CN115170813A - Network supervision fine-grained image identification method based on partial label learning - Google Patents

Network supervision fine-grained image identification method based on partial label learning Download PDF

Info

Publication number
CN115170813A
CN115170813A CN202210761418.9A CN202210761418A CN115170813A CN 115170813 A CN115170813 A CN 115170813A CN 202210761418 A CN202210761418 A CN 202210761418A CN 115170813 A CN115170813 A CN 115170813A
Authority
CN
China
Prior art keywords
label
noise
sample
image
correlation
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
Application number
CN202210761418.9A
Other languages
Chinese (zh)
Inventor
魏秀参
许玉燕
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.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
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 Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202210761418.9A priority Critical patent/CN115170813A/en
Publication of CN115170813A publication Critical patent/CN115170813A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a network supervision fine-grained image identification method based on partial label learning, which comprises the following steps: carrying out depth descriptor transformation by using a depth neural network model with pre-training to evaluate positive correlation between network images, detecting open-set label noise existing in a data set according to a correlation matrix, and removing the open-set label noise; the loss function is used for driving the model to show higher recall rate, so that the label set of each sample contains the labels of the real classes of the samples as much as possible; and selecting real labels of the samples from the label sets of the samples, correcting the label noise of the closed set, and putting the clean data and the closed set label noise with the corrected labels into a deep neural network for training. The method effectively removes the open set label noise, and simultaneously converts the closed set label noise into the training image with the accurate label by using the partial label learning, so that the number of available samples in the network data set is increased, and the learning performance of the neural network model is further improved.

Description

Network supervision fine-grained image identification method based on partial label learning
Technical Field
The invention belongs to the field of network supervision image recognition, and particularly relates to a network supervision fine-grained image recognition method based on partial label learning.
Background
Constructing a fine-grained dataset requires a specific domain expert to correctly classify through subtle differences between fine-grained subclasses, and is therefore a difficult task. To reduce the reliance on manual labeling for building fine-grained datasets and to learn more practical models, it is becoming increasingly popular to build and train network datasets directly from images of relevant classes collected from the internet. However, the constructed network data set has more data noise, and direct training can cause overfitting of the model, thereby affecting accuracy. There are generally two types of noise in a fine-grained network data set, namely, open-set label noise and closed-set label noise. Open-set label noise is typically caused by "cross-domain", i.e., noise does not belong to any class in the same fine-grained domain. Closed-set noise refers to images with false labels in a fine-grained domain.
The methods for processing general label noise include sample selection, soft label application or related loss functions, and although the methods have good classification effect, the methods have the risks of 1) discarding a part of clean images, and 2) still have the problems that closed set label noise images cannot be converted into accurate training images in the aspect of noise image utilization.
Disclosure of Invention
The invention aims to provide a network supervision fine-grained image identification method based on partial label learning.
The technical scheme for realizing the purpose of the invention is as follows: in a first aspect, the invention provides a network supervision fine-grained image identification method based on partial label learning, which comprises the following steps:
step 1, carrying out depth descriptor transformation by using a depth neural network model with pre-training to evaluate positive correlation between network images, detecting open-set label noise existing in a data set according to a correlation matrix reflecting the correlation, and removing the open-set label noise;
step 2, driving the model by using a loss function to enable the label set of each sample to contain the labels of the real types of the samples as much as possible;
and 3, selecting a real label of the sample from the label set of the sample by using the idea of partial label learning, thereby correcting the closed set label noise.
In a second aspect, the invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method of the first aspect when executing the program.
In a third aspect, the invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method of the first aspect.
In a fourth aspect, the invention provides a computer program product comprising a computer program which, when executed by a processor, performs the steps of the method of the first aspect.
Compared with the prior art, the invention has the following remarkable advantages: (1) The invention provides an open set label noise removal strategy and a closed set label noise correction strategy to process a practical but challenging network supervision fine-grained identification task; (2) Carrying out depth descriptor transformation by using a depth model trained in advance to estimate positive correlation between network images, and effectively detecting and removing open-set label noise according to a correlation value; (3) The candidate label set of each sample is automatically constructed, a loss function driving model is utilized to show higher recall rate, so that the sample real labels exist in the candidate label set of each sample as far as possible, and the performance of biased label learning is ensured; (4) The closed set label noise is converted into a training image with an accurate label by utilizing partial label learning, the conversion from the closed set label noise to the training data is generated while the closed set label noise is corrected, the number of available samples in a network data set is increased, and the improvement of the learning performance of a neural network model is ensured.
Drawings
Fig. 1 is a flowchart of a network supervision fine-grained image identification method based on partial label learning according to the present invention.
Detailed Description
With reference to fig. 1, a network supervision fine-grained image identification method based on partial label learning specifically includes the following steps:
step 1, performing depth descriptor transformation by using a depth neural network model with pre-training to evaluate positive correlation between network images, detecting open set tag noise existing in a data set according to a correlation matrix reflecting the correlation, and removing the open set tag noise;
is provided with
Figure BDA0003724146630000021
In order to be a space for the label,
Figure BDA0003724146630000022
for sample space, for each random batch containing n pictures
Figure BDA0003724146630000023
Figure BDA0003724146630000024
By having a pretrained convolutional neural network Φ pre Extracting a feature map set containing n feature maps
Figure BDA0003724146630000025
Figure BDA0003724146630000026
Wherein H, W and d respectively represent a characteristic diagram t i Height, width and depth. To be gathered in a feature map
Figure BDA0003724146630000027
Finding more universal high-grade feature expression, and generating feature vector with maximum feature value by using principal component analysis
Figure BDA0003724146630000028
For a given feature map set, each feature map is subjected to channel weighted summation with the feature vector p to obtain a heat map and form the heat map set
Figure BDA0003724146630000031
Calculating the ith feature map t i Corresponding to
Figure BDA0003724146630000032
The ith heatmap H in (II) i ：
Figure BDA0003724146630000033
Wherein
Figure BDA0003724146630000034
Each heat map is then up-sampled to the input image size to obtain a correlation matrix C. The correlation matrix is composed of correlation values, positive values representing the sum
Figure BDA0003724146630000035
The existing universal expression is in positive correlation with similarity, negative value represents negative correlation, and the larger the absolute value is, the stronger the correlation is. According to the quantity and the size of positive values in the correlation matrix, whether the samples are open set label noise can be effectively judged, therefore, a threshold value delta is set to judge each sample, and whether the ith sample is noise is judged:
Figure BDA0003724146630000036
if the sample does not satisfy the condition, it is regarded as open set label noise and removed from the sample space, so that a sample space composed of clean data and closed set label noise can be obtained
Figure BDA0003724146630000037
Step 2, a loss function driving model is used for showing a high recall rate, so that the label set of each sample contains the labels of the real types of the samples as much as possible;
the invention defines a label space as
Figure BDA0003724146630000038
The sample space is
Figure BDA0003724146630000039
Wherein
Figure BDA00037241466300000310
Indicating that the label belongs to class i y i A set of instances of (c). In the sample selection stage at the beginning of training, C categories are randomly selected to generate a batch
Figure BDA00037241466300000311
For each selected category y i Randomly select n * Samples of the category. On a batch basis
Figure BDA00037241466300000312
a＝n * X C, can be obtained by convolution with a neural network phi CNN Obtaining embedded features
Figure BDA00037241466300000313
For in
Figure BDA00037241466300000314
Sample of (1)
Figure BDA00037241466300000315
In other words, the neural network phi can be convolved with CNN Obtaining an embedding feature f i ：
Figure BDA00037241466300000316
Where c is the embedding feature f i Length of (d). Obtaining a similarity matrix by calculating cosine similarity between embedded features
Figure BDA00037241466300000317
The cosine similarity between the ith query image (query image) and the jth support image (support image) is calculated as follows:
Figure BDA00037241466300000318
similarity s obtained by each query image and other images q,: Arranging the first k pictures with high similarity into a set
Figure BDA00037241466300000319
In (1). The invention defines the query image as belonging to the same category but not in the set
Figure BDA00037241466300000320
The image in (2) is called positive image, in the set
Figure BDA00037241466300000321
An image that is not of the same category as the query image is called a negative image. Is not in the set
Figure BDA00037241466300000322
Positive images in (1) constitute a set
Figure BDA00037241466300000323
Wherein
Figure BDA00037241466300000324
To represent
Figure BDA00037241466300000325
In
Figure BDA00037241466300000326
Complement of, y q Is a label for the query image. The invention is provided with n * <k, a set of negative images can be obtained
Figure BDA0003724146630000041
Wherein
Figure BDA0003724146630000042
Is in the aggregate
Figure BDA0003724146630000043
Number of images outside of the same category as the query image, s n Is a matrix containing only the similarity scores of the negative images. Thus, the loss function is defined as
Figure BDA0003724146630000044
Applying this loss function ensures that each image generates as many tag sets as possible that contain true classes.
And 3, selecting a real label of the sample from the label set of the sample by using the idea of partial label learning, thereby correcting the closed set label noise.
Obtaining as many tag sets containing real categories as possible through step 2
Figure BDA0003724146630000045
Then, a true tag is determined from the closed set tag noise's tag set. In the encoding stage, an encoding matrix M epsilon { +1, -1} is constructed by randomly generating column encoding of N bits N×L Wherein N represents the number of classes, L represents the number of two classifiers, and the coding matrix is used for dividing samples in the training process. A randomly generated column code v = [ ] 1 ,v 2 ,…,v N ] T ∈{+1,-1} N The label space can be divided into positive label spaces
Figure BDA0003724146630000046
And a negative label space
Figure BDA0003724146630000047
Figure BDA0003724146630000048
Figure BDA0003724146630000049
Selecting positive and negative samples by utilizing positive and negative label space, and giving a training sample
Figure BDA00037241466300000410
Wherein
Figure BDA00037241466300000411
The invention visual label set
Figure BDA00037241466300000412
To help build a two classifier for a whole. When label set
Figure BDA00037241466300000413
All fall into
Figure BDA00037241466300000414
Or
Figure BDA00037241466300000415
Time, sample
Figure BDA00037241466300000416
Will be used as a positive or negative sample. These positive and negative samples then form a binary training set
Figure BDA00037241466300000417
At the decoding stage, a connected set is constructed for each class, and the connected set of the j-th class can be represented as:
Figure BDA00037241466300000418
according to a connected set epsilon y Generating a performance matrix G N×L To reflect the capabilities of the classifiers, class j at class t classifier g t The performance above is calculated as follows:
Figure BDA00037241466300000419
wherein
Figure BDA00037241466300000420
Figure BDA00037241466300000421
Is an indicator function, and in order to obtain the relative performance of the classifier on each class, the performance matrix G is normalized line by line:
Figure BDA00037241466300000422
wherein
Figure BDA00037241466300000423
For a closed set tag noise
Figure BDA00037241466300000424
Class prediction can be obtained by:
Figure BDA0003724146630000051
finally, closed set tag noise obtains pseudo tags
Figure BDA0003724146630000052
And combining the clean samples and the closed set label noise with the pseudo labels, and sending the combined signals into a convolutional neural network for training.
The above examples are only intended to illustrate the technical solution of the present invention, and not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (7)

1. A network supervision fine-grained image identification method based on partial label learning is characterized by comprising the following steps:
step 1, carrying out depth descriptor transformation by using a depth neural network model with pre-training to evaluate positive correlation between network images, detecting open-set label noise existing in a data set according to a correlation matrix reflecting the correlation, and removing the open-set label noise;
step 2, driving the model by using a loss function to enable the label set of each sample to contain the labels of the real types of the samples as much as possible;
and 3, selecting a real label of the sample from the label set of the sample by using the idea of partial label learning, thereby correcting the closed set label noise.
2. The method for identifying the network supervision fine-grained image based on the partial label learning according to the claim 1, characterized in that, in the step 1, the depth descriptor transformation is carried out by utilizing a depth neural network model with pre-training to evaluate the positive correlation between network images, the open set label noise existing in a data set is detected according to a correlation matrix reflecting the correlation, and the open set label noise is removed;
is provided with
Figure FDA0003724146620000011
In order to be a space for the label,
Figure FDA0003724146620000012
for sample space, for each random batch containing n pictures
Figure FDA0003724146620000013
Figure FDA0003724146620000014
By having a pretrained convolutional neural network Φ pre Extracting features containing nFeature graph set of graphs
Figure FDA0003724146620000015
Figure FDA0003724146620000016
Wherein H, W and d respectively represent a characteristic diagram t i Height, width and depth; feature vector with maximum feature value generated by principal component analysis
Figure FDA0003724146620000017
For a given feature map set, each feature map is subjected to channel weighted summation with the feature vector p to obtain a heat map and form the heat map set
Figure FDA0003724146620000018
Calculating the ith feature map t i Corresponding to
Figure FDA0003724146620000019
The ith heatmap H in (II) i ：
Figure FDA00037241466200000110
Wherein
Figure FDA00037241466200000111
Then, each heat map is up-sampled to the size of an input image, and a correlation matrix C is obtained; the correlation matrix is composed of correlation values, positive values representing the sum
Figure FDA00037241466200000112
The existing positive correlation of universal expression similarity, the negative value represents the negative correlation, and the larger the absolute value is, the stronger the correlation is; judging whether the noise is the open set label noise according to the number and the size of positive values in the correlation matrix, and setting oneJudging each sample by a threshold value delta, judging whether the ith sample is noise:
Figure FDA00037241466200000113
if the sample does not satisfy the condition, the open set label noise is regarded and removed from the sample space, thereby obtaining a sample space composed of clean data and closed set label noise
Figure FDA0003724146620000025
3. The method for identifying the network supervision fine-grained image based on the partial label learning according to the claim 2, wherein the step 2 shows a higher recall rate by using a loss function driven model.
Defining a tag space as
Figure FDA0003724146620000026
The sample space is
Figure FDA0003724146620000027
Wherein
Figure FDA0003724146620000028
Indicating that the label belongs to class i y i A set of instances of (a); in the sample selection stage at the beginning of training, C categories are randomly selected to generate a batch
Figure FDA0003724146620000029
For each selected category y i Randomly select n * Samples of the category; on a batch basis
Figure FDA00037241466200000210
By convolutional neural networks Φ CNN Obtaining embedded features
Figure FDA00037241466200000211
For in
Figure FDA00037241466200000212
Sample of (1)
Figure FDA00037241466200000213
By convolutional neural network phi CNN Obtaining an embedding feature f i ：
Figure FDA0003724146620000021
Where c is the embedding feature f i The length of (d); obtaining a similarity matrix by calculating cosine similarity between embedded features
Figure FDA00037241466200000214
The cosine similarity of the ith query image and the jth support image is calculated as follows:
Figure FDA0003724146620000022
similarity s obtained by each query image and other images q，： Arranging the first k pictures with high similarity into a set
Figure FDA00037241466200000215
The preparation method comprises the following steps of (1) performing; defining images that belong to the same category as the query image but are not in the collection
Figure FDA00037241466200000216
The image in (2) is called positive image, in the set
Figure FDA00037241466200000226
An image that is not of the same category as the query image is called a negative image; is not limited toIn the collection
Figure FDA00037241466200000218
Positive images in (1) constitute a set
Figure FDA00037241466200000217
Wherein
Figure FDA00037241466200000219
To represent
Figure FDA00037241466200000220
In
Figure FDA00037241466200000221
Complement of, y q Is a label of the query image; setting n * < k, get a set of negative images
Figure FDA00037241466200000222
Figure FDA00037241466200000223
Wherein
Figure FDA0003724146620000023
Is in the aggregate
Figure FDA00037241466200000224
Number of images belonging to the same category as the query image, s n Is a matrix containing only the similarity scores of negative images; thus, the loss function is defined as
Figure FDA0003724146620000024
Applying this loss function ensures that each image generates as much of a set of tags as possible that contain true classes.
4. The method for network supervision fine-grained image identification based on partial label learning according to claim 3, characterized in that in step 3, a real label is determined from the label set S obtained in step 2 by using partial label learning;
in the encoding stage, an encoding matrix M belongs to +1, -1 }is constructed by randomly generating column codes with N bits N×L Wherein N represents the number of classes, L represents the number of two classifiers, and the coding matrix is used for dividing samples in the training process; a randomly generated column code
Figure FDA00037241466200000225
The label space can be divided into positive label spaces
Figure FDA00037241466200000319
And a negative label space
Figure FDA00037241466200000320
Figure FDA0003724146620000031
Figure FDA0003724146620000032
Selecting positive and negative samples by utilizing positive and negative label space, and giving a training sample
Figure FDA00037241466200000316
Wherein
Figure FDA00037241466200000315
Visual label collection
Figure FDA00037241466200000318
Help build two classifiers for a whole; when label set
Figure FDA00037241466200000317
All fall into
Figure FDA00037241466200000314
Or
Figure FDA00037241466200000312
While, the sample
Figure FDA00037241466200000313
Will be used as positive or negative samples; these positive and negative samples then form a binary training set
Figure FDA00037241466200000311
At the decoding stage, a connected set is constructed for each class, the connected set of the j-th class being represented as:
Figure FDA0003724146620000033
according to a connected set epsilon y Generating a performance matrix G N×L To reflect the capabilities of the classifiers, class j at class t classifier g t The performance above is calculated as follows:
Figure FDA0003724146620000034
wherein
Figure FDA00037241466200000310
Figure FDA00037241466200000321
Is an indicator function, normalizes the performance matrix G line by line:
Figure FDA0003724146620000035
wherein
Figure FDA0003724146620000038
For a closed set tag noise
Figure FDA0003724146620000039
Obtaining a class prediction by:
Figure FDA0003724146620000036
finally, closed set tag noise obtains pseudo tags
Figure FDA0003724146620000037
And combining the clean samples and the closed set label noise with the pseudo labels and sending the combined signals into a convolutional neural network for training.
5. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method of any of claims 1-4 are implemented when the program is executed by the processor.
6. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 4.
7. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the method of any of claims 1 \u4.
CN202210761418.9A 2022-06-30 2022-06-30 Network supervision fine-grained image identification method based on partial label learning Pending CN115170813A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210761418.9A CN115170813A (en) 2022-06-30 2022-06-30 Network supervision fine-grained image identification method based on partial label learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210761418.9A CN115170813A (en) 2022-06-30 2022-06-30 Network supervision fine-grained image identification method based on partial label learning

Publications (1)

Publication Number Publication Date
CN115170813A true CN115170813A (en) 2022-10-11

Family

ID=83489216

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210761418.9A Pending CN115170813A (en) 2022-06-30 2022-06-30 Network supervision fine-grained image identification method based on partial label learning

Country Status (1)

Country Link
CN (1) CN115170813A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115564960A (en) * 2022-11-10 2023-01-03 南京码极客科技有限公司 Network image label denoising method combining sample selection and label correction

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115564960A (en) * 2022-11-10 2023-01-03 南京码极客科技有限公司 Network image label denoising method combining sample selection and label correction

Similar Documents

Publication Publication Date Title
CN112115995B (en) Image multi-label classification method based on semi-supervised learning
CN109754015B (en) Neural networks for drawing multi-label recognition and related methods, media and devices
CN113657425B (en) Multi-label image classification method based on multi-scale and cross-modal attention mechanism
CN109948735B (en) Multi-label classification method, system, device and storage medium
CN115937655A (en) Target detection model of multi-order feature interaction, and construction method, device and application thereof
CN110378911B (en) Weak supervision image semantic segmentation method based on candidate region and neighborhood classifier
CN112926675B (en) Depth incomplete multi-view multi-label classification method under double visual angle and label missing
CN115439685A (en) Small sample image data set dividing method and computer readable storage medium
CN111882000A (en) Network structure and method applied to small sample fine-grained learning
CN116486109A (en) Modal self-adaptive descriptive query pedestrian re-identification method and system
CN117437426B (en) Semi-supervised semantic segmentation method for high-density representative prototype guidance
CN115170813A (en) Network supervision fine-grained image identification method based on partial label learning
CN117315377A (en) Image processing method and device based on machine vision and electronic equipment
CN115422518A (en) Text verification code identification method based on data-free knowledge distillation
CN111914949B (en) Zero sample learning model training method and device based on reinforcement learning
CN117671261A (en) Passive domain noise perception domain self-adaptive segmentation method for remote sensing image
CN116069985A (en) Robust online cross-modal hash retrieval method based on label semantic enhancement
CN113762178B (en) Weak supervision abnormal event time positioning method for background suppression sampling
Escalera et al. Traffic sign recognition system with β-correction
CN114168780A (en) Multimodal data processing method, electronic device, and storage medium
CN113378707A (en) Object identification method and device
CN112926585A (en) Cross-domain semantic segmentation method based on regenerative kernel Hilbert space
CN116825210B (en) Hash retrieval method, system, equipment and medium based on multi-source biological data
CN116612466B (en) Content identification method, device, equipment and medium based on artificial intelligence
CN117520104B (en) System for predicting abnormal state of hard disk

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
CB03 Change of inventor or designer information

Inventor after: Xu Yuyan

Inventor after: Wei Xiucan

Inventor before: Wei Xiucan

Inventor before: Xu Yuyan

CB03 Change of inventor or designer information