask is divided into two learning tasks: the main task of one two classification, it is therefore an objective to by what is currently tracked
Cell target is distinguished with other objects in picture background, which can constantly update the part during cell tracker
Classifier；Another is the subtask of three classification, it is therefore an objective to by the primary categories of cell: acellular classification, ordinary cells,
Cell differentiation comes in division, can be when initializing multitask observation model just by distinguishing main cell class
Learn the main feature to cell, to improve tracking efficiency.
Such as the overall flow that Fig. 2 is cell tracker, specifically comprise the following steps:
(1) cell image data is collected, cell data are cut out, cell characteristic is extracted using convolutional neural networks, obtains
The multitask observation model of one initialization；Include the following steps:
(11) cell characteristic extract task: by construction one three classification task come define cell characteristic extract task, three
A classification includes cell-free classification, cell class in division, ordinary cells classification, wherein cell-free classification includes dead cell
And other organic matters；
(12) data acquisition:, will be right from the situation comprising three category features in step (11) is obtained in original microscope photograph
The Pictures location answered intercepts out, is then cut out to the picture slice unification of all interceptions to 32 × 32 size
Cell data；
(13) convolutional neural networks training: obtained cell data are input to convolutional neural networks learning characteristic and extracted and are appointed
Business, obtains the multitask observation model of initialization.The basic structure of convolutional neural networks can be indicated with following formula:
Wherein, P and Q defines the size of convolution kernel, and p, q indicate line number and columns, and for defining pixel position, f is
Convolution kernel function, l ∈ (1, L) indicate the current convolutional neural networks number of plies,The feature of l layers of i row j column is defined, k is fixed
The justice parameter of convolution kernel, b are corresponding bias functions.
The pondization operation of convolutional neural networks can be indicated with following formula:
a(l)=f (β(l)·down(a(l-1))+b(l-1)),
In above-mentioned formula, down illustrates the down-sampling operation of neural network, and β is relevant parameter.
(2) the first frame picture of microscope photograph sequence, and the accurate location in given first frame picture are given, in figure
The cell of square-shaped frame calibration；First frame picture is obtained in next frame picture (i.e. the second frame picture) using particle filter method
Multiple candidate cell positions near middle cell position；Include the following steps:
(21) the first frame picture of microscope photograph sequence, and the cell position in given first frame picture are given, will be given
Fixed cell position is input in a particle filter model；
(22) particle number that setting particle filter model generates, candidate particle position is obtained by particle filter algorithm
It sets, that is, obtains multiple candidate cell positions.
(3) positive sample and negative sample are up-sampled in first frame picture, the training one on the multitask observation model of initialization
A binary classifier predicts candidate cell position trained multitask observation model, obtains the prediction phase of cell
Like degree probability value；Multitask observation model is made of two parts.The subtask of three classification is used to initialization model and reaches extraction
The purpose of cell characteristic.The initialization of multitask observation model is under 30000 sample data sets, the convolution of one 4 layers of training
Neural network, the calibration of data include cell-free sample, ordinary cells sample, cell sample in division.Trained multitask
Observation model is used to learn a binary classifier to the cell in current image (first frame picture).The binary classifier is just
Sample, that is, the sample for belonging to cell are chosen from positive sample queue.The initialization of queue is from cell on first frame picture
The pixel of position 1 to 2 intercepts 10 positive samples.The selection of negative sample is that position acquisition 100 other than 50 pixels are negative
Sample.Trained observation model predicts candidate cell position, provides the similarity probability value of cell；Including as follows
Step:
(31) sample positive negative sample: 1 to 2 pixels of cell position intercept 10 positive samples on first frame picture,
100 negative samples of position acquisition other than 50 pixels；
(32) on the basis of step (1) obtains the multitask observation model of initialization, current classification task is relearned,
Obtain trained multitask observation model；
(33) the candidate particle position that trained multitask observation model is used to generate particle filter is predicted,
Provide the similarity probability value of each candidate cell position.
(4) the prediction maximum candidate cell position of similarity probability value is chosen as next picture (i.e. next frame picture)
Cell position, maximum predicted similarity probability value and corresponding cell position are saved in a model modification strategy,
One threshold value (threshold value chosen in our test is 0.8) is set, if updated maximum predicted similarity probability value
Less than threshold value, goes to step (2) and enter next picture (i.e. next frame picture, as next frame picture is the picture tracked
Next picture) tracking；Conversely, going to the training of step (3) resampling.Fig. 3 is showing in detail for model modification sample maintenance
Example diagram gives the process of ten positive samples of an Example maintenance.This method from initialize the queue, similarity decaying, queue
Insertion and delete operation process, model modification strategy specifically comprises the following steps:
(41) first to the queue initialization in model: N number of from 1 to 2 pixel interceptions of cell position on first frame picture
Positive sample carries out model initialization, and the similarity probability of N number of positive sample of initialization is 1, is then decayed by similarity regular
Formula decays to similarity, guarantee queue head cell similarity decay it is slow, the attenuation cell of rear of queue it is fast；
In above formula, βkIt is the similarity of cell, wherein (1, Z) k ∈, Z are the number of cell, in this example, Z=
10,It is a constant, for controlling the speed of decaying, in this example, we are used(42) have every time new
Picture enter queue will the similarity to cell in queue carry out decaying update, then will pass through one judgement: βnew>
βminIf new picture is also smaller than the smallest similarity in queue, which is added without, otherwise newly
Picture be added to queue head and be updated, new picture refers to the corresponding cell position of maximum predicted similarity probability value；
(43) if there is new picture is added in queue, the smallest cell sample of similarity is searched in the queue from queue
In delete, do not change the sequence of other samples of queue.
Fig. 4 is the effect picture of experiment, and this method is tested in the case where three types.Scheming (a) is compared in variation
The process of cell is tracked under complicated background；Figure (b) is the process that cell is tracked in the apparent situation of mobile comparison of cell；
Figure (c) is the process that cell is tracked in the case where the deformation of cell is relatively more violent.
Claims (4)
1. a kind of cell tracker method based on deep neural network, which comprises the steps of:
(1) cell image data is collected, cell data are cut out, cell characteristic is extracted using convolutional neural networks, obtains one
The multitask observation model of initialization；Include the following steps: in the step (1)
(11) cell characteristic extracts task: defining cell characteristic by one three classification task of construction and extracts task, three classes
Not Bao Kuo cell-free classification, cell class in division, ordinary cells classification, wherein cell-free classification include dead cell and
Other organic matters；
(12) data acquisition: from the situation comprising three Class Types in step (11) is obtained in cell image data, by corresponding figure
Piece position intercepts out, then to the size of the picture slice unification of all interceptions to 32 × 32, the cell number cut out
According to；
(13) convolutional neural networks training: being input to convolutional neural networks learning characteristic for obtained cell data and extract task,
Obtain the multitask observation model of initialization；
(2) the first frame picture of microscope photograph sequence, and the cell position in given first frame picture are given, is filtered using particle
Wave method obtains multiple candidate cell positions in present frame picture near cell position in next frame picture, wherein next frame
Picture refers to next frame picture in sequence；
(3) positive sample and negative sample are up-sampled in first frame picture, the just training one on the multitask observation model of initialization
Candidate cell position of the trained multitask observation model to subsequent pictures is predicted, obtains cell by binary classifier
Prediction similarity probability value；
(4) cell position of the prediction maximum candidate cell position of similarity probability value as next frame picture is chosen, it will be maximum
Prediction similarity probability value and corresponding cell position are saved in a model modification strategy, a threshold value are arranged, if more
Maximum predicted similarity probability value after new is less than threshold value, goes to the tracking that step (2) carry out i+1 frame picture, wherein i is indicated
Present frame picture；Conversely, going to the training of step (3) resampling；
Model modification strategy in the step (4) specifically comprises the following steps:
(41) first to the queue initialization in model: intercepting N number of positive sample from 1 to 2 pixels of cell position on first frame picture
This progress model initialization, the similarity probability of N number of positive sample of initialization are 1, are then decayed rule formula by similarity
Decay to similarity, guarantee queue head cell similarity decay it is slow, the attenuation cell of rear of queue it is fast；
In above formula, βkIt is the similarity of cell, wherein (1, Z) k ∈, Z are the numbers of cell,It is a constant, declines for controlling
The speed subtracted；
(42) have every time new cell picture enter queue will the similarity to cell in queue carry out decaying update, then want
Judged by one: βnew> βminIf new picture is also smaller than the smallest similarity in queue, the new cell
Picture discarding is added without, and otherwise new cell picture is added to queue head and is updated；
(43) if there is new picture is added in queue, the smallest cell sample of similarity is searched in the queue and is deleted in queue
It removes, does not change the sequence of other samples of queue.
2. a kind of cell tracker method based on deep neural network according to claim 1, which is characterized in that the step
Suddenly (2) include the following steps:
(21) the first frame picture of microscope photograph sequence, and the cell position in given first frame picture are given, by what is given
Cell position is input in a particle filter model；
(22) particle number that setting particle filter model generates, candidate particle position is obtained by particle filter algorithm, i.e.,
Obtain multiple candidate cell positions.
3. a kind of cell tracker method based on deep neural network according to claim 1, which is characterized in that the step
Suddenly (3) specifically comprise the following steps:
(31) sample positive negative sample: 1 to 2 pixels of cell position intercept N number of positive sample on first frame picture, in 50 pixels
M negative sample of position acquisition other than point；
(32) on the basis of step (1) obtains the multitask observation model of initialization, current classification task is relearned, is obtained
Trained multitask observation model；
(33) the candidate particle position that trained multitask observation model is used to generate particle filter is predicted, is provided
The similarity probability value of each candidate cell position.
4. a kind of cell tracker method based on deep neural network according to claim 2, which is characterized in that the step
Suddenly the basic structure of convolutional neural networks can be indicated with following formula in (13):
Wherein, P and Q defines the size of convolution kernel, and p, q indicate line number and columns, and for defining pixel position, f is convolution
Kernel function, I ∈ (1, L) indicate the current convolutional neural networks number of plies,The feature of I layers of i row j column is defined, k is defined
The parameter of convolution kernel, b are corresponding bias functions；
The pondization operation of convolutional neural networks is indicated with following formula:
a(l)=f (β(l)·down(a(l-1))+b(l-1)),
In above-mentioned formula, down illustrates the down-sampling operation of convolutional neural networks, and β is relevant parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610511744.9A CN106204642B (en) | 2016-06-29 | 2016-06-29 | A kind of cell tracker method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610511744.9A CN106204642B (en) | 2016-06-29 | 2016-06-29 | A kind of cell tracker method based on deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106204642A CN106204642A (en) | 2016-12-07 |
| CN106204642B true CN106204642B (en) | 2019-07-09 |
Family
ID=57464014
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610511744.9A Expired - Fee Related CN106204642B (en) | 2016-06-29 | 2016-06-29 | A kind of cell tracker method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106204642B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106846362B (en) * | 2016-12-26 | 2020-07-24 | 歌尔科技有限公司 | Target detection tracking method and device |
| CN108256408A (en) * | 2017-10-25 | 2018-07-06 | 四川大学 | A kind of stem cell method for tracing based on deep learning |
| CN108038435B (en) * | 2017-12-04 | 2022-01-04 | 中山大学 | Feature extraction and target tracking method based on convolutional neural network |
| CN110110799B (en) * | 2019-05-13 | 2021-11-16 | 广州锟元方青医疗科技有限公司 | Cell sorting method, cell sorting device, computer equipment and storage medium |
| CN110136149A (en) * | 2019-05-21 | 2019-08-16 | 闽江学院 | Leucocyte positioning and dividing method based on deep neural network |
| CN111753835B (en) * | 2019-08-19 | 2021-08-31 | 湖南大学 | Cell tracking method based on local graph matching and convolutional neural network |
| CN111047577B (en) * | 2019-12-12 | 2021-02-26 | 太原理工大学 | Abnormal urine red blood cell classification statistical method and system |
| CN111276181B (en) * | 2020-01-20 | 2023-06-06 | 中国科学院自动化研究所 | Noninvasive in-vivo stem cell tracing method and system based on convolutional neural network |
| CN111175301B (en) * | 2020-03-17 | 2023-03-31 | 桂林优利特医疗电子有限公司 | Clear image acquisition method for sheath flow microscopic full-thickness flow channel cells |
| CN112101575B (en) * | 2020-11-04 | 2021-04-30 | 江苏集萃微纳自动化系统与装备技术研究所有限公司 | Three-dimensional positioning method of micromanipulation platform for cell injection |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559724A (en) * | 2013-10-31 | 2014-02-05 | 苏州相城常理工技术转移中心有限公司 | Method for synchronously tracking multiple cells in high-adhesion cell environment |
| CN103955946A (en) * | 2013-08-16 | 2014-07-30 | 常熟理工学院 | Ant colony neighbor cell tracking system based on cooperation and competition mode and application thereof |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9173909B2 (en) * | 2013-05-23 | 2015-11-03 | Drvision Technologies Llc | Image guided protocol for cell generation |
-
2016
- 2016-06-29 CN CN201610511744.9A patent/CN106204642B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955946A (en) * | 2013-08-16 | 2014-07-30 | 常熟理工学院 | Ant colony neighbor cell tracking system based on cooperation and competition mode and application thereof |
| CN103559724A (en) * | 2013-10-31 | 2014-02-05 | 苏州相城常理工技术转移中心有限公司 | Method for synchronously tracking multiple cells in high-adhesion cell environment |
Non-Patent Citations (1)
| Title |
|---|
| Cell Tracking using Convolutional Neural Networks;Anton Jackson-Smith;《http://cs231n.stranford.edu/reports/2016/pdfs》;20160323;第4321-4327页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106204642A (en) | 2016-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106204642B (en) | A kind of cell tracker method based on deep neural network | |
| US20230127698A1 (en) | Automated stereology for determining tissue characteristics | |
| US20170109879A1 (en) | Computer-implemented methods, computer-readable media, and systems for tracking a plurality of spermatozoa | |
| CA2948499A1 (en) | System and method for classifying and segmenting microscopy images with deep multiple instance learning | |
| Saleem et al. | A performance-optimized deep learning-based plant disease detection approach for horticultural crops of New Zealand | |
| CN112633257A (en) | Potato disease identification method based on improved convolutional neural network | |
| CN112784767A (en) | Cell example segmentation algorithm based on leukocyte microscopic image | |
| Baykal et al. | Modern convolutional object detectors for nuclei detection on pleural effusion cytology images | |
| Ge et al. | Coarse-to-fine foraminifera image segmentation through 3D and deep features | |
| Patel | Bacterial colony classification using atrous convolution with transfer learning | |
| Bergler et al. | FIN-PRINT a fully-automated multi-stage deep-learning-based framework for the individual recognition of killer whales | |
| An et al. | Automated detection of tuberculosis bacilli using deep neural networks with sputum smear images | |
| Wang et al. | Multicell migration tracking within angiogenic networks by deep learning-based segmentation and augmented Bayesian filtering | |
| CN113128335A (en) | Method, system and application for detecting, classifying and discovering micro-body paleontological fossil image | |
| Angeline et al. | Faster region-based convolutional neural network for plant-parasitic and non-parasitic nematode detection | |
| Gupta et al. | Simsearch: A human-in-the-loop learning framework for fast detection of regions of interest in microscopy images | |
| Kalyani et al. | Remembrance of Monocotyledons Using Residual Networks | |
| Elvana et al. | Lymphatic filariasis detection using image analysis | |
| Michal et al. | Machine-learning approach to microbial colony localisation | |
| Soleimany et al. | Image segmentation of liver stage malaria infection with spatial uncertainty sampling | |
| Sai | A review on crop disease detection using deep learning | |
| US20230291992A1 (en) | Method and system for event-based imaging | |
| CN116012838B (en) | Artificial intelligence-based organoid activity recognition method and system | |
| Li et al. | A Deep Learning based Method for Microscopic Object Localization and Classification | |
| Vrbančič et al. | Profiling the Arion rufus snails with computer vision |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190709 Termination date: 20210629 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |