CN114758189B - Method and system for detecting antagonistic sample based on gradient thermodynamic diagram and key area - Google Patents

Method and system for detecting antagonistic sample based on gradient thermodynamic diagram and key area Download PDF

Info

Publication number
CN114758189B
CN114758189B CN202210353949.4A CN202210353949A CN114758189B CN 114758189 B CN114758189 B CN 114758189B CN 202210353949 A CN202210353949 A CN 202210353949A CN 114758189 B CN114758189 B CN 114758189B
Authority
CN
China
Prior art keywords
image
hot spot
key area
video
sample
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.)
Active
Application number
CN202210353949.4A
Other languages
Chinese (zh)
Other versions
CN114758189A (en
Inventor
刘小垒
胥迤潇
杨润
辛邦洲
王玉龙
殷明勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
COMPUTER APPLICATION RESEARCH INST CHINA ACADEMY OF ENGINEERING PHYSICS
Original Assignee
COMPUTER APPLICATION RESEARCH INST CHINA ACADEMY OF ENGINEERING PHYSICS
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by COMPUTER APPLICATION RESEARCH INST CHINA ACADEMY OF ENGINEERING PHYSICS filed Critical COMPUTER APPLICATION RESEARCH INST CHINA ACADEMY OF ENGINEERING PHYSICS
Priority to CN202210353949.4A priority Critical patent/CN114758189B/en
Publication of CN114758189A publication Critical patent/CN114758189A/en
Application granted granted Critical
Publication of CN114758189B publication Critical patent/CN114758189B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent

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 method and a system for detecting confrontation samples based on gradient thermodynamic diagrams and key areas, belongs to the technical field of confrontation sample identification, and solves the problems of high false alarm rate and high missing report rate of the existing confrontation sample detection method. The method is used for identifying the key area of the image or the video based on the image key area identification method; aiming at the deep learning model, carrying out hot spot region identification on the image or the video by using a hot spot region detection method; calculating the coincidence degree of the key area and the hot spot area to obtain a coincidence value; and judging whether the coincidence value is greater than a given boundary value, if so, determining that the image or the video is a normal sample, otherwise, determining that the image or the video is a countermeasure sample. The invention is used for resisting sample identification.

Description

Method and system for detecting countermeasure sample based on gradient thermodynamic diagram and key area
Technical Field
A method and a system for detecting a confrontation sample based on a gradient thermodynamic diagram and a key area are used for confronting sample identification and belong to the technical field of confronting sample identification.
Background
Existing research shows that a deep learning model based on a deep neural network is easy to resist sample attack. The general definition of an attack against a sample is: an attacker adds tiny disturbances invisible to the human eye to input samples of the deep learning model, so that the deep learning model outputs wrong results or makes wrong responses with high probability. The application safety of the deep learning model is seriously threatened by the resisting sample, and particularly when the deep learning model is applied to safety key scenes, such as the fields of automatic driving, malicious flow analysis and the like. Therefore, the robustness of the deep learning model to the resisting sample is enhanced, and the basis of further wide application of the deep learning model is provided.
The existing mainstream confrontation sample defense method is confrontation training, namely, the confrontation sample and a normal sample are added into a deep learning model at the same time for training so as to improve the robustness of the deep learning model to the confrontation sample. However, the prior art of adopting the confrontation training to detect the confrontation sample has the following technical problems:
1. the false alarm rate and the missing report rate of the existing countermeasure sample detection method are high;
2. the defense method for carrying out the countermeasure sample based on the existing countermeasure training needs to introduce new countermeasure sample data, so that the difficulty of data collection is increased;
3. adopting the countertraining can reduce the performance of the deep learning model on a normal sample data set, for example, for an image classification model, the performance refers to the accuracy of image classification, and for an image generation model, the performance refers to the quality of image generation;
4. the prior art cannot identify and early warn the anti-sample attack event.
Disclosure of Invention
In view of the above research problems, an object of the present invention is to provide a method and a system for detecting a challenge sample based on a gradient thermodynamic diagram and a key region, so as to solve the problem of high false alarm rate and high false negative rate of the conventional method for detecting a challenge sample.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method for detecting challenge samples based on gradient thermodynamic diagrams and key regions, comprising the steps of:
step 1, identifying key areas of an image or a video based on an image key area identification method;
step 2, aiming at the deep learning model, carrying out hot spot area identification on the image or the video by using a hot spot area detection method;
step 3, calculating the coincidence degree of the key area and the hot spot area to obtain a coincidence degree value;
and 4, judging whether the coincidence value is larger than a given boundary value or not, if so, judging the image or the video to be a normal sample, otherwise, judging the image or the video to be a countermeasure sample.
Further, the image key area identification method in the step 1 is an AC algorithm, an HC algorithm, an LC algorithm or an FT algorithm.
Further, the formula for identifying the key area in step 1 is as follows:
s＝Salient(x)
where s represents a key region, x represents an image or video, and yield (.) represents a computer vision method.
Further, the hot spot region detection method in step 2 is a gradient thermodynamic diagram method G-CAM.
Further, the specific steps of step 2 are:
inputting an image or video x into a deep learning model C to obtain an input/response y of the deep learning model:
y＝C(x)
identifying a hot spot region h of an image or video x and an output/response y in the deep learning model C by using a hot spot region detection method:
h＝Gradmap(C，x，y)
wherein, gradmap (.) represents a hot spot region detection method.
Further, the calculation formula of the coincidence value in step 3 is as follows:
value＝SUM(s·h)
where value represents the coincidence value, and-represents the dot product of the vectors, i.e., the multiplication of the values of the corresponding positions of s and h.
A system for detecting challenge samples based on gradient thermodynamic diagrams and critical areas, comprising:
a key area identification unit: performing key area identification on the image or the video based on an image key area identification method;
a hot spot area identification unit: identifying the hot spot area of the image or the video by using a hot spot area detection method;
a contact ratio calculation unit: calculating the coincidence degree of the key area and the hot spot area to obtain a coincidence value;
a confrontation sample identification unit: and judging whether the coincidence value is greater than a given boundary value, if so, determining that the image or the video is a normal sample, otherwise, determining that the image or the video is a countermeasure sample.
Further, the image key area identification method in the key area identification unit is an AC algorithm, an HC algorithm, an LC algorithm or an FT algorithm.
Further, the hot spot region detection method in the hot spot region identification unit is a gradient thermodynamic diagram method G-CAM;
the hot spot area identification unit comprises the following specific steps:
inputting an image or video x into a deep learning model C to obtain an input/response y of the deep learning model:
y＝C(x)
identifying a hot spot region h of an image or video x and an output/response y in the deep learning model C by using a hot spot region detection method:
h＝Gradmap(C，x，y)
wherein Gradmap (.) represents a hot spot region detection method.
Further, the calculation formula of the coincidence degree value in the coincidence degree calculation unit is as follows:
value＝SUM(s·h)
where value represents the coincidence value,. Represents the dot product of the vector, i.e. the value of the corresponding position of s and h.
Compared with the prior art, the invention has the beneficial effects that:
1. compared with the existing method for detecting the confrontation sample, the confrontation sample obtained based on the gradient thermodynamic diagram and the key area detection reduces the false alarm rate and the missing report rate by more than 10%;
2. the invention does not need to introduce new countermeasure sample data for countermeasure training, thereby reducing the difficulty of data collection;
3. the performance of the confrontation sample obtained based on the gradient thermodynamic diagram and the key region detection is higher than that of the confrontation sample obtained on a normal sample data set through a deep learning model;
4. in the case of the countermeasure training, part of the countermeasure samples can be correctly classified by the countermeasure training, but the samples cannot be judged to be the countermeasure samples, so that the attack events cannot be identified and early warned.
Drawings
FIG. 1 is a schematic diagram of the framework of the present invention;
FIG. 2 is a schematic diagram of thermodynamic diagrams and key areas of a normal sample and a challenge sample in the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings and the detailed description.
The invention provides a confrontation sample detection method based on gradient thermodynamic diagram and key area identification. The method is based on the general definition of the confrontation sample, namely, the invisibility of the confrontation disturbance to human eyes, and provides a key area identification technology based on a traditional computer vision method, wherein the key areas identified on the normal sample and the corresponding confrontation sample are the same, and the hot spot areas in the deep learning model are different. Meanwhile, the key area of the normal sample and the hot spot area have the coincident characteristic, and the anti-sample is detected.
The key area coincidence rate of the confrontation sample and the corresponding normal sample identified by the key area by using the traditional computer vision method can reach more than 99 percent.
The hot spot regions acting on the decision in the deep learning model are different for the confrontation sample and the corresponding normal sample, and the decision hot spot region of the normal sample in the deep learning model is usually included in the key region.
The contact ratio of the hot spot area of the countermeasure sample in the deep learning model and the key area identified by the image key area identification method in the computer vision method is lower than that of the normal sample.
Therefore, the detection of the confrontation sample can be carried out by calculating the coincidence degree of the image or video decision hot spot region and the key region.
A method for detecting challenge samples based on gradient thermodynamic diagrams and key regions, comprising the steps of:
step 1, identifying key areas of an image or a video based on an image key area identification method; the image key area identification method comprises an AC algorithm, an HC algorithm, an LC algorithm or an FT algorithm and the like, wherein the LC algorithm adopts the sum of Euclidean distances between a single pixel point and a global pixel point as the weight of the pixel point, and a significant area (key area) of an image is obtained after normalization.
The formula for key area identification is:
s＝Salient(x)
where s represents a key region, x represents an image or video, and yield (.) represents a computer vision method.
Step 2, aiming at a deep learning model (the deep learning model is an attacked target model), carrying out hot spot region identification on an input image by using a hot spot region detection method; the hot spot region detection method is a gradient thermodynamic diagram method G-CAM, and the gradient thermodynamic diagram method G-CAM evaluates the influence of a single pixel point in an image or video x on a decision result of a final model (a deep learning model) through gradient forward and backward propagation.
The method comprises the following specific steps:
inputting an image or video x into a deep learning model C to obtain an input/response y of the deep learning model:
y＝C(x)
identifying a hot spot region h of an image or video x and an output/response y in the deep learning model C by using a hot spot region detection method:
h＝Gradmap(C，x，y)
wherein, gradmap (.) represents a hot spot region detection method. Gradmap (C, x, y) represents that the G-CAM method is used for identifying which parts of the image or video x have large contribution to the classification result in the process of classifying the image or video x by using the deep learning model C, namely identifying the hot spot region.
Step 3, calculating the coincidence degree of the key area and the hot spot area to obtain a coincidence degree value; the formula for calculating the coincidence value is as follows:
value＝SUM(s·h)
where value represents the coincidence value,. Represents the dot product of the vector, i.e. the value of the corresponding position of s and h.
The deep learning model C may be an image classification model, a video classification model, or a target detection model, etc.
And 4, judging whether the coincidence value is larger than a given boundary value or not, if so, judging as a normal sample, and otherwise, judging as a countermeasure sample.
Examples
Given an image/video classification model F, a single image/video sample x is to be determined whether x is a challenge sample. Firstly, identifying a key area of x by using an image key area identification method LC:
s＝Salient(x)
and identifying and classifying hot spot regions (namely hot spot regions) based on the gradient thermodynamic diagram:
h＝Gradmap(F，x，y)
and finally, calculating the contact ratio of the key area and the classified hot spot area:
value＝SUM(s·h)
when the coincidence value is smaller than the set boundary value, x can be judged as a confrontation sample, otherwise, x is judged as a normal sample.
As shown in fig. 2 (from the Imagenet data set), the first graph is the thermodynamic diagram of the normal sample, the second graph is the thermodynamic diagram of the confrontation sample, the third graph is the key area of the normal sample, and the fourth graph is the key area of the confrontation sample, and the thermodynamic diagram is overlapped with the key area to calculate the repetition value, so that the normal sample and the confrontation sample can be identified, namely, the overlap ratio of the white parts of the first graph and the third graph, and the second graph and the fourth graph is calculated. (since the first and second diagrams cannot be represented in color, the red and green portions are adjusted to be displayed in white, and then the coincidence value is calculated for each of the third and fourth diagrams).
The above are merely representative examples of the many specific applications of the present invention, and do not limit the scope of the invention in any way. All the technical solutions formed by the transformation or the equivalent substitution fall within the protection scope of the present invention.

Claims (10)

1. A method for detecting challenge samples based on gradient thermodynamic diagrams and key regions, comprising the steps of:
step 1, giving an image/video classification model
Figure 349031DEST_PATH_IMAGE001
Single picture/video sample>
Figure 864326DEST_PATH_IMAGE002
Is to be judged and/or judged>
Figure 177496DEST_PATH_IMAGE002
Whether the image is a countermeasure sample or not is judged, key area identification is carried out on the image or the video on the basis of an image key area identification method, the image key area identification method is an LC algorithm, the LC algorithm adopts the sum of Euclidean distances between a single pixel point and a global pixel point as the weight of the pixel point, and the key area of the image is obtained after normalization;
step 2, aiming at the deep learning model, carrying out hot spot area identification on the image or the video by using a hot spot area detection method;
step 3, calculating the coincidence degree of the key area and the hot spot area to obtain a coincidence degree value;
and 4, judging whether the coincidence value is larger than a given boundary value, if so, judging the image or the video as a normal sample, otherwise, judging the image or the video as a confrontation sample.
2. The method for detecting the challenge sample based on the gradient thermodynamic diagram and the key area according to claim 1, wherein the image key area identification method in the step 1 is replaced by an LC algorithm, an HC algorithm or an FT algorithm.
3. The method for detecting the challenge sample based on the gradient thermodynamic diagram and the key area as claimed in claim 2, wherein the formula for identifying the key area in step 1 is as follows:
Figure DEST_PATH_IMAGE003
wherein the content of the first and second substances,
Figure 940501DEST_PATH_IMAGE004
indicates a key region, <' > or>
Figure 73542DEST_PATH_IMAGE002
Represents an image or video, is>
Figure 221889DEST_PATH_IMAGE005
Representing a computer vision method.
4. The method of claim 3, wherein the hot spot region detection method in step 2 is a gradient thermodynamic diagram method G-CAM.
5. The method for detecting the challenge sample based on the gradient thermodynamic diagram and the key area as claimed in claim 4, wherein the specific steps of the step 2 are as follows:
rendering images or videosxInput deep learning modelCObtaining input/responses for deep learning models
Figure 123986DEST_PATH_IMAGE006
：
Figure 666963DEST_PATH_IMAGE007
Deep learning model identification by hot spot region detection methodCFor images or videosxAnd heat output/responseDot areah：
Figure 224983DEST_PATH_IMAGE008
Wherein, the first and the second end of the pipe are connected with each other,
Figure 409977DEST_PATH_IMAGE009
indicating a hot spot area detection method.
6. The method for detecting the challenge sample based on the gradient thermodynamic diagram and the critical area according to claim 3, wherein the coincidence value in step 3 is calculated by the formula:
Figure 868378DEST_PATH_IMAGE010
wherein, the first and the second end of the pipe are connected with each other,
Figure 582256DEST_PATH_IMAGE011
represents a coincidence value>
Figure 955468DEST_PATH_IMAGE012
Representing point multiplication of vectors, i.e. < - >>
Figure 412995DEST_PATH_IMAGE004
And &>
Figure 492946DEST_PATH_IMAGE013
The values of the corresponding positions are multiplied.
7. A system for detecting challenge samples based on gradient thermodynamic diagrams and critical areas, comprising:
a key area identification unit: given an image/video classification model
Figure 348032DEST_PATH_IMAGE001
Single picture/video sample>
Figure 942961DEST_PATH_IMAGE002
Is to be judged and/or judged>
Figure 407441DEST_PATH_IMAGE002
Whether the image is a countermeasure sample or not is judged, key area identification is carried out on the image or the video on the basis of an image key area identification method, the image key area identification method is an LC algorithm, the LC algorithm adopts the sum of Euclidean distances between a single pixel point and a global pixel point as the weight of the pixel point, and the key area of the image is obtained after normalization;
a hot spot area identification unit: identifying the hot spot area of the image or the video by using a hot spot area detection method;
a contact ratio calculation unit: calculating the coincidence degree of the key area and the hot spot area to obtain a coincidence value;
a confrontation sample identification unit: and judging whether the coincidence value is greater than a given boundary value, if so, determining that the image or the video is a normal sample, otherwise, determining that the image or the video is a countermeasure sample.
8. The system for detecting the antagonistic sample based on the gradient thermodynamic diagram and the key area as claimed in claim 7, wherein the image key area identification method in the key area identification unit is replaced by an LC algorithm, an HC algorithm, an LC algorithm or an FT algorithm.
9. The system for detecting the challenge sample based on the gradient thermodynamic diagram and the key area according to claim 8, wherein the hot spot area detection method in the hot spot area identification unit is a gradient thermodynamic diagram method G-CAM;
the hot spot area identification unit comprises the following specific steps:
rendering images or videosxInput deep learning modelCObtaining input/responses for deep learning models
Figure 138636DEST_PATH_IMAGE006
：
Figure 663159DEST_PATH_IMAGE007
Deep learning model identification by hot spot region detection methodCFor images or videosxAnd hot spot area of output/responseh：
Figure 948646DEST_PATH_IMAGE008
Wherein the content of the first and second substances,
Figure 252370DEST_PATH_IMAGE009
indicating a hot spot area detection method.
10. The system for detecting the confrontation sample based on the gradient thermodynamic diagram and the key area as claimed in claim 9, wherein the coincidence value in the coincidence degree calculating unit is calculated by the formula:
Figure 306913DEST_PATH_IMAGE010
wherein the content of the first and second substances,
Figure 64654DEST_PATH_IMAGE011
represents a coincidence value>
Figure 103017DEST_PATH_IMAGE012
Representing a dot multiplication of a vector, i.e. < >>
Figure 909299DEST_PATH_IMAGE004
And &>
Figure 585393DEST_PATH_IMAGE013
Corresponding bitThe values of the bits are multiplied. />
CN202210353949.4A 2022-04-01 2022-04-01 Method and system for detecting antagonistic sample based on gradient thermodynamic diagram and key area Active CN114758189B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210353949.4A CN114758189B (en) 2022-04-01 2022-04-01 Method and system for detecting antagonistic sample based on gradient thermodynamic diagram and key area

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210353949.4A CN114758189B (en) 2022-04-01 2022-04-01 Method and system for detecting antagonistic sample based on gradient thermodynamic diagram and key area

Publications (2)

Publication Number Publication Date
CN114758189A CN114758189A (en) 2022-07-15
CN114758189B true CN114758189B (en) 2023-04-07

Family

ID=82328929

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210353949.4A Active CN114758189B (en) 2022-04-01 2022-04-01 Method and system for detecting antagonistic sample based on gradient thermodynamic diagram and key area

Country Status (1)

Country Link
CN (1) CN114758189B (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990357A (en) * 2021-04-16 2021-06-18 中国工程物理研究院计算机应用研究所 Black box video countermeasure sample generation method based on sparse disturbance

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11373093B2 (en) * 2019-06-26 2022-06-28 International Business Machines Corporation Detecting and purifying adversarial inputs in deep learning computing systems

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990357A (en) * 2021-04-16 2021-06-18 中国工程物理研究院计算机应用研究所 Black box video countermeasure sample generation method based on sparse disturbance

Also Published As

Publication number Publication date
CN114758189A (en) 2022-07-15

Similar Documents

Publication Publication Date Title
CN111627044B (en) Target tracking attack and defense method based on deep network
US7574039B2 (en) Video based fire detection system
CN109543760B (en) Confrontation sample detection method based on image filter algorithm
CN107506734A (en) One kind of groups unexpected abnormality event detection and localization method
CN113011367A (en) Abnormal behavior analysis method based on target track
CN102760230B (en) Flame detection method based on multi-dimensional time domain characteristics
Chen et al. Research on image fire detection based on support vector machine
Fan et al. Lightweight forest fire detection based on deep learning
CN112668557A (en) Method for defending image noise attack in pedestrian re-identification system
CN113127857A (en) Deep learning model defense method for adversarial attack and deep learning model
CN113011322A (en) Detection model training method and detection method for specific abnormal behaviors of monitoring video
Li et al. Detecting adversarial patch attacks through global-local consistency
CN114170686A (en) Elbow bending behavior detection method based on human body key points
CN114758189B (en) Method and system for detecting antagonistic sample based on gradient thermodynamic diagram and key area
Long et al. Transmission: a new feature for computer vision based smoke detection
CN113240028A (en) Anti-sample block attack detection method based on class activation graph
CN117152486A (en) Image countermeasure sample detection method based on interpretability
Deldjoo et al. A novel fuzzy-based smoke detection system using dynamic and static smoke features
Chi et al. Public-attention-based adversarial attack on traffic sign recognition
Yin et al. An improved algorithm based on convolutional neural network for smoke detection
CN108710886A (en) A kind of multiimage matching process based on SIFT algorithms
CN114694090A (en) Campus abnormal behavior detection method based on improved PBAS algorithm and YOLOv5
CN112052912A (en) Intelligent flame combustion state identification method for fire-fighting robot
CN108416768B (en) Binary-based foreground image similarity evaluation method
CN114663730B (en) Game interaction-based confrontation sample detection method and system

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
GR01 Patent grant
GR01 Patent grant