CN107958216A - Based on semi-supervised multi-modal deep learning sorting technique - Google Patents
Based on semi-supervised multi-modal deep learning sorting technique Download PDFInfo
- Publication number
- CN107958216A CN107958216A CN201711202305.0A CN201711202305A CN107958216A CN 107958216 A CN107958216 A CN 107958216A CN 201711202305 A CN201711202305 A CN 201711202305A CN 107958216 A CN107958216 A CN 107958216A
- Authority
- CN
- China
- Prior art keywords
- sample
- mode
- classification
- semi
- implicit
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
-
- 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
Abstract
The present invention proposes the classification contribution otherness of the multi-modal information that sample-rich is considered while being classified with deep learning and each mode,Solve the problems, such as that sample is insufficient using semi-supervised method,High spectrum image different modalities data are respectively fed to deep neural network by the present invention,And a large amount of unmarked samples are utilized using semi-supervised method,Feature learning is carried out using the deep neural network based on own coding,All marks and unlabelled data are sent into own coding depth network to be learnt,Respectively different modalities design similar network,Respective initiation parameter is obtained by own coding reconstruct,The implicit attributive classification of marked sample is obtained by the method for cluster,To Unlabeled data,First pass through multiple target depth network calculations its depth characteristics,It is then based on cluster labels and finds approximate marked sample,The mark of unmarked sample is finally predicted according to the label information of marked sample.
Description
Technical field
The present invention relates to the multi-modal information that sample-rich is considered while being classified with deep learning and respectively
The classification contribution otherness of mode, a kind of based on semi-supervised multi-modal of sample deficiency is solved the problems, such as using semi-supervised method
Deep learning sorting technique.
Background technology
The multiple modalities information such as texture, spatial coherence and spectrum that high-spectrum remote sensing has, fusion are so more
The characteristic information of aspect allows high-spectrum remote-sensing to detect more terrestrial object informations, greatly improves the mankind to objective world
Cognitive ability.
More researchers directly merge part modal information using depth model at present, and less researcher is directed to
The classification of mode contributes otherness to build the depth network architecture.In addition, the classification performance of image relies primarily on substantial amounts of training
Sample, and reality often lacks enough marked samples.How on the basis of less marked sample is possessed, depth is kept
The research for learning generalization ability is also fewer.
In known technology, now with many depth models, such as SAE, DBN, DCNN etc..In 2015, Li et al. people
The depth framework based on SAE is proposed to combine bottom level characteristics.Chen et al. propose using another new depth model DBN into
Row classification, DBN model can extract more robust feature, obtain the nicety of grading better than other models.
In terms of depth model DCNN, DCNN can efficiently extract spatial information using local connection, and by shared
Weight reduces training parameter.Zhao et al. by local discriminant embedding grammar compressed spectrum feature, is learnt empty by CNN first
Between feature, the spectral signature after processing and space characteristics are then put into multi-categorizer.Chen et al. proposes DCNN end to end
Frame carries out feature learning and classification hyperspectral imagery.A.Romero et al. proposes to analyze using unsupervised convolutional neural networks
Remote sensing images, this method build a DCNN model by unsupervised greed layering pre-training method.It is few for training sample
And over-fitting problem caused by dimension height, Chen et al. is solved using L2 regularizations and dropout.Li et al. people proposes to utilize
CNN carries out the deficiency that pixel carrys out the thought of tagsort offset data.
The content of the invention
In conclusion it is more that directly part modal information is carried out using depth model in existing known technology research
Fusion, less researcher build the depth network architecture for the classification contribution otherness of mode.The shortcomings that known technology is,
The different interlayers of many depth models such as SAE, DBN use full connection mode, it is necessary to train substantial amounts of parameter, and reality is often
Lack a large amount of available mark training samples.In addition, spatial information in the training stage, is simply expressed as vectorial shape by SAE and DBN
Formula, thus cannot more effectively extract spatial information.
For this model of DANN, although it can efficiently extract spatial information using local connection, and by shared
The mode of weights reduces the quantity of parameter.But it needs many hidden layers just to acquire more complicated data characteristics, may
There is still a need for the parameter that study is very more, especially often leads to over-fitting problem when training samples number is inadequate.
In contrast, existing known technology cannot comprehensively consider how possessing the basis of less marked sample
On, keep the generalization ability of deep learning and how the multi-modal information of sample and few marker samples combined into structure
Deep learning framework.
The defects of in order to overcome in the prior art, proposes the present invention.Present invention incorporates the space field feature of sample,
Textural characteristics, the multi-modal information of spectral signature and build depth network rack with the classification of each modal information contribution otherness
Structure, the problem of also contemplating a small amount of marker samples in addition, utilizes the sample predictions marked on a small quantity big using semi-supervised method
Measure unlabelled sample.
The cost is classification accuracy rate.
The technical solution adopted in the present invention is：Based on semi-supervised multi-modal deep learning sorting technique, including it is as follows
Step：
(1) high spectrum image is divided into spectrum mode, spatial texture mode and spatial coherence mode, for every kind of mode
Separately design corresponding depth convolutional neural networks；
(2) diversity factor, the integrated classification device F of design three mode of fusion are contributed according to the classification of mode；Each mode sheet
Body has the implicit attribute of oneself, therefore different modalities data can be clustered, to obtain its inherent attribute classification, each
Mode all corresponds to an implicit attribute grader, respectively spectrum implicit attribute grader S, texture implicit attribute grader T, with
And correlation implicit attribute grader R.
(3) all marker samples are directed to, its implicit attributive classification is obtained using clustering method；
(4) two labels are possessed by each sample after cluster, the correspondence of the two classifications can be obtained；
(5) cluster analysis is carried out to unlabelled sample, first passes through multiple target depth network calculations its depth characteristics, then
Approximate marked sample is found based on cluster labels, unmarked sample is finally predicted according to the label information of marked sample
Mark；
The step (5) needs to set the confidence standard of predictive marker, does not give a forecast if beyond standard, when certain batch
After secondary unmarked sample is predicted out label information, the classification contribution diversity factor of each mode is adaptively adjusted.
Beneficial effects of the present invention：
The present invention relates to based on semi-supervised multi-modal deep learning sorting technique, it is multi-modal that sample is more considered comprehensively
Information and each mode classification contribution diversity factor, and it is combined with semi-supervised method and solves the problems, such as that marker samples are few,
To improve the generalization ability of depth model.
The multi-modal information of sample-rich and the classification of each mode are considered while being classified with deep learning
Otherness is contributed, a kind of based on semi-supervised multi-modal deep learning point of sample deficiency is solved the problems, such as using semi-supervised method
Class method.This method considers the textural characteristics possessed by pixel, space characteristics, and the multi-modal information of spectral signature melts
Close, while also contemplate contribution diversity factor of each modal information to the depth network architecture, and by multi-modal information to classification
Contribution diversity factor and predict that unmarked sample combines to improve the generalization ability of depth model, be finally reached raising bloom
Compose the accuracy of image classification.
Brief description of the drawings
The present invention is further illustrated below in conjunction with the accompanying drawings.
It is above-mentioned purpose of the invention, excellent by reference to the detailed description below in conjunction with attached drawing to used embodiment
Point and feature will become more apparent from, wherein：
Fig. 1 is the schematic diagram of the overall flow structure of the present invention；
Fig. 2 is the concrete practice step of the semi-supervised multi-modal deep learning frame of embodiment of the present invention；
Fig. 3 is the concrete practice step of the semi-supervised multi-modal deep learning frame of embodiment of the present invention；
Fig. 4 is the particular flow sheet of the half network architecture parameters initialization based on own coding of embodiment of the present invention；
Fig. 5 is the particular flow sheet of the unmarked sample collaborative forecasting method based on multiple target of embodiment of the present invention.
Embodiment
Carry out detailed description of the present invention embodiment below with reference to the accompanying drawings.
Fig. 1 shows the system structure of embodiment of the present invention.102 be high spectrum image sample and label information it is defeated
Enter, 103 represent the pretreatment that our samples to reception carry out, and 104 represent us by the sample packet after processing, 105 be by
Sample after being divided into group is sent in semi-supervised multi-modal deep learning frame respectively to be learnt, and 106 represent each depth
The result of habit frame, which combines, does a decision-making treatment, and 107 be to obtain final classification results.
Fig. 2 and Fig. 3 shows the concrete practice step of the semi-supervised multi-modal deep learning frame of embodiment of the present invention.This
High spectrum image different modalities data are respectively fed to deep neural network by scheme, and are utilized largely using semi-supervised method
Unmarked sample.201 to 203 represent high spectrum image to be divided into spectrum mode, spatial texture mode and spatial coherence mode,
It is respectively 204 to 206 to separately design corresponding depth convolutional neural networks for every kind of mode, and wherein depth network uses
AlexNet frameworks, AlexNet frameworks possess five convolutional layers, two full articulamentums.Diversity factor is contributed such as according to the classification of mode
Shown in 304,306 be the integrated classification device F of design three mode of fusion.Each mode has the implicit attribute of oneself in itself, than
It can be drawn as texture classes and do one kind, therefore different modalities data can clustered, to obtain its inherent attribute classification, often
A mode all corresponds to an implicit attribute grader, and respectively 305 be spectrum implicit attribute grader S, and 307 be that texture implies category
Property grader T, and 308 be correlation implicit attribute grader R.Deep neural network is with sample class and implicit attribute classification
Two class objects carry out the study of abstract characteristics, and specific more class object learning strategies (shown in 207) are as shown in Figure 3.With light
Exemplified by composing mode 301, the learning strategy of more class objects is illustrated, wherein 306 be integrated classification device F, it is with sample class
As target, spectrum implicit attribute grader S is using the implicit attribute of spectrum target, to pass through fusion two as shown in 305
Class object, can acquire the implicit attribute classification based on sample class as shown in 309.
Fig. 4 shows the concrete practice step of the initialization of the network architecture parameters based on own coding of embodiment of the present invention.
High-spectrum remote sensing usually only has the label information of a small amount of high quality, has been difficult with measure of supervision study to suitable net
Network initiation parameter, in order to learn relatively reasonable initial network parameter, will using based on the deep neural network of own coding into
Row feature learning, 402 be that all marks and unlabelled data are all sent into own coding depth network and are learnt, this depth net
Network framework includes five layers of convolutional layer and four layers of coding and self-editing code layer, and 403 be that respectively different modalities design similar network,
404 be that respective initiation parameter is obtained by own coding reconstruct, for certain modal data v of high spectrum image, accordingly
Output be405 be by minimizing squared reconstruction error
Carry out training network, wherein M and N are the quantity of marked sample and unmarked sample respectively, and r represents mode.406 be that we use
Standard back-propagation algorithm based on stochastic gradient descent is come optimized reconstruction network.407 be when the network of each mode reaches receipts
After holding back, the convolution layer parameter in the network structure based on own coding is as corresponding mode in corresponding semi-supervised depth network structure
Convolution layer parameter.
Fig. 5 shows the concrete practice step of the unmarked sample collaborative forecasting method based on multiple target of embodiment of the present invention
Suddenly.For a certain modal data of EO-1 hyperion, 503 be to be directed to all marker samples, and it is implicit to obtain it using clustering method
Attributive classification, two labels are possessed by each sample after cluster：Original class label c and the implicit attribute class after cluster
Distinguishing label c ', 504 be to be directed to marker samples, can obtain the correspondence of the two classifications.For unmarked sample, cluster
Purpose is exactly：According to the cluster result (obtaining c ' labels) of unmarked sample (c labels are unknown), according to c ' in marked sample
Label finds its similar c label.In order to preferably balance class label c and implicit attribute class label c ', 505 kinds I
Using both as target carry out depth model training.Based on the consideration to multi-modal data, for the class object of class label c
By the way of multi-modal fusion, convergence strategy contributes diversity factor, the study of multiple target depth model using above-mentioned classification
Object function be：Here r represents mould
State, wrRepresent the classification contribution degree of r mode.After completing network training based on multiple target loss function, 506 be to be directed to unmarked number
According to first passing through multiple target depth network calculations its depth characteristics, then 507 be to find approximately to have mark sample based on cluster labels
This, 508 be to predict the mark of unmarked sample, it is necessary to set the confidence mark of predictive marker according to the label information of marked sample
Standard, if not giving a forecast if the standard, 509 be after the unmarked sample of certain batch is predicted out label information, can be with
The classification contribution diversity factor of each mode is adaptively adjusted, 510 be when network training tends towards stability, and does not have and meets prediction standard
New samples occur, can deconditioning.After obtaining respective classification results by depth model, determined most by decision combinations
Whole classification results.
Although having been combined the preferred embodiment of the present invention above shows the present invention, those skilled in the art will
It will be appreciated that without departing from the spirit and scope of the present invention, various modifications can be carried out to the present invention, replaces and changes
Become.Therefore, the present invention should not be limited by above-described embodiment, and should be limited by appended claims and its equivalent.
Claims (2)
1. based on semi-supervised multi-modal deep learning sorting technique, it is characterised in that include the following steps：
(1) high spectrum image is divided into spectrum mode, spatial texture mode and spatial coherence mode, distinguished for every kind of mode
Design corresponding depth convolutional neural networks；
(2) diversity factor, the integrated classification device F of design three mode of fusion are contributed according to the classification of mode；Each mode has in itself
There is the implicit attribute of oneself, therefore different modalities data can be clustered, to obtain its inherent attribute classification, each mode
An implicit attribute grader is all corresponded to, is respectively spectrum implicit attribute grader S, texture implicit attribute grader T, Yi Jixiang
Closing property implicit attribute grader R.
(3) all marker samples are directed to, its implicit attributive classification is obtained using clustering method；
(4) two labels are possessed by each sample after cluster, the correspondence of the two classifications can be obtained；
(5) cluster analysis is carried out to unlabelled sample, first passes through multiple target depth network calculations its depth characteristics, be then based on
Cluster labels find approximate marked sample, and the mark of unmarked sample is finally predicted according to the label information of marked sample
Note.
It is 2. according to claim 1 based on semi-supervised multi-modal deep learning sorting technique, it is characterised in that the step
Suddenly (5) need to set the confidence standard of predictive marker, do not give a forecast if beyond standard, when the unmarked sample of certain batch
After being predicted out label information, the classification contribution diversity factor of each mode is adaptively adjusted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711202305.0A CN107958216A (en) | 2017-11-27 | 2017-11-27 | Based on semi-supervised multi-modal deep learning sorting technique |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711202305.0A CN107958216A (en) | 2017-11-27 | 2017-11-27 | Based on semi-supervised multi-modal deep learning sorting technique |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107958216A true CN107958216A (en) | 2018-04-24 |
Family
ID=61961887
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711202305.0A Withdrawn CN107958216A (en) | 2017-11-27 | 2017-11-27 | Based on semi-supervised multi-modal deep learning sorting technique |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107958216A (en) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108664999A (en) * | 2018-05-03 | 2018-10-16 | 北京图森未来科技有限公司 | A kind of training method and its device, computer server of disaggregated model |
| CN109146003A (en) * | 2018-09-30 | 2019-01-04 | 合刃科技（深圳）有限公司 | A kind of classifying identification method based on high light spectrum image-forming, apparatus and system |
| CN109361404A (en) * | 2018-09-28 | 2019-02-19 | 华南理工大学 | A kind of LDPC decoding system and interpretation method based on semi-supervised deep learning network |
| CN109460708A (en) * | 2018-10-09 | 2019-03-12 | 东南大学 | A kind of Forest fire image sample generating method based on generation confrontation network |
| CN109784405A (en) * | 2019-01-16 | 2019-05-21 | 山东建筑大学 | Cross-module state search method and system based on pseudo label study and semantic consistency |
| CN109831392A (en) * | 2019-03-04 | 2019-05-31 | 中国科学技术大学 | Semi-supervised net flow assorted method |
| CN109858557A (en) * | 2019-02-13 | 2019-06-07 | 安徽大学 | A kind of new hyperspectral image data semisupervised classification algorithm |
| CN109978071A (en) * | 2019-04-03 | 2019-07-05 | 西北工业大学 | Hyperspectral image classification method based on data augmentation and Multiple Classifier Fusion |
| CN110009015A (en) * | 2019-03-25 | 2019-07-12 | 西北工业大学 | EO-1 hyperion small sample classification method based on lightweight network and semi-supervised clustering |
| CN110119447A (en) * | 2019-04-26 | 2019-08-13 | 平安科技（深圳）有限公司 | From coding Processing with Neural Network method, apparatus, computer equipment and storage medium |
| WO2019210737A1 (en) * | 2018-05-04 | 2019-11-07 | 上海商汤智能科技有限公司 | Object prediction method and apparatus, electronic device and storage medium |
| CN110479636A (en) * | 2019-07-19 | 2019-11-22 | 深圳市微蓝智能科技有限公司 | Method and device based on neural network automatic sorting tobacco leaf |
| CN111553890A (en) * | 2020-04-22 | 2020-08-18 | 上海全景云医学影像诊断有限公司 | X-ray positive chest radiography multi-task detection method based on incremental learning |
| CN111652177A (en) * | 2020-06-12 | 2020-09-11 | 中国计量大学 | Signal feature extraction method based on deep learning |
| CN111768366A (en) * | 2020-05-20 | 2020-10-13 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging system, BI-RADS classification method and model training method |
| CN112001438A (en) * | 2020-08-19 | 2020-11-27 | 四川大学 | Multi-mode data clustering method for automatically selecting clustering number |
| CN112182225A (en) * | 2020-10-14 | 2021-01-05 | 上海宝钿科技产业发展有限公司 | Knowledge management method for multi-modal scene target based on semi-supervised deep learning |
| CN113140273A (en) * | 2021-03-31 | 2021-07-20 | 厦门大学 | ICU patient electronic medical record analysis method and system based on deep learning |
| CN109242033B (en) * | 2018-09-21 | 2021-08-20 | 长鑫存储技术有限公司 | Wafer defect mode classification method and device, storage medium and electronic equipment |
| CN114742132A (en) * | 2022-03-17 | 2022-07-12 | 湖南工商大学 | Deep multi-view clustering method, system and equipment based on common difference learning |
| WO2022166361A1 (en) * | 2021-02-04 | 2022-08-11 | 浙江师范大学 | Deep clustering method and system based on cross-modal fusion |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101980480A (en) * | 2010-11-04 | 2011-02-23 | 西安电子科技大学 | Semi-supervised anomaly intrusion detection method |
| CN101980202A (en) * | 2010-11-04 | 2011-02-23 | 西安电子科技大学 | Semi-supervised classification method of unbalance data |
| CN102646200A (en) * | 2012-03-08 | 2012-08-22 | 武汉大学 | Image classifying method and system for self-adaption weight fusion of multiple classifiers |
| CN102968639A (en) * | 2012-09-28 | 2013-03-13 | 武汉科技大学 | Semi-supervised image clustering subspace learning algorithm based on local linear regression |
| CN103605990A (en) * | 2013-10-23 | 2014-02-26 | 江苏大学 | Integrated multi-classifier fusion classification method and integrated multi-classifier fusion classification system based on graph clustering label propagation |
| CN103942779A (en) * | 2014-03-27 | 2014-07-23 | 南京邮电大学 | Image segmentation method based on combination of graph theory and semi-supervised learning |
| CN104598813A (en) * | 2014-12-09 | 2015-05-06 | 西安电子科技大学 | Computer intrusion detection method based on integrated study and semi-supervised SVM |
| CN104751182A (en) * | 2015-04-02 | 2015-07-01 | 中国人民解放军空军工程大学 | DDAG-based SVM multi-class classification active learning algorithm |
| CN105303198A (en) * | 2015-11-17 | 2016-02-03 | 福州大学 | Remote-sensing image semi-supervision classification method based on customized step-size learning |
-
2017
- 2017-11-27 CN CN201711202305.0A patent/CN107958216A/en not_active Withdrawn
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101980480A (en) * | 2010-11-04 | 2011-02-23 | 西安电子科技大学 | Semi-supervised anomaly intrusion detection method |
| CN101980202A (en) * | 2010-11-04 | 2011-02-23 | 西安电子科技大学 | Semi-supervised classification method of unbalance data |
| CN102646200A (en) * | 2012-03-08 | 2012-08-22 | 武汉大学 | Image classifying method and system for self-adaption weight fusion of multiple classifiers |
| CN102968639A (en) * | 2012-09-28 | 2013-03-13 | 武汉科技大学 | Semi-supervised image clustering subspace learning algorithm based on local linear regression |
| CN103605990A (en) * | 2013-10-23 | 2014-02-26 | 江苏大学 | Integrated multi-classifier fusion classification method and integrated multi-classifier fusion classification system based on graph clustering label propagation |
| CN103942779A (en) * | 2014-03-27 | 2014-07-23 | 南京邮电大学 | Image segmentation method based on combination of graph theory and semi-supervised learning |
| CN104598813A (en) * | 2014-12-09 | 2015-05-06 | 西安电子科技大学 | Computer intrusion detection method based on integrated study and semi-supervised SVM |
| CN104751182A (en) * | 2015-04-02 | 2015-07-01 | 中国人民解放军空军工程大学 | DDAG-based SVM multi-class classification active learning algorithm |
| CN105303198A (en) * | 2015-11-17 | 2016-02-03 | 福州大学 | Remote-sensing image semi-supervision classification method based on customized step-size learning |
Non-Patent Citations (1)
| Title |
|---|
| 韩玉想: ""基于聚类分析的标签传播半监督学习研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108664999A (en) * | 2018-05-03 | 2018-10-16 | 北京图森未来科技有限公司 | A kind of training method and its device, computer server of disaggregated model |
| WO2019210737A1 (en) * | 2018-05-04 | 2019-11-07 | 上海商汤智能科技有限公司 | Object prediction method and apparatus, electronic device and storage medium |
| US11593596B2 (en) | 2018-05-04 | 2023-02-28 | Shanghai Sense Time Intelligent Technology Co., Ltd. | Object prediction method and apparatus, and storage medium |
| CN109242033B (en) * | 2018-09-21 | 2021-08-20 | 长鑫存储技术有限公司 | Wafer defect mode classification method and device, storage medium and electronic equipment |
| CN109361404A (en) * | 2018-09-28 | 2019-02-19 | 华南理工大学 | A kind of LDPC decoding system and interpretation method based on semi-supervised deep learning network |
| CN109361404B (en) * | 2018-09-28 | 2020-07-28 | 华南理工大学 | L DPC decoding system and method based on semi-supervised deep learning network |
| CN109146003A (en) * | 2018-09-30 | 2019-01-04 | 合刃科技（深圳）有限公司 | A kind of classifying identification method based on high light spectrum image-forming, apparatus and system |
| CN109146003B (en) * | 2018-09-30 | 2022-06-07 | 合刃科技（深圳）有限公司 | Hyperspectral imaging-based classification identification method, device and system |
| CN109460708A (en) * | 2018-10-09 | 2019-03-12 | 东南大学 | A kind of Forest fire image sample generating method based on generation confrontation network |
| CN109784405B (en) * | 2019-01-16 | 2020-09-08 | 山东建筑大学 | Cross-modal retrieval method and system based on pseudo-tag learning and semantic consistency |
| CN109784405A (en) * | 2019-01-16 | 2019-05-21 | 山东建筑大学 | Cross-module state search method and system based on pseudo label study and semantic consistency |
| CN109858557B (en) * | 2019-02-13 | 2022-12-27 | 安徽大学 | Novel semi-supervised classification method for hyperspectral image data |
| CN109858557A (en) * | 2019-02-13 | 2019-06-07 | 安徽大学 | A kind of new hyperspectral image data semisupervised classification algorithm |
| CN109831392A (en) * | 2019-03-04 | 2019-05-31 | 中国科学技术大学 | Semi-supervised net flow assorted method |
| CN110009015A (en) * | 2019-03-25 | 2019-07-12 | 西北工业大学 | EO-1 hyperion small sample classification method based on lightweight network and semi-supervised clustering |
| CN109978071A (en) * | 2019-04-03 | 2019-07-05 | 西北工业大学 | Hyperspectral image classification method based on data augmentation and Multiple Classifier Fusion |
| CN110119447A (en) * | 2019-04-26 | 2019-08-13 | 平安科技（深圳）有限公司 | From coding Processing with Neural Network method, apparatus, computer equipment and storage medium |
| CN110119447B (en) * | 2019-04-26 | 2023-06-16 | 平安科技（深圳）有限公司 | Self-coding neural network processing method, device, computer equipment and storage medium |
| CN110479636B (en) * | 2019-07-19 | 2021-11-09 | 深圳市微蓝智能科技有限公司 | Method and device for automatically sorting tobacco leaves based on neural network |
| CN110479636A (en) * | 2019-07-19 | 2019-11-22 | 深圳市微蓝智能科技有限公司 | Method and device based on neural network automatic sorting tobacco leaf |
| CN111553890B (en) * | 2020-04-22 | 2021-01-08 | 上海全景云医学影像诊断有限公司 | X-ray positive chest radiography multi-task detection method based on incremental learning |
| CN111553890A (en) * | 2020-04-22 | 2020-08-18 | 上海全景云医学影像诊断有限公司 | X-ray positive chest radiography multi-task detection method based on incremental learning |
| CN111768366A (en) * | 2020-05-20 | 2020-10-13 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging system, BI-RADS classification method and model training method |
| CN111652177A (en) * | 2020-06-12 | 2020-09-11 | 中国计量大学 | Signal feature extraction method based on deep learning |
| CN112001438A (en) * | 2020-08-19 | 2020-11-27 | 四川大学 | Multi-mode data clustering method for automatically selecting clustering number |
| CN112001438B (en) * | 2020-08-19 | 2023-01-10 | 四川大学 | Multi-mode data clustering method for automatically selecting clustering number |
| CN112182225A (en) * | 2020-10-14 | 2021-01-05 | 上海宝钿科技产业发展有限公司 | Knowledge management method for multi-modal scene target based on semi-supervised deep learning |
| WO2022166361A1 (en) * | 2021-02-04 | 2022-08-11 | 浙江师范大学 | Deep clustering method and system based on cross-modal fusion |
| CN113140273A (en) * | 2021-03-31 | 2021-07-20 | 厦门大学 | ICU patient electronic medical record analysis method and system based on deep learning |
| CN113140273B (en) * | 2021-03-31 | 2022-05-10 | 厦门大学 | ICU patient electronic medical record analysis method and system based on deep learning |
| CN114742132A (en) * | 2022-03-17 | 2022-07-12 | 湖南工商大学 | Deep multi-view clustering method, system and equipment based on common difference learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107958216A (en) | Based on semi-supervised multi-modal deep learning sorting technique | |
| CN109344736B (en) | Static image crowd counting method based on joint learning | |
| CN110321923B (en) | Target detection method, system and medium for fusion of different-scale receptive field characteristic layers | |
| CN111368896B (en) | Hyperspectral remote sensing image classification method based on dense residual three-dimensional convolutional neural network | |
| CN107609601B (en) | Ship target identification method based on multilayer convolutional neural network | |
| CN108021947B (en) | A kind of layering extreme learning machine target identification method of view-based access control model | |
| CN106815604B (en) | Method for viewing points detecting based on fusion of multi-layer information | |
| CN107527068A (en) | Model recognizing method based on CNN and domain adaptive learning | |
| CN110796168A (en) | Improved YOLOv 3-based vehicle detection method | |
| CN104573669B (en) | Image object detection method | |
| CN107818302A (en) | Non-rigid multiple dimensioned object detecting method based on convolutional neural networks | |
| CN108830334A (en) | A kind of fine granularity target-recognition method based on confrontation type transfer learning | |
| CN106650806A (en) | Cooperative type deep network model method for pedestrian detection | |
| CN106919920A (en) | Scene recognition method based on convolution feature and spatial vision bag of words | |
| CN106991382A (en) | A kind of remote sensing scene classification method | |
| CN104992142A (en) | Pedestrian recognition method based on combination of depth learning and property learning | |
| CN105160400A (en) | L21 norm based method for improving convolutional neural network generalization capability | |
| CN108596203A (en) | Optimization method of the pond layer in parallel to pantograph carbon slide surface abrasion detection model | |
| CN104134068B (en) | Monitoring vehicle characteristics based on sparse coding represent and sorting technique | |
| CN108629367A (en) | A method of clothes Attribute Recognition precision is enhanced based on depth network | |
| CN110163077A (en) | A kind of lane recognition method based on full convolutional neural networks | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| CN105046269A (en) | Multi-instance multi-label scene classification method based on multinuclear fusion | |
| CN109948696A (en) | A kind of multilingual scene character recognition method and system | |
| CN107103308A (en) | A kind of pedestrian's recognition methods again learnt based on depth dimension from coarse to fine |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20180424 |