CN107145893A - A kind of image recognition algorithm and system based on convolution depth network - Google Patents
A kind of image recognition algorithm and system based on convolution depth network Download PDFInfo
- Publication number
- CN107145893A CN107145893A CN201710144957.7A CN201710144957A CN107145893A CN 107145893 A CN107145893 A CN 107145893A CN 201710144957 A CN201710144957 A CN 201710144957A CN 107145893 A CN107145893 A CN 107145893A
- Authority
- CN
- China
- Prior art keywords
- ccdbn
- training
- picture
- input
- image
- 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
- 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
Abstract
The present invention a kind of image recognition algorithm (CCDBN models) and system based on convolution depth network, it is used as sample set CCDBN by picture of the structure with label to be trained, the neutral net trained is preserved, using picture to be identified as input, result is identified according to output vector.Image is recognized by CCDBN, it is to avoid artificial characteristic extraction procedure, directly using picture as the input of network, recognition accuracy is higher；Network one is trained can Reusability, treatment effeciency height.
Description
Technical field
The present invention relates to pattern-recognition and machine learning field, and in particular to depth confidence net (DBN), convolutional neural networks
(CNN), concept and formula and the C++ programming languages such as restricted Boltzmann machine (RBM), recruitment factor, conditional Gaussian distribution
Speech.
Background technology
Machine learning is to study how to make a subject of computer simulation mankind's learning behavior.Machine learning is based on study
Strategy can be divided into rote learning, analogical learning, deductive learning, the study based on explanation, inductive learning, based on neutral net
Practise etc..The emphasis studied herein is at artificial neural network (ANN).Neutral net is a kind of network of parallel distributed information processing
Structure, with very strong non-linear mapping capability and higher fault-tolerant ability.ANN can trace back to nineteen forty-three neuropsychologist
The neuron models (M-P models) that Mcculloch and mathematician Pitts is proposed from the angle of mathematical logic, ANN starts hair since then
Exhibition.At present, common ANN model species is a lot, and we can be divided into three kinds of basic network models according to ANN structure：It is single
Layer feedforward network, Multi-layered Feedforward Networks, Recursive Networks.As multilayer neural network, deep learning model is due to can be from input
Sample directly approaches the Nonlinear Mapping of complexity, and is widely used in many fields, and common model has convolutional neural networks (CNN)
With the autocoder model (SAE) and depth confidence net (DBN) depth Boltzmann machine (DBM) of stacking etc..Convolutional Neural
Network is designed exclusively for processing 2-D data, it is considered to be first deep learning side using multi-layer Network Framework
Method, achieves huge success in field of image recognition in recent years, and based on CNN, extends many deep learning frames
Frame and system, typical framework is including Caffe, Tensorflow etc..Meanwhile, convolution operation can be regarded as a kind of regularization side
Method, due to being shared using local connection and weights, is kept greatly reducing network parameter again while network deep structure, makes mould
There is type good generalization ability to be easier to training again.Therefore, as a kind of regularization method, CNN has also been used in voice
Deng other field and cross discipline, such as deeply study and depth migration study.
It is usually to extract feature in advance in classical pattern-recognition.Extract after all multiple features, these features are carried out
Correlation analysis, the extraction of these features too relies on the experience and subjective consciousness of people, and the difference for the feature extracted is to classification
Performance impact is very big.Meanwhile, the quality of image preprocessing also influences whether the feature extracted.And deep learning algorithm need not pair
Image carries out complicated pretreatment operation, can easily using image as input, by substantial amounts of data come learning characteristic,
The feature extraction of display is avoided, it is more more reliable than conventional artificial selected characteristic.
Limited Boltzmann machine (RBM) is a kind of typical unsupervised learning model, comes from the nothing in probability graph model
To graph model.By training RBM models and successively stacking, a typical depth model structure can be obtained：Depth confidence net
(DBN).But RBM is a kind of two-value model, in order to model real value image, conditional Gaussian distribution is introduced in RBM models, I
According to conditional Gaussian be distributed and energy function form, obtain following activation primitive：
Wherein, v represents visible layer unit, and c represents recruitment factor, and h represents to hide layer unit.Obtained hiding node layer is made
For next layer of input of network, build RBM models and continue pre-training process.
The content of the invention
In order to preferably solve the problems, such as the identification of image, the present invention proposes a kind of image recognition based on convolution depth network
Algorithm and system, it is to avoid explicit extraction feature, directly using digitized image pixel as input, training is obtained based on volume
Long-pending supplement DBN model (CCDBN), then increases an associative memory layer as recognition result, effectively realizes image
Identification and reconstruct.
The present invention is achieved by the following scheme：
The present invention relates to a kind of image recognition algorithm based on convolution depth network, made by the training set for building tape label
Pre-training and training are carried out to convolutional neural networks for sample set, it is then to be identified using the convolutional neural networks processing trained
Picture, the output vector finally according to neutral net judges recognition result.
The present invention is comprised the following steps that：
Step 1, it is simple to pre-process training set and regard pixel as input：Image data set is divided first
Batch, each small batch include 100 samples, then by image normalization and adjusting size is 32*32 again；
Step 2, construction depth learning model network structure：The network includes：Input layer, 3 convolutional layers, 3 pond layers,
1 full linking layer and 1 output layer, wherein：Input layer is the pixel of two dimensional image, and first convolutional layer has 24 convolution spies
Figure is levied, second convolutional layer there are 48 convolution characteristic patterns, and output layer is set to 10 nodes.Output is also the prediction of system；
Step 3, deep neural network is trained：Weights, convolution kernel, biasing are initialized first.Then instructed in advance
Practice process, the first step：RBM models are improved, conditional Gaussian distribution is introduced, structure is restricted Boltzmann machine based on recruitment factor
(CRBM) convolution operation, is then introduced, a CRBM (CCRBM) based on convolution is built, its network topology is as shown in Figure 1.Connect
Get off using weight uncertainty methods and to sdpecific dispersion algorithm to train CCRBM models to alleviate over-fitting problem.
Then, based on one supplement depth confidence net (CCDBN) based on convolution of CCRBM model constructions, its network topology such as Fig. 2 institutes
Show.Pre-training is carried out using view data, after pre-training terminates, using CCDBN as deep neural network model, input picture,
Obtain result.Finally, weight and biasing are adjusted with reference to BP back-propagation algorithms, detailed process is as follows：
Step 3.1, to netinit：Random initializtion is carried out to weights, convolution kernel and biasing；
Step 3.2, the pre-training of network：Training sample is imported to the network initialized and carries out pre-training, is improved first
RBM models, introduce conditional Gaussian distribution, and structure is restricted Boltzmann machine (CRBM) based on recruitment factor, then introduces volume
Product operation, builds a CRBM (CCRBM) based on convolution.Next trained using weight uncertainty methods
CCRBM models are to alleviate over-fitting problem.Finally, based on one supplement depth confidence net based on convolution of CCRBM model constructions
(CCDBN).Pre-training is carried out using view data；
Step 3.3, reality output is contrasted with label, obtains error, using CCDBN as neutral net, utilized
Weight uncertainty BP algorithms are finely adjusted, the neural network model trained.
Step 4, the identification module of image：The view data being under the jurisdiction of in CIFAR-10 classifications is obtained, place is normalized
Reason, is then adjusted to 32*32 Pixel Dimensions.It is then enter into the convolutional neural networks trained, finally gives knowledge
Other result.
By above content, the application provides a kind of image recognition algorithm based on convolution depth network and is
System, makes training set and label, then the parameter such as number of plies of planned network, carries out pre-training afterwards according to actual needs first,
Then the adjustment to network weight and biasing is completed using weight uncertainty BP algorithms, recently enters image, pre- place
Neutral net is inputted after reason, the identification to image is completed.The application passes through neural network recognization image, it is to avoid the feature of display
Extract, directly using picture as the input of network, recognition accuracy is higher；And network one it is trained can Reusability, place
Manage efficiency high；Training time is short.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is the schematic diagram of CCRBM models used in this application.
Fig. 2 is deep neural network structural representation used in this application.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present application, the technical scheme in the embodiment of the present application is carried out clear, complete
Site preparation is described, and described embodiment is only some embodiments of the present application, rather than whole embodiments.Based on the application
In embodiment, all other implementation that those of ordinary skill in the art are obtained under the premise of creative work is not paid
Example, belongs to the scope of the application protection.
Embodiment 1
The present embodiment comprises the following steps：
Step 1, picture is pre-processed：
Step 1.1, handwritten numeral image is normalized；
Step 1.2, it is 32*32 sizes the image normalization obtained by step 1.2, and is stored in training set, then
Corresponding tally set, 1,0*1 one digital label of matrix representative are made according to training set.
Step 2, CCDBN depth models are built：
The CCDBN models used in the present embodiment are the neutral nets of a multilayer, by input layer, intermediate layer and output layer
Etc. multilayer composition, every layer is made up of multiple node units.Construction multilayer neural network as shown in Figure 2, each layer is due to being all
A kind of probability graph model, therefore, based on energy function, the activation primitive of each hiding layer unit is all the shape of sigmoid functions
Formula；
Step 3, training convolutional neural networks：
Step 3.1, with different small random numbers (between 0-1) pair can training parameter initialize, to biasing be initialized as 0；
Step 3.2, the activation new probability formula that network carries out pre-training network model can be expressed as follows：
Wherein, h represents to hide layer unit, and v represents visible layer unit, and W represents weight matrix, and b represents biasing, then introduced
Weight uncertainty algorithms, so, the calculating of derivative are changed to following form：
According to above-mentioned formula, CCDBN pre-training process is completed.
Step 3.3, residual error is calculated, adjustable parameters and biasing are updated with reference to backpropagation BP algorithm, is completed to CCDBN's
Whole training process.
Step 4, the identification module of image：
Step 4.1, the view data being under the jurisdiction of in CIFAR-10 classifications is obtained, is normalized, is then adjusted to
32*32 Pixel Dimensions；
Step 4.2, the picture after being pre-processed, is input in the CCDBN networks trained, waits to be output,
Take output vector maximum to obtain line number for recognition result, that is, complete the identification to real value image.
Claims (8)
1. a kind of image recognition algorithm and system based on convolution depth network, it is characterised in that by building with label
Picture (CIFAR-10 data sets) is trained as sample set to neutral net, and the neutral net trained is preserved, will be treated
Recognize that picture, as input, result is identified according to output vector.
2. according to the method described in claim 1, it is characterized in that, described deep neural network for multilayer convolution depth nerve
Network, including input layer, 3 convolutional layers, 3 pond layers and 1 output layer, wherein：Input layer is the pixel of two dimensional image,
First convolutional layer has 24 convolution characteristic patterns, and second convolutional layer has 48 convolution characteristic patterns, and output layer is set to 10 sections
Point.
3. according to the method described in claim 1, it is characterized in that, signified training refers to：By sample set (including image and mark
Label) neutral net set is input to, pre-training is carried out using restricted Boltzmann machine (RBM) innovatory algorithm, then
To obtained models coupling backpropagation BP algorithm adjusting parameter and biasing, the complete training process to neutral net is completed.
4. the method according to claim 1 or 3, it is characterized in that, described training includes：
4.1st, first to netinit；Random initializtion is carried out to convolution kernel, full articulamentum weights and biasing；
The 4.2nd, training sample and tally set are imported to the network initialized and carries out pre-training；RBM models are improved first, introduce bar
Part Gaussian Profile, structure is restricted Boltzmann machine (CRBM) based on recruitment factor, then introduces convolution operation, builds one
CRBM (CCRBM) based on convolution.Next CCRBM models are trained to alleviate using weight uncertainty methods
Fitting problems.Finally, based on one supplement depth confidence net (CCDBN) based on convolution of CCRBM model constructions.Utilize picture number
According to pre-training is carried out, after pre-training terminates, using CCDBN as deep neural network model, input picture obtains result.
4.3rd, reality output and label are contrasted, obtains error, using CCDBN as neutral net, utilize weight
Uncertainty BP algorithms are finely adjusted, the neural network model trained.
5. method according to claim 4, it is characterized in that, described training sample includes：Input two dimensional image and label.
Process is divided into pre-training process and global weights trim process.Pre-training process is successively carried out, and is unsupervised learning；Trained
Cheng Zhong, input picture is by successively converting, in output layer output, obtains reality output vector.
6. according to the method described in claim 1, it is characterized in that, described identification includes：Picture to be identified input is trained
CCDBN models in, output vector is obtained to recognize classification belonging to picture.
7. the method according to claim 1 or 6, it is characterized in that, described identification includes：
7.1st, obtain correlation type input image data, picture is pre-processed, normalization and adjust Pixel Dimensions (as
Element is the most suitable for 32*32)；
7.2nd, the picture (being derived from the classification that CIFAR-10 data sets are included) after being pre-processed, is input to and has trained
CCDBN neutral nets in, wait to be output, take output vector maximum line number be recognition result, that is, complete the reality to input
It is worth the identification of image.
8. a kind of system for realizing any of the above-described claim methods described, it is characterised in that：CCDBN neural network modules and from
The identification module of right image, wherein CCDBN is trained to one by CCDBN mixed-media network modules mixed-medias can recognize the classification of limited natural image
Device (class that the image to be recognized belongs in CIFAR-10 data set sample labels), picture recognition module obtains image and inputted
The CCDBN networks that train are identified.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710144957.7A CN107145893A (en) | 2017-03-13 | 2017-03-13 | A kind of image recognition algorithm and system based on convolution depth network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710144957.7A CN107145893A (en) | 2017-03-13 | 2017-03-13 | A kind of image recognition algorithm and system based on convolution depth network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107145893A true CN107145893A (en) | 2017-09-08 |
Family
ID=59783786
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710144957.7A Pending CN107145893A (en) | 2017-03-13 | 2017-03-13 | A kind of image recognition algorithm and system based on convolution depth network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107145893A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107633296A (en) * | 2017-10-16 | 2018-01-26 | 中国电子科技集团公司第五十四研究所 | A kind of convolutional neural networks construction method |
| CN108319456A (en) * | 2018-01-29 | 2018-07-24 | 徐磊 | A kind of development approach for exempting to program deep learning application |
| CN108549899A (en) * | 2018-03-07 | 2018-09-18 | 中国银联股份有限公司 | A kind of image-recognizing method and device |
| CN108918527A (en) * | 2018-05-15 | 2018-11-30 | 佛山市南海区广工大数控装备协同创新研究院 | A kind of printed matter defect inspection method based on deep learning |
| CN108985268A (en) * | 2018-08-16 | 2018-12-11 | 厦门大学 | Conclusion formula high Resolution Range Profile Identification of Radar method based on depth migration study |
| CN108985366A (en) * | 2018-07-06 | 2018-12-11 | 武汉兰丁医学高科技有限公司 | B ultrasound image recognition algorithm and system based on convolution depth network |
| CN109658399A (en) * | 2018-12-13 | 2019-04-19 | 深圳先进技术研究院 | A kind of neck patch image-recognizing method and device |
| CN110619342A (en) * | 2018-06-20 | 2019-12-27 | 鲁东大学 | Rotary machine fault diagnosis method based on deep migration learning |
| CN111291634A (en) * | 2020-01-17 | 2020-06-16 | 西北工业大学 | Unmanned aerial vehicle image target detection method based on convolution limited Boltzmann machine |
| CN111429215A (en) * | 2020-03-18 | 2020-07-17 | 北京互金新融科技有限公司 | Data processing method and device |
| CN111492406A (en) * | 2017-10-11 | 2020-08-04 | 通用电气公司 | Image generation using machine learning |
| TWI706334B (en) * | 2018-12-13 | 2020-10-01 | 鴻海精密工業股份有限公司 | Storage device, electronic device and method for classifying images |
| CN111741491A (en) * | 2020-08-12 | 2020-10-02 | 绿漫科技有限公司 | Automatic statistical method for Bluetooth connection success rate |
| US11055578B2 (en) | 2018-12-13 | 2021-07-06 | Hon Hai Precision Industry Co., Ltd. | Method for classifying images and electronic device employing the method |
-
2017
- 2017-03-13 CN CN201710144957.7A patent/CN107145893A/en active Pending
Non-Patent Citations (6)
| Title |
|---|
| HONGLAK LEE ET AL.: "Unsupervised feature learning for audio classification using convolutional deep belief networks", 《23RD ANNUAL CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS》 * |
| JIAN ZHANG ET AL.: "Incremental extreme learning machine based on deep feature embedded", 《INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS》 * |
| JIAN ZHANG ET AL.: "Weight Uncertainty in Boltzmann Machine", 《COGNITIVE COMPUTATION》 * |
| NAN ZHANG ET AL.: "Multi layer ELM-RBF for multi-label learning", 《APPLIED SOFT COMPUTING》 * |
| SHIFEI DING ET AL.: "Boltzmann Machine and its Applications in Image Recognition", 《INTERNATIONAL FEDERATION FOR INFORMATION PROCESSING 2016》 * |
| 陈雯柏: "《人工神经网络原理与实践》", 31 January 2016 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111492406A (en) * | 2017-10-11 | 2020-08-04 | 通用电气公司 | Image generation using machine learning |
| CN111492406B (en) * | 2017-10-11 | 2023-12-22 | 通用电气公司 | Method for training machine learning algorithm, image processing system and image reconstruction method |
| CN107633296B (en) * | 2017-10-16 | 2019-10-22 | 中国电子科技集团公司第五十四研究所 | A kind of convolutional neural networks construction method |
| CN107633296A (en) * | 2017-10-16 | 2018-01-26 | 中国电子科技集团公司第五十四研究所 | A kind of convolutional neural networks construction method |
| CN108319456A (en) * | 2018-01-29 | 2018-07-24 | 徐磊 | A kind of development approach for exempting to program deep learning application |
| CN108319456B (en) * | 2018-01-29 | 2021-03-09 | 徐磊 | Development method of programming-free deep learning application |
| CN108549899A (en) * | 2018-03-07 | 2018-09-18 | 中国银联股份有限公司 | A kind of image-recognizing method and device |
| CN108918527A (en) * | 2018-05-15 | 2018-11-30 | 佛山市南海区广工大数控装备协同创新研究院 | A kind of printed matter defect inspection method based on deep learning |
| CN110619342B (en) * | 2018-06-20 | 2023-02-03 | 鲁东大学 | Rotary machine fault diagnosis method based on deep migration learning |
| CN110619342A (en) * | 2018-06-20 | 2019-12-27 | 鲁东大学 | Rotary machine fault diagnosis method based on deep migration learning |
| CN108985366A (en) * | 2018-07-06 | 2018-12-11 | 武汉兰丁医学高科技有限公司 | B ultrasound image recognition algorithm and system based on convolution depth network |
| CN108985268A (en) * | 2018-08-16 | 2018-12-11 | 厦门大学 | Conclusion formula high Resolution Range Profile Identification of Radar method based on depth migration study |
| CN108985268B (en) * | 2018-08-16 | 2021-10-29 | 厦门大学 | Inductive radar high-resolution range profile identification method based on deep migration learning |
| TWI706334B (en) * | 2018-12-13 | 2020-10-01 | 鴻海精密工業股份有限公司 | Storage device, electronic device and method for classifying images |
| CN109658399A (en) * | 2018-12-13 | 2019-04-19 | 深圳先进技术研究院 | A kind of neck patch image-recognizing method and device |
| US11055578B2 (en) | 2018-12-13 | 2021-07-06 | Hon Hai Precision Industry Co., Ltd. | Method for classifying images and electronic device employing the method |
| CN111291634A (en) * | 2020-01-17 | 2020-06-16 | 西北工业大学 | Unmanned aerial vehicle image target detection method based on convolution limited Boltzmann machine |
| CN111429215A (en) * | 2020-03-18 | 2020-07-17 | 北京互金新融科技有限公司 | Data processing method and device |
| CN111429215B (en) * | 2020-03-18 | 2023-10-31 | 北京互金新融科技有限公司 | Data processing method and device |
| CN111741491B (en) * | 2020-08-12 | 2021-04-27 | 绿漫科技有限公司 | Automatic statistical method for Bluetooth connection success rate |
| CN111741491A (en) * | 2020-08-12 | 2020-10-02 | 绿漫科技有限公司 | Automatic statistical method for Bluetooth connection success rate |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107145893A (en) | A kind of image recognition algorithm and system based on convolution depth network | |
| CN107729872A (en) | Facial expression recognition method and device based on deep learning | |
| CN107153810A (en) | A kind of Handwritten Numeral Recognition Method and system based on deep learning | |
| CN107506722A (en) | One kind is based on depth sparse convolution neutral net face emotion identification method | |
| CN104537393B (en) | A kind of traffic sign recognition method based on multiresolution convolutional neural networks | |
| Teow | Understanding convolutional neural networks using a minimal model for handwritten digit recognition | |
| JP6788264B2 (en) | Facial expression recognition method, facial expression recognition device, computer program and advertisement management system | |
| CN110399821B (en) | Customer satisfaction acquisition method based on facial expression recognition | |
| CN106803069A (en) | Crowd's level of happiness recognition methods based on deep learning | |
| CN112801040B (en) | Lightweight unconstrained facial expression recognition method and system embedded with high-order information | |
| CN110378208B (en) | Behavior identification method based on deep residual error network | |
| CN106326899A (en) | Tobacco leaf grading method based on hyperspectral image and deep learning algorithm | |
| CN107657204A (en) | The construction method and facial expression recognizing method and system of deep layer network model | |
| CN112784763A (en) | Expression recognition method and system based on local and overall feature adaptive fusion | |
| CN107633293A (en) | A kind of domain-adaptive method and device | |
| CN107766934A (en) | A kind of depth convolutional neural networks method based on convolution unit arranged side by side | |
| Xu et al. | Face expression recognition based on convolutional neural network | |
| CN113627376A (en) | Facial expression recognition method based on multi-scale dense connection depth separable network | |
| CN112597873A (en) | Dual-channel facial expression recognition method based on deep learning | |
| Danukusumo et al. | Indonesia ancient temple classification using convolutional neural network | |
| Gao et al. | Natural scene recognition based on convolutional neural networks and deep Boltzmannn machines | |
| CN109508640A (en) | A kind of crowd's sentiment analysis method, apparatus and storage medium | |
| Teow | A minimal convolutional neural network for handwritten digit recognition | |
| He et al. | Global and local fusion ensemble network for facial expression recognition | |
| CN112800979B (en) | Dynamic expression recognition method and system based on characterization flow embedded network |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170908 |