CN107909095A - A kind of image-recognizing method based on deep learning - Google Patents
A kind of image-recognizing method based on deep learning Download PDFInfo
- Publication number
- CN107909095A CN107909095A CN201711084130.8A CN201711084130A CN107909095A CN 107909095 A CN107909095 A CN 107909095A CN 201711084130 A CN201711084130 A CN 201711084130A CN 107909095 A CN107909095 A CN 107909095A
- Authority
- CN
- China
- Prior art keywords
- image
- deep learning
- layer
- model
- training
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image-recognizing method based on deep learning, belong to machine learning techniques field.This method shows good deep learning model VGG models in image processing field by improving, and full articulamentum is replaced using convolutional layer;Deep learning feature extraction is carried out to image, then the deep learning feature extracted, is trained by support vector machines (SVM), finally carries out Classification and Identification.The present invention proves that image-recognizing method of the invention can be effectively according to different criteria for classifications, and to judge the medical image generic of input, while the invention can greatly reduce calculating cost, to aid in diagnosis disease by the identification to medical image.Method using the present invention, can aid in diagnosis disease from objective angle, meet the diagnosis requirement of doctor, diagnosis efficiency is improved, so as to effectively reduce misdiagnosis rate.
Description
Technical field
Technical field belonging to the present invention is art of image analysis, is specially a kind of image recognition side based on deep learning
Method.
Background technology
Machine learning method is widely used in graphical analysis, by the data-oriented collection training pattern complete new data
On particular task, such as classification, identification and segmentation etc..Common algorithm has support vector machines (SVM), Hidden Markov
(HMM) and artificial neural network etc..However, traditional machine learning algorithm needs to utilize priori people from initial data
Work extracts feature, so that training pattern.Since Feature Selection difficulty is larger, there may be over-fitting problem, generalization ability for model
It is difficult to ensure that;On the other hand, conventional model is difficult in adapt to large-scale dataset, model poor expandability.
Deep learning is a new field in machine learning research, its motivation, which is to establish, simulates human brain is divided
Analysis study.Deep learning is a kind of data driven type model, can simulate human brain vision mechanism automatically learn it is each to data
The abstract characteristics of level, so as to preferably reflect the substantive characteristics of data.Good result of the deep learning model in every field
The upsurge for carrying out data mining and analysis using the technology in more areas is triggered, has also caused in medicine and biological cognitive domain
Pay attention to.Deep learning has begun to set foot in the lesion classification of medical image, segmentation, identification and cerebral function research etc. at present
Aspect.Deep learning venture company Enlitic develops the cancer detection system based on deep learning, the lung on chest CT image
Cancer recall rate exceedes doctor.IBM proposes Watson for Oncology, by learning mass data and experience, so as to analyze disease
The medical information of people, helps doctor to formulate reliable medical scheme.The Google subsidiaries DeepMind for developing AlphaGo is near
Day announces DeepMind Health projects, and effective health care science and technology is further developed using deep learning.
In place of the deficiencies in the prior art:
(1) manual extraction is needed to be adapted to the feature of medical image, but it is suitable special for different medical image extractions
Sign requires a great deal of time, and needs the experience of substantial amounts of association area.
(2) in order to obtain more preferable deep learning model, it is necessary to which substantial amounts of training data, data volume is too small may to cause mould
The over-fitting of type.Therefore a kind of method of good EDS extended data set is proposed in the present invention.Designed with reference to the present invention
Frame, can obtain preferable result on small-scale data set
The content of the invention
Present invention aims to use convolutional neural networks, the deep learning model of training image, extracts the depth of image
Spend learning characteristic and be built into database, Classification and Identification is carried out to feature finally by SVM, each process includes several and walks
Suddenly, comprise the following steps that:
Step 1, gather and expand image data set:Limited image is pre-processed, the sample of image is expanded
Fill;
Step 2, convolutional neural networks are designed:The convolutional neural networks model is on the basis of existing VGG-16 models
It is upper to replace full connection layer building using convolutional layer;
Step 3, training convolutional neural networks:Using back-propagation algorithm and stochastic gradient descent method, according to preceding to biography
The size for the loss values broadcast, to carry out the weight that backpropagation iteration updates each layer, until the loss values of model are intended to receive
When holding back, deconditioning model, obtains deep learning model;
Step 4, the feature of image is extracted:Every piece image in data set is input to the deep learning described in step 3
In model, for the image of input, in the full convolutional layer extraction deep learning feature of the layer second from the bottom of image;
Step 5, image is identified:For giving any one image to be identified, trained deep learning mould is input to
In type, the deep learning feature of sample is extracted, effectively differentiates which classification the image belongs to by method trained twice.
Further, the method for collection and the expansion of data set includes at least one of following methods in the step 1:
Mirror image switch:Original image is subjected to left and right reversion, by 2 times of data extending;
Add salt-pepper noise:Some salt-pepper noises will be added in original image, by 2 times of data extending;
Segmentation figure:Target area in artwork is split, other regions are replaced with 0, by 2 times of data extending.
Add illumination:Image is rotated by 90 °, 180 degree, 270 degree, by 3 times of data extending.
Further, the convolutional neural networks model of the step 2 is by input layer, hidden layer, output layer composition:
The input layer, by the data set obtained after expansion, data set is converted into by the image that step 1 proposes
Input of the lmdb forms as the model;
The hidden layer includes convolutional layer and pond layer;Convolution kernel is sized to 1*1 in the first layer of the convolutional layer,
Full connection is replaced using full convolution in layer second from the bottom, characteristic dimension is reduced to the half connected entirely.The pond layer uses
The convolution kernel of 2*2, is applied in combination in first five layer and convolutional layer of the model;
The output layer is connected with last full convolutional layer, the classification number phase of the dimension of output and image to be identified
Deng.
Further, the method that transfer learning is used during step 3 training convolutional neural networks, will be trained in advance
The weight of VGG models is migrated to the convolutional neural networks model.
Further, the training convolutional neural networks described in the step 3 further include:In the caffe frames of Ubuntu systems
Under, view data is divided into training set, training set label, test set, test set label.
Further, the step 4 further includes:Each sample in data set is corresponded into a row vector, and is sticked pair
The label answered, is built into the database of a deep learning feature.
Further, the dimension of the deep learning feature is 2048 dimensions.
Further, the training twice described in the step 5 is respectively:Training for the first time is the side by deep learning
Method, extracts an effective deep learning feature;Second of training is by the deep learning data having had been built up in step 4
Storehouse, by the use of SVM methods using the data in database and label as the training set of SVM, the depth obtained further according to arbitrary sample
Spend learning characteristic, the test set as SVM.
Beneficial effects of the present invention:
(1) test can be automatically identified according to the classification of trained deep learning model and SVM in the present invention
The classification of sample, without inputting any parameter.
(2) by deep learning and the dual training of SVM, the accuracy rate of image recognition can be greatly improved, with more reality
The property used.Apply in medical image recognition, can quickly provide identification image as a result, help diagnosis lesion.
Brief description of the drawings
Fig. 1 is the flow chart that the present invention is implemented.
Fig. 2 is the structure chart of deep learning network.
Embodiment
The invention will be further described with specific embodiment below in conjunction with the accompanying drawings.The present invention is identified as with medical image
Embodiment illustrates, but the present invention is not limited thereto.
As shown in Figure 1, method proposed by the present invention includes the following steps:
Step 1, the collection and expansion of medical images data sets
The purpose of collection and the expansion of medical image be in order to expand the sample size of deep learning training set, prevent due to
Training data is smaller to cause model over-fitting to occur.Mainly in the following ways:
Traditional medical image amount is less, and training deep learning model needs substantial amounts of training sample.Therefore this hair
It is bright by being pre-processed to limited medical image, to expand the sample of medical image.The medical image of the present invention
Collection and the purpose of expansion be to increase the sample size of deep learning training set, prevent since training data is less, can
Model over-fitting can be caused to occur.The collection and expansion of the data set mainly use following methods:
Mirror image switch:Original image is subjected to left and right reversion, by 2 times of data extending.
Add salt-pepper noise:Some salt-pepper noises will be added in original image, by 2 times of data extending.
Segmentation figure:Target area in artwork is split, other regions are replaced with 0, by 2 times of data extending.
Illumination is added, image is rotated by 90 °, 180 degree, 270 degree, data set is expanded 3 times.
Step 2, the design of convolutional neural networks
As shown in Fig. 2, convolutional neural networks model of the present invention is improved on the basis of existing VGG-16 models.
The mode input layer, hidden layer, output layer composition.But since the model in latter two layers of hidden layer is full articulamentum, and
The dimension of full articulamentum be 4096 dimensions, result in that the training parameter of the model is larger, and speed trained under equal conditions
Relatively slow, the convergence rate of model is slower.Therefore the present invention has abandoned full articulamentum, and full articulamentum is replaced with convolutional layer, because volume
Product operation also can effectively extract the feature of image.2048 will be tuned under latter two layers of characteristic dimension, so can preferably carried
The convergence rate of high model, and more preferable model can be obtained.And use average Chi Huadai in softmax layers of preceding layers
For maximum pond.
A. input layer
Input layer, by the data set obtained after expansion, data set is converted into by the medical image that step 1 proposes
Input of the lmdb forms as the model.
B. hidden layer,
Convolutional layer:First layer convolutional layer by the size of convolution kernel by 3*3 under be tuned into 1*1 so that convolution operation obtains
Characteristic pattern can more hold the detailed information of image.In the layer second from the bottom of image connection, and by spy entirely is replaced with full convolution
Sign dimension is reduced to the half connected entirely, substantially reduces the computation complexity of model, and improve the convergence rate of model.It is each
Layer port number is identical with input layer.
Pond layer:Using the convolution kernel of 2*2, it is applied in combination in first five layer and convolutional layer of the model.
C. output layer:
Output layer is connected with last full convolutional layer, and the dimension of output is equal with the classification number of medical image.The figure
The classification number of picture corresponds to the species number of disease.
Step 3, the training of convolutional neural networks
Under the caffe frames of Ubuntu systems, medical image is divided into training set, training set label, test set,
Test set label.Using back-propagation algorithm and stochastic gradient descent method, according to the size of the loss values of propagated forward, into
Row backpropagation iteration updates each layer of weight.When the loss values of model are intended to convergence, deconditioning model.In order to
Improve the convergence rate of model of the present invention, the method for introducing transfer learning, by the weight of trained VGG models in advance migrate to
On the model of the present invention, the weight of each layer of initial random assignment is avoided.Trained cycle, Yi Jineng can effectively be shortened
Access more preferable deep learning model.
Step 4, the feature extraction of medical image
Every piece image in data set is input in the deep learning model described in step 3.For the image of input,
The deep learning feature of 2048 dimensions is extracted in the full convolutional layer of the layer second from the bottom of image.Therefore a medical image sample
The vector of one 2048 dimension is can be obtained by, each sample in data set is corresponded into a row vector.And stick corresponding
Label, is built into the database of a deep learning feature.
Step 5, the identification of medical image.
For giving any one medical image to be identified, it is input in trained deep learning model, extracts sample
This deep learning feature.The present invention is trained, training for the first time is to pass through depth twice in order to improve the accuracy rate of identification
The method of study, extracts an effective deep learning feature;Second of training is by the depth having had been built up in step 4
Learning database, the training set by the use of traditional SVM methods using the data in database and label as SVM, further according to any
The deep learning feature that sample obtains, the test set as SVM.The medicine figure is effectively differentiated by method trained twice
As which classification belonged to.Experiment shows that this method can effectively improve the recognition accuracy of medical image.
Fig. 2 illustrates the method deep learning network structure using the present invention.The embodiment of the present invention is using improved
VGG-16 models.Comprise the following steps that:
1) manually medical images data sets are expanded, sets training set, training set label, test set, test set mark
Label;
2) by ready-made data set by under caffe environment, passing through the classification of the improved VGG-16 model specifications needs of Fig. 2
Num_oupt value, start training data, using back-propagation algorithm learning algorithm, and stochastic gradient descent method.According to
The size of the loss values of propagated forward, to carry out the weight that backpropagation iteration updates each layer.Until the loss values of model become
To in convergence when, deconditioning model.
3) batch input image, in the preceding layer of the softmax classification layers of model, 2048 dimensions of full convolutional layer extraction image
Characteristic of division, and as the training set of SVM.
4) for the training set of SVM obtained above, we set the label of training set, and our inputs of test set are then
It is the feature of the corresponding deep learning of test sample of test required for us.The accuracy rate that may finally be classified.
Those listed above is a series of to be described in detail only for feasibility embodiment of the invention specifically
Bright, they simultaneously are not used to limit the scope of the invention, all equivalent implementations made without departing from skill spirit of the present invention
Or change should all be included in the protection scope of the present invention.
Claims (9)
1. a kind of image-recognizing method based on deep learning, it is characterised in that include the following steps:
Step 1, gather and expand image data set:Limited image is pre-processed, the sample of image is expanded;
Step 2, convolutional neural networks are designed:The convolutional neural networks model is sharp on the basis of existing VGG-16 models
Replaced connecting layer building entirely with convolutional layer;
Step 3, training convolutional neural networks:Using back-propagation algorithm and stochastic gradient descent method, according to propagated forward
The size of loss values, to carry out the weight that backpropagation iteration updates each layer, when the loss values of model are intended to convergence,
Deconditioning model, obtains deep learning model;
Step 4, the feature of image is extracted:Every piece image in data set is input to the deep learning model described in step 3
In, for the image of input, in the full convolutional layer extraction deep learning feature of the layer second from the bottom of image;
Step 5, image is identified:For giving any one image to be identified, it is input in trained deep learning model,
The deep learning feature of sample is extracted, effectively differentiates which classification the image belongs to by method trained twice.
A kind of 2. image-recognizing method based on deep learning according to claim 1, it is characterised in that the step 1
The method of collection and the expansion of middle data set includes at least one of following methods:
Mirror image switch:Original image is subjected to left and right reversion, by 2 times of data extending;
Add salt-pepper noise:Some salt-pepper noises will be added in original image, by 2 times of data extending;
Segmentation figure:Target area in artwork is split, other regions are replaced with 0, by 2 times of data extending.
Add illumination:Image is rotated by 90 °, 180 degree, 270 degree, by 3 times of data extending.
A kind of 3. image-recognizing method based on deep learning according to claim 1, it is characterised in that the step 2
Convolutional neural networks model by input layer, hidden layer, output layer composition:
Data set, by the data set obtained after expansion, is converted into lmdb lattice by the input layer by the image that step 1 proposes
Input of the formula as the model;
The hidden layer includes convolutional layer and pond layer;Convolution kernel is sized to 1*1 in the first layer of the convolutional layer, is falling
The number second layer replaces full connection using full convolution, and characteristic dimension is reduced to the half connected entirely.The pond layer is using 2*2's
Convolution kernel, is applied in combination in first five layer and convolutional layer of the model;
The output layer is connected with last full convolutional layer, and the dimension of output is equal with the classification number of image to be identified.
4. a kind of image-recognizing method based on deep learning according to claim 1, it is characterised in that step 3 is trained
The method that transfer learning is used during convolutional neural networks, the weight of advance trained VGG models is migrated to described
On convolutional neural networks model.
A kind of 5. image-recognizing method based on deep learning according to claim 1, it is characterised in that the step 3
The training convolutional neural networks further include:Under the caffe frames of Ubuntu systems, view data is divided into training set,
Training set label, test set, test set label.
A kind of 6. image-recognizing method based on deep learning according to claim 1, it is characterised in that the step 4
Further include:Each sample in data set is corresponded into a row vector, and sticks corresponding label, is built into a depth
Practise the database of feature.
A kind of 7. image-recognizing method based on deep learning according to claim 1, it is characterised in that the depth
The dimension for practising feature is 2048 dimensions.
A kind of 8. image-recognizing method based on deep learning according to claim 1, it is characterised in that the step 5
Described in training twice be respectively:Training for the first time is the method by deep learning, extracts an effective deep learning
Feature;Second training is by the deep learning database having had been built up in step 4, using SVM methods by database
Training set as SVM of data and label, the deep learning feature obtained further according to arbitrary sample, the test as SVM
Collection.
9. according to a kind of image-recognizing method based on deep learning of claim 1-8 any one of them, it is characterised in that institute
State image-recognizing method and be applied to medical image recognition.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711084130.8A CN107909095A (en) | 2017-11-07 | 2017-11-07 | A kind of image-recognizing method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711084130.8A CN107909095A (en) | 2017-11-07 | 2017-11-07 | A kind of image-recognizing method based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107909095A true CN107909095A (en) | 2018-04-13 |
Family
ID=61842759
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711084130.8A Pending CN107909095A (en) | 2017-11-07 | 2017-11-07 | A kind of image-recognizing method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107909095A (en) |
Cited By (38)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647741A (en) * | 2018-05-18 | 2018-10-12 | 湖北工业大学 | A kind of image classification method and system based on transfer learning |
CN108665457A (en) * | 2018-05-16 | 2018-10-16 | 腾讯科技(深圳)有限公司 | Image-recognizing method, device, storage medium and computer equipment |
CN108875794A (en) * | 2018-05-25 | 2018-11-23 | 中国人民解放军国防科技大学 | Image visibility detection method based on transfer learning |
CN108985222A (en) * | 2018-07-12 | 2018-12-11 | 天津艾思科尔科技有限公司 | A kind of deep learning network model and system for making and receiving calls identification |
CN109064461A (en) * | 2018-08-06 | 2018-12-21 | 长沙理工大学 | A kind of detection method of surface flaw of steel rail based on deep learning network |
CN109117822A (en) * | 2018-08-31 | 2019-01-01 | 贵州大学 | A kind of part case segmentation recognition method based on deep learning |
CN109190721A (en) * | 2018-09-13 | 2019-01-11 | 深圳增强现实技术有限公司 | Fastener recognition methods, system and electronic equipment |
CN109272107A (en) * | 2018-08-10 | 2019-01-25 | 广东工业大学 | A method of improving the number of parameters of deep layer convolutional neural networks |
CN109272045A (en) * | 2018-09-25 | 2019-01-25 | 华南农业大学 | A kind of fruit image classification method and device based on neural network and transfer learning |
CN109299733A (en) * | 2018-09-12 | 2019-02-01 | 江南大学 | The method for carrying out image recognition using compact depth convolutional neural networks |
CN109344884A (en) * | 2018-09-14 | 2019-02-15 | 腾讯科技(深圳)有限公司 | The method and device of media information classification method, training picture classification model |
CN109359551A (en) * | 2018-09-21 | 2019-02-19 | 深圳市璇玑实验室有限公司 | A kind of nude picture detection method and system based on machine learning |
CN109859203A (en) * | 2019-02-20 | 2019-06-07 | 福建医科大学附属口腔医院 | Defect dental imaging recognition methods based on deep learning |
CN110021022A (en) * | 2019-02-21 | 2019-07-16 | 哈尔滨理工大学 | A kind of thyroid gland nuclear medical image diagnostic method based on deep learning |
CN110211118A (en) * | 2019-06-04 | 2019-09-06 | 沈阳智朗科技有限公司 | Gastric cancer image-recognizing method and system based on deep learning |
CN110309841A (en) * | 2018-09-28 | 2019-10-08 | 浙江农林大学 | A kind of hickory nut common insect pests recognition methods based on deep learning |
CN110309846A (en) * | 2019-04-26 | 2019-10-08 | 哈尔滨理工大学 | A kind of image-recognizing method based on deep space domain network |
CN110378408A (en) * | 2019-07-12 | 2019-10-25 | 台州宏创电力集团有限公司 | Power equipment image-recognizing method and device based on transfer learning and neural network |
CN110399815A (en) * | 2019-07-12 | 2019-11-01 | 淮阴工学院 | A kind of CNN-SVM Handwritten Signature Recognition Method based on VGG16 |
CN110472518A (en) * | 2019-07-24 | 2019-11-19 | 杭州晟元数据安全技术股份有限公司 | A kind of fingerprint image quality judgment method based on full convolutional network |
CN110728284A (en) * | 2018-06-29 | 2020-01-24 | 南京汇承科技有限公司 | Edge calculation firework identification method based on deep learning and intelligent terminal |
CN110859642A (en) * | 2019-11-26 | 2020-03-06 | 北京华医共享医疗科技有限公司 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
CN111093140A (en) * | 2019-12-11 | 2020-05-01 | 上海闻泰信息技术有限公司 | Method, device, equipment and storage medium for detecting defects of microphone and earphone dust screen |
CN111105010A (en) * | 2018-10-26 | 2020-05-05 | 斯特拉德视觉公司 | Method and apparatus for using multiple tagged databases with different tag sets |
CN111261278A (en) * | 2018-11-30 | 2020-06-09 | 上海图灵医疗科技有限公司 | Heart disease detection method based on deep learning model of three-dimensional image |
CN111353947A (en) * | 2018-12-24 | 2020-06-30 | 深圳先进技术研究院 | Magnetic resonance parallel imaging method and related equipment |
CN111462082A (en) * | 2020-03-31 | 2020-07-28 | 重庆金山医疗技术研究院有限公司 | Focus picture recognition device, method and equipment and readable storage medium |
CN111476301A (en) * | 2019-12-26 | 2020-07-31 | 山东中科先进技术研究院有限公司 | Medical image classification method and system based on machine learning |
CN111536571A (en) * | 2020-04-23 | 2020-08-14 | 山东建筑大学 | Heat storage and supply system and leakage detection method |
CN111783802A (en) * | 2020-07-21 | 2020-10-16 | 南昌航空大学 | Metal fracture image identification method based on deep learning |
CN111860585A (en) * | 2020-06-11 | 2020-10-30 | 南京简凡科技有限公司 | Automatic fatty liver detection method based on ultrasonic image artificial intelligence algorithm |
CN112508914A (en) * | 2020-12-10 | 2021-03-16 | 内江师范学院 | Rib fracture image detection method based on small sample deep learning |
CN113194292A (en) * | 2021-04-30 | 2021-07-30 | 招商局金融科技有限公司 | Coordination management method, device, equipment and storage medium for intelligent equipment |
CN113516097A (en) * | 2021-07-29 | 2021-10-19 | 东北大学秦皇岛分校 | Plant leaf disease identification method based on improved EfficentNet-V2 |
CN113704388A (en) * | 2021-03-05 | 2021-11-26 | 腾讯科技(深圳)有限公司 | Training method and device for multi-task pre-training model, electronic equipment and medium |
CN113822231A (en) * | 2021-11-08 | 2021-12-21 | 中国人民解放军陆军特色医学中心 | Inter-trochanter fracture operation auxiliary system based on deep learning image recognition |
CN115631386A (en) * | 2022-12-19 | 2023-01-20 | 天津医之本医疗科技有限公司 | Pathological image classification method and system based on machine learning |
WO2023143129A1 (en) * | 2022-01-30 | 2023-08-03 | 北京字跳网络技术有限公司 | Image processing method and apparatus, electronic device, and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6961719B1 (en) * | 2002-01-07 | 2005-11-01 | The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration | Hybrid neural network and support vector machine method for optimization |
CN106599907A (en) * | 2016-11-29 | 2017-04-26 | 北京航空航天大学 | Multi-feature fusion-based dynamic scene classification method and apparatus |
CN106951872A (en) * | 2017-03-24 | 2017-07-14 | 江苏大学 | A kind of recognition methods again of the pedestrian based on unsupervised depth model and hierarchy attributes |
CN107301406A (en) * | 2017-07-13 | 2017-10-27 | 珠海多智科技有限公司 | Fast face angle recognition method based on deep learning |
CN107341506A (en) * | 2017-06-12 | 2017-11-10 | 华南理工大学 | A kind of Image emotional semantic classification method based on the expression of many-sided deep learning |
-
2017
- 2017-11-07 CN CN201711084130.8A patent/CN107909095A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6961719B1 (en) * | 2002-01-07 | 2005-11-01 | The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration | Hybrid neural network and support vector machine method for optimization |
CN106599907A (en) * | 2016-11-29 | 2017-04-26 | 北京航空航天大学 | Multi-feature fusion-based dynamic scene classification method and apparatus |
CN106951872A (en) * | 2017-03-24 | 2017-07-14 | 江苏大学 | A kind of recognition methods again of the pedestrian based on unsupervised depth model and hierarchy attributes |
CN107341506A (en) * | 2017-06-12 | 2017-11-10 | 华南理工大学 | A kind of Image emotional semantic classification method based on the expression of many-sided deep learning |
CN107301406A (en) * | 2017-07-13 | 2017-10-27 | 珠海多智科技有限公司 | Fast face angle recognition method based on deep learning |
Non-Patent Citations (3)
Title |
---|
SHELHAMER E ETAL.: "Fully Convolutional Networks for Semantic Segmentation", 《IEEE TRANSACTIONS ON PATTERNANALYSIS & MACHINE INTELLIGENCE》 * |
周先得: "HEp-2样本图片阴阳性分类算法研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
戚超 等: "基于卷积神经网络的运输船舶分类识别方法", 《微型机与应用》 * |
Cited By (47)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108665457A (en) * | 2018-05-16 | 2018-10-16 | 腾讯科技(深圳)有限公司 | Image-recognizing method, device, storage medium and computer equipment |
CN108665457B (en) * | 2018-05-16 | 2023-12-19 | 腾讯医疗健康(深圳)有限公司 | Image recognition method, device, storage medium and computer equipment |
CN108647741A (en) * | 2018-05-18 | 2018-10-12 | 湖北工业大学 | A kind of image classification method and system based on transfer learning |
CN108875794A (en) * | 2018-05-25 | 2018-11-23 | 中国人民解放军国防科技大学 | Image visibility detection method based on transfer learning |
CN108875794B (en) * | 2018-05-25 | 2020-12-04 | 中国人民解放军国防科技大学 | Image visibility detection method based on transfer learning |
CN110728284A (en) * | 2018-06-29 | 2020-01-24 | 南京汇承科技有限公司 | Edge calculation firework identification method based on deep learning and intelligent terminal |
CN108985222B (en) * | 2018-07-12 | 2024-02-20 | 天津艾思科尔科技有限公司 | Deep learning network model and system for recognition of incoming calls |
CN108985222A (en) * | 2018-07-12 | 2018-12-11 | 天津艾思科尔科技有限公司 | A kind of deep learning network model and system for making and receiving calls identification |
CN109064461A (en) * | 2018-08-06 | 2018-12-21 | 长沙理工大学 | A kind of detection method of surface flaw of steel rail based on deep learning network |
CN109272107A (en) * | 2018-08-10 | 2019-01-25 | 广东工业大学 | A method of improving the number of parameters of deep layer convolutional neural networks |
CN109117822A (en) * | 2018-08-31 | 2019-01-01 | 贵州大学 | A kind of part case segmentation recognition method based on deep learning |
CN109299733A (en) * | 2018-09-12 | 2019-02-01 | 江南大学 | The method for carrying out image recognition using compact depth convolutional neural networks |
CN109190721A (en) * | 2018-09-13 | 2019-01-11 | 深圳增强现实技术有限公司 | Fastener recognition methods, system and electronic equipment |
CN109344884A (en) * | 2018-09-14 | 2019-02-15 | 腾讯科技(深圳)有限公司 | The method and device of media information classification method, training picture classification model |
CN109344884B (en) * | 2018-09-14 | 2023-09-12 | 深圳市雅阅科技有限公司 | Media information classification method, method and device for training picture classification model |
CN109359551A (en) * | 2018-09-21 | 2019-02-19 | 深圳市璇玑实验室有限公司 | A kind of nude picture detection method and system based on machine learning |
CN109272045A (en) * | 2018-09-25 | 2019-01-25 | 华南农业大学 | A kind of fruit image classification method and device based on neural network and transfer learning |
CN110309841A (en) * | 2018-09-28 | 2019-10-08 | 浙江农林大学 | A kind of hickory nut common insect pests recognition methods based on deep learning |
CN111105010B (en) * | 2018-10-26 | 2023-09-29 | 斯特拉德视觉公司 | Method and apparatus for using multiple tagged databases with different tag sets |
CN111105010A (en) * | 2018-10-26 | 2020-05-05 | 斯特拉德视觉公司 | Method and apparatus for using multiple tagged databases with different tag sets |
CN111261278A (en) * | 2018-11-30 | 2020-06-09 | 上海图灵医疗科技有限公司 | Heart disease detection method based on deep learning model of three-dimensional image |
CN111353947A (en) * | 2018-12-24 | 2020-06-30 | 深圳先进技术研究院 | Magnetic resonance parallel imaging method and related equipment |
CN109859203A (en) * | 2019-02-20 | 2019-06-07 | 福建医科大学附属口腔医院 | Defect dental imaging recognition methods based on deep learning |
CN109859203B (en) * | 2019-02-20 | 2022-10-14 | 福建医科大学附属口腔医院 | Defect tooth image identification method based on deep learning |
CN110021022A (en) * | 2019-02-21 | 2019-07-16 | 哈尔滨理工大学 | A kind of thyroid gland nuclear medical image diagnostic method based on deep learning |
CN110309846A (en) * | 2019-04-26 | 2019-10-08 | 哈尔滨理工大学 | A kind of image-recognizing method based on deep space domain network |
CN110211118A (en) * | 2019-06-04 | 2019-09-06 | 沈阳智朗科技有限公司 | Gastric cancer image-recognizing method and system based on deep learning |
CN110399815A (en) * | 2019-07-12 | 2019-11-01 | 淮阴工学院 | A kind of CNN-SVM Handwritten Signature Recognition Method based on VGG16 |
CN110378408A (en) * | 2019-07-12 | 2019-10-25 | 台州宏创电力集团有限公司 | Power equipment image-recognizing method and device based on transfer learning and neural network |
CN110472518A (en) * | 2019-07-24 | 2019-11-19 | 杭州晟元数据安全技术股份有限公司 | A kind of fingerprint image quality judgment method based on full convolutional network |
CN110472518B (en) * | 2019-07-24 | 2022-05-17 | 杭州晟元数据安全技术股份有限公司 | Fingerprint image quality judgment method based on full convolution network |
CN110859642A (en) * | 2019-11-26 | 2020-03-06 | 北京华医共享医疗科技有限公司 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
CN110859642B (en) * | 2019-11-26 | 2024-01-23 | 北京华医共享医疗科技有限公司 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
CN111093140A (en) * | 2019-12-11 | 2020-05-01 | 上海闻泰信息技术有限公司 | Method, device, equipment and storage medium for detecting defects of microphone and earphone dust screen |
CN111476301A (en) * | 2019-12-26 | 2020-07-31 | 山东中科先进技术研究院有限公司 | Medical image classification method and system based on machine learning |
CN111462082A (en) * | 2020-03-31 | 2020-07-28 | 重庆金山医疗技术研究院有限公司 | Focus picture recognition device, method and equipment and readable storage medium |
CN111536571A (en) * | 2020-04-23 | 2020-08-14 | 山东建筑大学 | Heat storage and supply system and leakage detection method |
CN111860585A (en) * | 2020-06-11 | 2020-10-30 | 南京简凡科技有限公司 | Automatic fatty liver detection method based on ultrasonic image artificial intelligence algorithm |
CN111783802A (en) * | 2020-07-21 | 2020-10-16 | 南昌航空大学 | Metal fracture image identification method based on deep learning |
CN112508914A (en) * | 2020-12-10 | 2021-03-16 | 内江师范学院 | Rib fracture image detection method based on small sample deep learning |
CN113704388A (en) * | 2021-03-05 | 2021-11-26 | 腾讯科技(深圳)有限公司 | Training method and device for multi-task pre-training model, electronic equipment and medium |
CN113194292A (en) * | 2021-04-30 | 2021-07-30 | 招商局金融科技有限公司 | Coordination management method, device, equipment and storage medium for intelligent equipment |
CN113516097A (en) * | 2021-07-29 | 2021-10-19 | 东北大学秦皇岛分校 | Plant leaf disease identification method based on improved EfficentNet-V2 |
CN113822231A (en) * | 2021-11-08 | 2021-12-21 | 中国人民解放军陆军特色医学中心 | Inter-trochanter fracture operation auxiliary system based on deep learning image recognition |
WO2023143129A1 (en) * | 2022-01-30 | 2023-08-03 | 北京字跳网络技术有限公司 | Image processing method and apparatus, electronic device, and storage medium |
CN115631386B (en) * | 2022-12-19 | 2023-03-10 | 天津医之本医疗科技有限公司 | Pathological image classification method and system based on machine learning |
CN115631386A (en) * | 2022-12-19 | 2023-01-20 | 天津医之本医疗科技有限公司 | Pathological image classification method and system based on machine learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107909095A (en) | A kind of image-recognizing method based on deep learning | |
CN109598733B (en) | Retinal fundus images dividing method based on the full convolutional neural networks of depth | |
Su et al. | Lung nodule detection based on faster R-CNN framework | |
Sahu et al. | FINE_DENSEIGANET: Automatic medical image classification in chest CT scan using Hybrid Deep Learning Framework | |
CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
CN106682616A (en) | Newborn-painful-expression recognition method based on dual-channel-characteristic deep learning | |
CN104346617B (en) | A kind of cell detection method based on sliding window and depth structure extraction feature | |
CN109272048A (en) | A kind of mode identification method based on depth convolutional neural networks | |
CN107506761A (en) | Brain image dividing method and system based on notable inquiry learning convolutional neural networks | |
CN111340819A (en) | Image segmentation method, device and storage medium | |
CN109727253A (en) | Divide the aided detection method of Lung neoplasm automatically based on depth convolutional neural networks | |
CN106296699A (en) | Cerebral tumor dividing method based on deep neural network and multi-modal MRI image | |
CN107945173A (en) | A kind of skin disease detection method and system based on deep learning | |
CN107368859A (en) | Training method, verification method and the lesion pattern recognition device of lesion identification model | |
CN106204587A (en) | Multiple organ dividing method based on degree of depth convolutional neural networks and region-competitive model | |
CN107492095A (en) | Medical image pulmonary nodule detection method based on deep learning | |
CN109345527A (en) | A kind of tumor of bladder detection method based on MaskRcnn | |
CN108363979A (en) | Neonatal pain expression recognition method based on binary channels Three dimensional convolution neural network | |
Chang | Skin cancer reorganization and classification with deep neural network | |
CN106991355A (en) | The face identification method of the analytical type dictionary learning model kept based on topology | |
CN108573284A (en) | Deep learning facial image extending method based on orthogonal experiment analysis | |
CN106022287A (en) | Over-age face verification method based on deep learning and dictionary representation | |
CN109214388B (en) | Tumor segmentation method and device based on personalized fusion network | |
CN107909588A (en) | Partition system under MRI cortex based on three-dimensional full convolutional neural networks | |
CN109215040A (en) | A kind of tumor of breast dividing method based on multiple dimensioned weighting study |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180413 |