CN107784319A - A kind of pathological image sorting technique based on enhancing convolutional neural networks - Google Patents
A kind of pathological image sorting technique based on enhancing convolutional neural networks Download PDFInfo
- Publication number
- CN107784319A CN107784319A CN201710883838.3A CN201710883838A CN107784319A CN 107784319 A CN107784319 A CN 107784319A CN 201710883838 A CN201710883838 A CN 201710883838A CN 107784319 A CN107784319 A CN 107784319A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- weak classifier
- image
- classification
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
A kind of pathological image sorting technique based on enhancing convolutional neural networks, the pathological image sorting technique comprise the following steps:Fixed size image block is divided into healthy two class pathological pictures by ill, assigns two class image block corresponding labels as training sample;Multiple depth convolutional neural networks are trained as Weak Classifier;The multiple Weak Classifiers trained are integrated into strong classifier, realize the classification to pathological image.This method trains multiple depth convolutional neural networks as Weak Classifier, with sample re -training another grader with classification error, finally multiple Weak Classifiers are integrated, it is possible to achieve classification is complementary, the classification performance of network is effectively improved, is improved to pathological picture classification accuracy.
Description
Technical field
The present invention relates to Signal and Information Processing technical field, more particularly to a kind of disease based on enhancing convolutional neural networks
Manage image classification method.
Background technology
In Pathologic image analysis, the problem of characteristics of image is individual highly difficult is extracted, doctor will analyze a large amount of pathology daily
Figure comes whether diagnosis cell occurs lesion, and so huge workload needs to take a significant amount of time, and needs what doctor enriched
Experience ensures the accuracy rate of diagnosis, so a kind of design extraction pathology figure feature for being capable of precise and high efficiency, so as to by it is ill with
The algorithm that the pathology figure of health is correctly classified is significant.
In extraction feature and classification:Monga etc. learns the dictionary of particular category under sparse constraint in document [1], real
Now pathological image is classified;The deep neural network automatic detection cell compartment such as Gadepalli in document [2];In document [3]
Hatipoglu etc. is realized to image classification with convolutional neural networks study nuclear characteristics;The use such as Khosla in document [4]
Method is depth e-learning data characteristics, and data are classified, and the sample of misclassification then is added into network training sample next time
In this, training network again.
Inventor is during the present invention is realized, discovery at least has the following disadvantages in the prior art and deficiency:
Feature extraction of the conventional art to pathological image is highly difficult, it is difficult to correctly distinguish image category;Conventional depth is rolled up
Product neutral net can automatically extract characteristics of image, but only train a depth e-learning sample characteristics, realize to image
Classification, classification results can be caused to depend on the performance of the depth network, classification accuracy is relatively low.
The content of the invention
The invention provides a kind of pathological image sorting technique based on enhancing convolutional neural networks, the present invention is using training
Multiple depth convolutional neural networks integrate multiple Weak Classifiers and obtain the method for strong classifier, realization pair as Weak Classifier
The classification of pathological image;Multiple Weak Classifiers can realize that classification is complementary, effectively improve classification performance, can be examined accordingly used in medical science
The fields such as disconnected and pathological analysis, it is described below:
It is a kind of based on enhancing convolutional neural networks pathological image sorting technique, the pathological image sorting technique include with
Lower step:
Fixed size image block is divided into healthy two class pathological pictures by ill, two class image block corresponding labels is assigned and makees
For training sample;
Multiple depth convolutional neural networks are trained as Weak Classifier;The multiple Weak Classifiers trained are integrated into strong point
Class device, realize the classification to pathological image.
Wherein, two class image block corresponding labels of the imparting are specially as training sample:
Still image block size is N × N, overlaid pixel s, from top to bottom, from left to right slides and takes block, and it is suitable to press block
Sequence is saved as training sample;
Training sample is normalized, three passages of red, green, blue of image block are normalized respectively, make image block
Each passage average value is 0, variance 1;
It is 1 to assign label for normalized ill image block, and it is -1 that healthy image block, which assigns label,.
Wherein, the multiple depth convolutional neural networks of training are specially as Weak Classifier:
Multiple convolutional neural networks are built as Weak Classifier, initialization sample weight;Weak point is obtained according to weight sampling
The training sample of class device, train a Weak Classifier;
After the completion of training, all training samples are input in the Weak Classifier;The Weak Classifier is calculated to training sample
Classification error rate and the Weak Classifier shared weight in strong classifier;
Update the weight of training sample;Until the training of all Weak Classifiers is completed.
Wherein, the convolutional neural networks are specially:
Including an input layer, multiple convolutional layers, multiple pond layers and multiple full articulamentums;
Each convolutional layer is followed by a pond layer, and last pond layer is followed by multiple full articulamentums, the output of preceding layer
As the input of later layer, network exports the classification predicted value of training sample.
Further, it is described the multiple Weak Classifiers trained are integrated into strong classifier to be specially:
All Weak Classifier output labels trained are weighted summation, form strong classifier.
The present invention is used as Weak Classifier by multiple convolutional neural networks, and is integrated into strong classifier, has with following
Beneficial effect:
1st, the convolution operation of convolutional neural networks can effectively extract the exclusive form of different classes of image, color and wheel
Wide feature, ensure the accuracy of feature;
2nd, multiple convolutional neural networks (CNN) are trained to be used as Weak Classifier, every time by previous Weak Classifier classification error
Sample be added in the training sample of this Weak Classifier, it is complementary to realize the classification results of each Weak Classifier, ensure that classification
Accuracy rate;
3rd, present procedure is simple, it is easy to accomplish.
Brief description of the drawings
Fig. 1 is a kind of flow chart of the pathological image sorting technique based on enhancing convolutional neural networks;
Fig. 2 is convolutional neural networks structural representation.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, embodiment of the present invention is made below further
It is described in detail on ground.
Embodiment 1
In order to realize the feature extraction of high-accuracy and classification, the embodiment of the present invention proposes a kind of based on enhancing convolution god
Pathological image sorting technique through network, it is described below referring to Fig. 1:
101:Fixed size image block is divided into healthy two class pathological pictures by ill, two class image blocks of imparting are corresponding to be marked
Label are used as training sample;
Wherein, the step is specially:
1) still image block size is N × N, and overlaid pixel s, pathological image size is m × n, according to formula, OK
Block number is pr=(m-s)/(N-s), and row block number is pc=(n-s)/(N-s);
The image block of all training images is arranged in column form, image block total amount is K;In practical application, N, s, m, n
Setting, the embodiment of the present invention are without limitation as needed;
2) image block caused by step 1) is normalized, is divided into three passages of red, green, blue, each passage
It is 0 that pixel, which is normalized into average, and variance is 1 sample set X;
Above-mentioned processing procedure is known to those skilled in the art, and the embodiment of the present invention is not repeated this.
3) label is assigned for the sample set X after normalization;
Wherein, it is 1 to assign label for ill image block, and it is -1 that healthy image block, which assigns label,.
102:Multiple depth convolutional neural networks are trained as Weak Classifier;
Wherein, the step is specially:
Training sample weight is initialized, kth (k=1,2 ..., K) individual training sample weight is Dk=1/K.
By taking the individual Weak Classifier training of jth (j=1,2 ..., H) as an example, H is total Weak Classifier number, illustrates training process:
1) build a convolutional neural networks includes an input layer, Duo Gejuan as Weak Classifier (referring to Fig. 2), network
Lamination, multiple pond layers and multiple full articulamentums;Each convolutional layer is followed by a pond layer, and last pond layer is followed by multiple
Full articulamentum;In a network, each layer of output is all its next layer input, and network output is predicted for the classification of training sample
Value.
The activation primitive of convolutional layer and full articulamentum selects as needed, and network output uses softmax functions, output two
Individual value, the network of samples target output value that label is 1 are (1,0), and the network of samples target output value that label is -1 is (0,1).
2) sample set X weight is D={ D1,D2,...,DK, sampled according to weight, obtain the training sample of Weak Classifier
This;
3) network training takes batch gradient descent method, is trained by minimizing convolutional neural networks cost function formula (1)
Weak Classifier:
In formula, J be cost function value, ykFor the network objectives output valve of k-th of training sample, pkFor k-th of training sample
This network real output value;K is training sample number;Equation latter is weight regularization, βlFor coefficient, Wfc (l)For l
Individual full articulamentum weight, L is full articulamentum number, | | | |2For 2- norms.
4) after the completion of training, sample set X is input in the Weak Classifier trained, obtains network output valve;The output valve
There are two, the big prediction label of first value is 1, is otherwise -1;Classification error rate is calculated according to formula (2):
In formula, ejFor the classification error rate of j-th of Weak Classifier;Cj(Xk) for j-th Weak Classifier to kth (k=1,
2 ..., K) individual training sample prediction label;YkFor k-th of training sample physical tags.
According to classification error rate ejCalculate weight shared by the Weak Classifier:
In formula, αjFor weight shared by j-th of Weak Classifier.
5) training sample set X weight is recalculated by formula (4), exemplified by k-th:
In formula, DkFor k-th of training sample weight, Z is standardizing factor:
6) above step is repeated, until the training completion of H Weak Classifier, H number is not done be not limited here.
103:The multiple Weak Classifiers trained are integrated into strong classifier, realized to image classification.
Wherein, concretely comprise the following steps:
The Weak Classifier trained is integrated by formula (6), obtains strong classifier:
In formula, C () is strong classifier, and sign () is sign function.
In summary, the embodiment of the present invention, which uses, trains multiple depth convolutional neural networks as Weak Classifier, and integrates
Multiple Weak Classifiers obtain the method for strong classifier, realize the classification to pathological image;Multiple Weak Classifiers can realize that classification is mutual
Mend, effectively improve classification performance, the fields such as medical diagnosis and pathological analysis can be used in accordingly.
Embodiment 2
The scheme in embodiment 1 is described in detail with reference to specific accompanying drawing 1,2 and calculation formula, referred to
It is described below:
201:Pretreatment;
The image pattern of 100 kidneys of selection regards training sample, including 50 illness and 50 healthy images, to original
Image zooms in and out operation, and the image size after scaling is 512 × 680.
202:Fixed size image block is divided into healthy two class pathological pictures by ill, two class image blocks of imparting are corresponding to be marked
Label are used as training sample;
1) tile size is set as 80 × 80, and overlaid pixel 40, image size is 512 × 680, and side of being expert at can produce
Pr=(512-40)/(80-40)=11 image block, pc=(680-40)/(80-40)=16 image can be produced on row side
Block.Piece image can produce 176 image blocks altogether.
2) all take block to handle 100 pictures for doing scaling processing, obtain training sample set X={ X1,X2,...,
X17600, wherein, Xk∈R80×80(k=1,2 ..., 17600);
3) normalized is done to training sample set X, is divided into three passages of red, green, blue, the pixel normalization of each passage
It is 0 into average, variance 1;
4) label is assigned for normalized sample set X.It is 1 to assign label for ill image block, and healthy image block assigns mark
Sign as -1.
203:A depth convolutional neural networks are trained as Weak Classifier;
1) convolutional neural networks are built, including an input layer, two convolutional layers, two pond layers and two connect entirely
Layer is connect, wherein:
Input layer size be 80 × 80 × 3,3 be training sample Color Channel;First convolutional layer convolution kernel size be
40 × 40, convolution kernel number is 10;Second convolutional layer convolution kernel size is 20 × 20, and convolution kernel number is 20;Pond size
For 2 × 2, step-length 2, using maximum pond;Two full articulamentum weights are respectively Wfc (1)∈R8000×500, Wfc (2)∈R500×2;
The activation primitive of convolutional layer and first full articulamentum is sigmoid functions;Second full articulamentum uses
Softmax functions, network export two values, and the network of samples target output value that label is 1 is (1,0), and label is -1 sample
Network objectives output valve is (0,1);E-learning calibration is 10-4, iterations is 2 × 103。
2) training sample initial weight D is initializedk=1/17600, k=1,2 ..., 17600;Sampled, obtained according to weight
Take 5 × 103Individual training sample;
3) network training uses batch gradient descent method, by minimizing cost function (7) training convolutional neural networks:
After training, by all training samples by the network, network output valve is obtained;Network output valve has two,
The big prediction label of first value is 1, is otherwise -1;Classification error rate is calculated according to formula (8):
In formula, e1For the 1st Weak Classifier classification error rate, C1(Xk) for the 1st Weak Classifier to k-th of training sample
Prediction label;YkFor the physical tags of k-th of training sample.
According to e1Calculate weight shared by the Weak Classifier:
4) training sample weight is updated, exemplified by k-th:
In formula, DkFor the weight of k-th of training sample;Z is standardizing factor.
204:Train the Weak Classifier of multiple only two full articulamentums;
In order to reduce the training time, the forward part of the full articulamentum of the convolutional neural networks trained in 203 is fixed, is only instructed
Practice full articulamentum as Weak Classifier.
Training sample X is input in 203 Weak Classifiers trained, the output of second pond layer is taken out, obtains
The feature that 17600 sizes are 20 × 20 × 20, the training sample as following Weak Classifier;
By taking jth (j=2,3,4,5) individual Weak Classifier as an example, Weak Classifier number is 4, illustrates Weak Classifier training process:
1) 4 Weak Classifier structures are identical with the full connection Rotating fields of convolutional neural networks in 203;Parameter setting also phase
Together;
2) sampled according to weight, obtain 5 × 103Individual training sample;Training sample is inputted in j-th of Weak Classifier, adopted
Random batch gradient descent method is taken, j-th of Weak Classifier is trained by minimizing cost function (1):
3) training is completed, and training sample is fully entered in the Weak Classifier, obtains network output valve, output valve has two
Individual, the big prediction label of first value is 1, is otherwise -1;Classification error rate is calculated according to formula (11):
In formula, ejFor j-th of Weak Classifier classification error rate, Cj(Xk) for j-th of Weak Classifier to k-th of training sample
Prediction label;YkFor the physical tags of k-th of training sample.
According to ejCalculate weight shared by the Weak Classifier:
4) training sample weight is updated according to formula (12):
Wherein, Z is standardizing factor.
5) repeat step 2), 3), 4) and, until 4 classifier trainings are completed.
205:Multiple Weak Classifiers are integrated, strong classifier is formed, realizes the classification to image;
Totally 5 Weak Classifiers trained in 203 and 204 are formed into strong classifier by integrated formula (13):
In summary, the embodiment of the present invention, which uses, trains multiple depth convolutional neural networks as Weak Classifier, and integrates
Multiple Weak Classifiers obtain the method for strong classifier, realize the classification to pathological image;Multiple Weak Classifiers can realize that classification is mutual
Mend, effectively improve classification performance, the fields such as medical diagnosis and pathological analysis can be used in accordingly.
Embodiment 3
Feasibility checking is carried out to the scheme in Examples 1 and 2 below by experimental data, it is described below:
120 width images are selected to make as test image, including 60 illness and 60 healthy images to test image following
Processing:
1) image by image scaling for 512 × 680 sizes;
2) image after scaling is carried out taking block to handle:Tile size is 80 × 80, overlaid pixel 40, per pictures
176 image blocks can be produced;
3) two class testing data inputs are obtained into classification results into strong classifier;
4) classification results are judged:Selected threshold value is 0.5, i.e., for piece image, if having more than total 0.5 times of image block
Image block exports ill label, then it is ill image to judge the figure;Conversely, then judge the image for healthy image.
The accuracy rate of image level is following formula:
In formula:M is certain a kind of overview image number, and m is such correct picture number of classification.
Test result is:P (illness)=0.967, P (health)=0.95.
As a result prove, train multiple convolutional neural networks that it is strong to integrate Weak Classifier classification results composition as Weak Classifier
Grader, there is higher classification accuracy to two class pathological images.Therefore, Pathologic image analysis is may be used as to lead with pathological diagnosis
Domain.Bibliography
[1]Vu T,Mousavi H,Monga V,et al.Histopathological Image
Classification Using Discriminative Feature-oriented Dictionary Learning[J]
.IEEE Transactions on Medical Imaging,2015,35(3):738-751.
[2]Liu Y,Gadepalli K,Norouzi M,et al.Detecting Cancer Metastases
onGigapixelPathology Images[J].arXiv:1703.02442[cs.CV]8Mar 2017.
[3]Hatipoglu N&Bilgin G.Classification of Histopathological Images
Using Convolutional Neural Network.International Conference on Image
Processing Theory,Tools and Applications.IEEE.pp.1-6,2015.
[4]Wang D,Khosla A,Gargeya R,et al.Deep Learning for Identifying
Metastatic Breast Cancer[J].arXiv:1606.05718v1[q-bio.QM]18Jun 2016.
It will be appreciated by those skilled in the art that accompanying drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Sequence number is for illustration only, does not represent the quality of embodiment.
The foregoing is only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent substitution and improvements made etc., it should be included in the scope of the protection.
Claims (5)
- A kind of 1. pathological image sorting technique based on enhancing convolutional neural networks, it is characterised in that the pathological image classification Method comprises the following steps:Fixed size image block is divided into healthy two class pathological pictures by ill, assigns two class image block corresponding labels as instruction Practice sample;Multiple depth convolutional neural networks are trained as Weak Classifier;The multiple Weak Classifiers trained are integrated into strong classification Device, realize the classification to pathological image.
- 2. a kind of pathological image sorting technique based on enhancing convolutional neural networks according to claim 1, its feature exist In two class image block corresponding labels of the imparting are specially as training sample:Still image block size is N × N, overlaid pixel s, from top to bottom, from left to right slides and takes block, and presses block order and protect Store away as training sample;Training sample is normalized, three passages of red, green, blue of image block are normalized respectively, make each of image block Passage average value is 0, variance 1;It is 1 to assign label for normalized ill image block, and it is -1 that healthy image block, which assigns label,.
- 3. a kind of pathological image sorting technique based on enhancing convolutional neural networks according to claim 1, its feature exist In the multiple depth convolutional neural networks of training are specially as Weak Classifier:Multiple convolutional neural networks are built as Weak Classifier, initialization sample weight;Sampled according to weight and obtain Weak Classifier Training sample, train a Weak Classifier;After the completion of training, all training samples are input in the Weak Classifier;The Weak Classifier is calculated to divide training sample Class error rate and the Weak Classifier shared weight in strong classifier;Update the weight of training sample;Until the training of all Weak Classifiers is completed.
- 4. a kind of pathological image sorting technique based on enhancing convolutional neural networks according to claim 3, its feature exist In the convolutional neural networks are specially:Including an input layer, multiple convolutional layers, multiple pond layers and multiple full articulamentums;Each convolutional layer is followed by a pond layer, and last pond layer is followed by multiple full articulamentums, the output conduct of preceding layer The input of later layer, network export the classification predicted value of training sample.
- 5. a kind of pathological image sorting technique based on enhancing convolutional neural networks according to claim 1, its feature exist In described the multiple Weak Classifiers trained are integrated into strong classifier to be specially:All Weak Classifier output labels trained are weighted summation, form strong classifier.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710883838.3A CN107784319A (en) | 2017-09-26 | 2017-09-26 | A kind of pathological image sorting technique based on enhancing convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710883838.3A CN107784319A (en) | 2017-09-26 | 2017-09-26 | A kind of pathological image sorting technique based on enhancing convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107784319A true CN107784319A (en) | 2018-03-09 |
Family
ID=61433902
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710883838.3A Pending CN107784319A (en) | 2017-09-26 | 2017-09-26 | A kind of pathological image sorting technique based on enhancing convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107784319A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596882A (en) * | 2018-04-10 | 2018-09-28 | 中山大学肿瘤防治中心 | The recognition methods of pathological picture and device |
CN108922599A (en) * | 2018-06-27 | 2018-11-30 | 西南交通大学 | A kind of accurate mask method of medical image lesion point based on MIL |
CN109165544A (en) * | 2018-07-05 | 2019-01-08 | 丁彦青 | Training method, diagnostic equipment and the storage medium of glomerulonephritis pathology Intelligence Classifier |
CN109614869A (en) * | 2018-11-10 | 2019-04-12 | 天津大学 | A kind of pathological image classification method based on multi-scale compress rewards and punishments network |
CN110136113A (en) * | 2019-05-14 | 2019-08-16 | 湖南大学 | A kind of vagina pathology image classification method based on convolutional neural networks |
CN110288573A (en) * | 2019-06-13 | 2019-09-27 | 天津大学 | A kind of mammalian livestock illness automatic testing method |
CN110852325A (en) * | 2019-10-31 | 2020-02-28 | 上海商汤智能科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
CN111583203A (en) * | 2020-04-27 | 2020-08-25 | 齐鲁工业大学 | Pathological image labeling method and system based on deep learning model |
CN111680553A (en) * | 2020-04-29 | 2020-09-18 | 北京联合大学 | Pathological image identification method and system based on depth separable convolution |
CN113436192A (en) * | 2021-08-26 | 2021-09-24 | 深圳科亚医疗科技有限公司 | Pathological image classification learning method, classification system and readable medium |
CN114373095A (en) * | 2021-12-09 | 2022-04-19 | 山东师范大学 | Alzheimer disease classification system and method based on image information |
-
2017
- 2017-09-26 CN CN201710883838.3A patent/CN107784319A/en active Pending
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596882B (en) * | 2018-04-10 | 2019-04-02 | 中山大学肿瘤防治中心 | The recognition methods of pathological picture and device |
CN108596882A (en) * | 2018-04-10 | 2018-09-28 | 中山大学肿瘤防治中心 | The recognition methods of pathological picture and device |
US10970842B2 (en) | 2018-04-10 | 2021-04-06 | Sun Yat-Sen University Cancer Center | Method and device for identifying pathological picture |
CN108922599A (en) * | 2018-06-27 | 2018-11-30 | 西南交通大学 | A kind of accurate mask method of medical image lesion point based on MIL |
CN109165544A (en) * | 2018-07-05 | 2019-01-08 | 丁彦青 | Training method, diagnostic equipment and the storage medium of glomerulonephritis pathology Intelligence Classifier |
CN109614869A (en) * | 2018-11-10 | 2019-04-12 | 天津大学 | A kind of pathological image classification method based on multi-scale compress rewards and punishments network |
CN109614869B (en) * | 2018-11-10 | 2023-02-28 | 天津大学 | Pathological image classification method based on multi-scale compression reward and punishment network |
CN110136113B (en) * | 2019-05-14 | 2022-06-07 | 湖南大学 | Vagina pathology image classification method based on convolutional neural network |
CN110136113A (en) * | 2019-05-14 | 2019-08-16 | 湖南大学 | A kind of vagina pathology image classification method based on convolutional neural networks |
CN110288573A (en) * | 2019-06-13 | 2019-09-27 | 天津大学 | A kind of mammalian livestock illness automatic testing method |
CN110852325A (en) * | 2019-10-31 | 2020-02-28 | 上海商汤智能科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
CN110852325B (en) * | 2019-10-31 | 2023-03-31 | 上海商汤智能科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
CN111583203A (en) * | 2020-04-27 | 2020-08-25 | 齐鲁工业大学 | Pathological image labeling method and system based on deep learning model |
CN111680553A (en) * | 2020-04-29 | 2020-09-18 | 北京联合大学 | Pathological image identification method and system based on depth separable convolution |
CN113436192A (en) * | 2021-08-26 | 2021-09-24 | 深圳科亚医疗科技有限公司 | Pathological image classification learning method, classification system and readable medium |
CN114373095A (en) * | 2021-12-09 | 2022-04-19 | 山东师范大学 | Alzheimer disease classification system and method based on image information |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107784319A (en) | A kind of pathological image sorting technique based on enhancing convolutional neural networks | |
CN107506761B (en) | Brain image segmentation method and system based on significance learning convolutional neural network | |
CN107665492B (en) | Colorectal panoramic digital pathological image tissue segmentation method based on depth network | |
CN107133616B (en) | Segmentation-free character positioning and identifying method based on deep learning | |
CN108846426B (en) | Polarization SAR classification method based on deep bidirectional LSTM twin network | |
CN111931811B (en) | Calculation method based on super-pixel image similarity | |
CN110647875B (en) | Method for segmenting and identifying model structure of blood cells and blood cell identification method | |
CN109190643A (en) | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment | |
CN104484886B (en) | A kind of dividing method and device of MR images | |
CN106845418A (en) | A kind of hyperspectral image classification method based on deep learning | |
CN101930533B (en) | Device and method for performing sky detection in image collecting device | |
CN108090403A (en) | Face dynamic identification method and system based on 3D convolutional neural network | |
CN105469111B (en) | The object classification method of small sample set based on improved MFA and transfer learning | |
CN105955708A (en) | Sports video lens classification method based on deep convolutional neural networks | |
Guo et al. | Liver steatosis segmentation with deep learning methods | |
CN107909102A (en) | A kind of sorting technique of histopathology image | |
CN104331714A (en) | Image data extraction and neural network modeling-based platinum flotation grade estimation method | |
CN109614869A (en) | A kind of pathological image classification method based on multi-scale compress rewards and punishments network | |
CN113469119A (en) | Cervical cell image classification method based on visual converter and graph convolution network | |
CN113378792A (en) | Weak supervision cervical cell image analysis method fusing global and local information | |
CN104680190B (en) | Object detection method and device | |
CN116630971A (en) | Wheat scab spore segmentation method based on CRF_Resunate++ network | |
CN115546466A (en) | Weak supervision image target positioning method based on multi-scale significant feature fusion | |
CN110188592B (en) | Urine formed component cell image classification model construction method and classification method | |
CN105761207B (en) | Image Super-resolution Reconstruction method based on the insertion of maximum linear block neighborhood |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180309 |
|
WD01 | Invention patent application deemed withdrawn after publication |