CN114998960B - Expression recognition method based on positive and negative sample contrast learning - Google Patents
Expression recognition method based on positive and negative sample contrast learning Download PDFInfo
- Publication number
- CN114998960B CN114998960B CN202210595007.7A CN202210595007A CN114998960B CN 114998960 B CN114998960 B CN 114998960B CN 202210595007 A CN202210595007 A CN 202210595007A CN 114998960 B CN114998960 B CN 114998960B
- Authority
- CN
- China
- Prior art keywords
- sample
- samples
- positive
- negative
- expression
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 230000001815 facial effect Effects 0.000 claims abstract description 13
- 238000010801 machine learning Methods 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 14
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000010586 diagram Methods 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 2
- 230000008569 process Effects 0.000 abstract description 2
- 238000005065 mining Methods 0.000 abstract 1
- 230000008921 facial expression Effects 0.000 description 3
- 238000005286 illumination Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000001747 exhibiting effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/174—Facial expression recognition
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an expression recognition method based on positive and negative sample contrast learning, which comprises the following steps of S1, collecting facial images; s2, inputting the facial image into a trained machine learning model to identify the expression in the facial image; s3, outputting expression categories of the facial images; in the process of machine learning a model, a negative sample is introduced in the structure similarity comparison learning method, and the distance between a positive sample and a negative sample in a batch of samples is considered to be zoomed out. Meanwhile, in consideration of unbalance between data volumes of positive samples and negative samples in a batch of samples, the negative samples are subjected to difficult sample mining, so that the model reduces constraint on the negative samples which are most similar to the positive samples while increasing constraint on structural similarity of the positive samples, and further plays a comparison learning effect, and the expression recognition accuracy is improved.
Description
Technical Field
The invention relates to the technical field of expression recognition, in particular to an expression recognition method based on positive and negative sample contrast learning.
Background
The facial expression recognition technology is used for recognizing the emotion of a person in an image through a computer, and has important practical application values, including aspects of intelligent medical treatment, safe driving detection, online education, psychological coaching, entertainment interaction and the like. But in view of the expression recognition accuracy in the actual scene, the accuracy of facial expression recognition does not reach the level of human being at present.
Expression recognition in natural scenes is more challenging than the expression recognition task of controlled experimental scenes because the acquired images in natural conditions contain more environmental factor differences such as illumination, resolution and larger amplitude of pose changes. Secondly, because the acquisition conditions are not uniform, the expression recognition task in the natural scene presents a larger problem of large difference in similar classes, such as 'surprise' expression, and the external appearance presents a large difference because of different shooting angles, illumination conditions and character expression modes. Also, images from different classes have similar appearances, exhibiting large inter-class similarities. Another example is the "happy" category, which is the same for smiles and laughter due to different acquisition conditions, but has a large difference in image appearance, and a small difference in appearance for images of the "wounded" and "fear" categories. Finally, the expression data set collected by the natural scene usually needs labeling personnel to be labeled manually, which consumes a great deal of manpower, and more importantly, different people can make different judgments on the same image due to subjective influence of the people, so that the label of the expression has large ambiguity.
The key of the facial expression recognition task in the natural scene is to extract the characteristics with expression meaning, and many methods for learning and optimizing the local characteristics and methods based on measurement learning achieve good effects. Although many methods have been previously proposed to solve the problems of large intra-class variance and tag ambiguity for inter-class similarity, they are all trained in a traditional supervised manner. Under a natural scene, the problem that the label is ambiguous in the expression recognition task is not negligible, and the original label is simply utilized to possibly introduce wrong information, so that the ambiguous label can mislead the learning of the expression characteristics by the network model, and the supervised learning can not fully mine the fine expression discrimination characteristics.
Therefore, how to propose an expression recognition method based on positive and negative sample contrast learning, which can effectively improve the expression recognition accuracy, is a problem that needs to be solved by those skilled in the art.
Disclosure of Invention
In view of the above, the invention provides an expression recognition method based on positive and negative sample contrast learning, and aims to provide a more accurate expression recognition method.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
an expression recognition method based on positive and negative sample contrast learning comprises the following steps:
s1, collecting a face image;
s2, inputting the facial image into a trained machine learning model to identify the expression in the facial image;
s3, outputting expression categories of the facial images;
the training method of the machine learning model in the S2 specifically comprises the following steps:
s21, generating a plurality of new samples for each sample;
for a batch of samplesObtaining ∈10 after performing strong data enhancement>Obtaining ∈10 after weak data enhancement>Wherein x is i Is a face image, y i Is x i Corresponding expression labels; obtaining a current weak enhancement sample x i ' and strong enhancement sample x i Considered as a positive sample pair; will be identical to the current sample x i Enhancement samples of all other samples of the' different classes are taken as negative sample pairs, the set of negative sample pairs of the ith sample is denoted +.>
S22, inputting a deep neural network, and extracting the characteristics of each sample;
for weak enhancement sample x' i And strong enhancement sample x i After feature extraction, feature representations are obtained respectively: u= { U i |i＝1,2,……HW},V＝{v j I j=1, 2, … … HW }, where u i ,v j A feature vector representation as a supply point and a destination point, respectively;
s23, calculating a loss function of the deep neural network;
defining the negative sample similarity set of the ith sample asThe aggregate size is K i ；
The total loss function is:
L＝L cls +βL hard
wherein beta is a hyper-parameter representing a balance coefficient; l (L) cls Classifying loss functions for softmax, L hard Adaptive weighting function for negative samples:
wherein N is the number of samples in the batch;converting probability for structural similarity, gamma is scaling coefficient, s i,j Representing the structural similarity between the ith sample and the jth enhanced sample; the sum of probabilities of all samples is 1;
s24, optimizing parameters of the deep neural network according to the loss function.
Preferably, in S23, the structural similarity S between the ith feature representation and the jth enhanced feature i,j The specific calculation method of (a) is as follows:
s i,j ＝-s(U i ,V j )
wherein H and W are the height and width of convolution feature maps U and V, respectively, c ij Representing the transmission cost between source supply side node i to destination node j, U i A convolution characteristic diagram representing the ith sample, V j A convolution signature representing the jth sample,representing an optimal transportation scheme, wherein the optimal transportation scheme obtains an optimal solution by calculating the following formulaThe method comprises the following steps:
subject to f ij ≥0,i＝1,…,m,j＝1,…,k
wherein,the local weights of the source node and the destination node are represented, and the calculation mode is solved by the following formula:
wherein G is avg Representing a global average pooling operation.
Preferably, c ij The calculation mode of (a) is as follows:
wherein u is i ，v j Is a feature vector representation of the feed point and the destination point.
Preferably, in S23, the relative magnitude p of the structural similarity between the ith sample and the kth enhanced sample i,k The specific calculation method of (a) is as follows:
when p is i,k The larger the value, the greater the learning weight given to the corresponding portion of the sample.
Compared with the prior art, the invention discloses an expression recognition method based on positive and negative sample contrast learning, which has the following beneficial effects:
1. according to the invention, the EMD distance is acted on the feature map, so that the model can be guided to pay attention to the region related to the expression, and the region such as noisy background and the like which is irrelevant to the expression is eliminated, thereby effectively paying attention to the expression region.
2. The invention designs the constraint loss of the structural similarity by adopting the self-supervision thought, and utilizes the contrast learning of the positive and negative samples of the enhanced image to optimize the model and the classification loss together under the condition of not depending on the original label so as to learn the expression characteristics with more generalization.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of an expression recognition method based on positive and negative sample contrast learning provided by the invention;
FIG. 2 is a schematic diagram of a training method of a machine learning model in an expression recognition method based on positive and negative sample contrast learning;
fig. 3 is a schematic diagram of self-adaptive weighting of negative samples in an expression recognition method based on positive and negative sample contrast learning.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The embodiment of the invention discloses an expression recognition method based on positive and negative sample contrast learning, which is shown in fig. 1-3 and comprises the following steps:
s1, collecting a face image;
s2, inputting the facial image into a trained machine learning model to identify the expression in the facial image;
s3, outputting expression categories of the facial images;
the training method of the machine learning model in S2 specifically comprises the following steps:
s21, generating a plurality of new samples for each sample;
for a batch of samplesObtaining ∈10 after performing strong data enhancement>Obtaining ∈10 after weak data enhancement>Wherein x is i Is a face image, y i Is x i Corresponding expression labels; obtaining a current weak enhancement sample x' i And its strong enhancement sample x i Considered as a positive sample pair; will be identical to the current sample x' i Enhancement samples of all other samples of different classes are taken as negative sample pairs, the set of negative sample pairs of the ith sample is denoted +.>
Wherein images of two different viewing angles are obtained by weak enhancement and strong enhancement. Weak enhancement refers to the fact that the image is scaled and flipped to a small extent, and the strong enhancement data enhancement mode also comprises rotation and color transformation.
S22, inputting a deep neural network, and extracting the characteristics of each sample;
for weak enhancement sample x' i And strong enhancement sample x i After feature extraction, feature representations are obtained respectively: u= { U i |i＝1,2,……HW},V＝{v j I j=1, 2, … … HW }, where u i ,v j A feature vector representation as a supply point and a destination point, respectively;
s23, calculating a loss function of the deep neural network;
defining the negative sample similarity set of the ith sample asThe aggregate size is K i ；
The total loss function is:
L＝L cls +βL hard
wherein beta is a hyper-parameter representing a balance coefficient; l (L) cls Classifying loss functions for softmax, L hard Adaptive weighting function for negative samples:
wherein N is the number of samples in the batch;for the structural similarity transition probabilities, γ is a scaling factor, and γ affects the degree of equalization between sample classification probabilities. The larger the value of γ, the larger the difference in classification probabilities, and the smaller γ, the closer the classification probabilities between samples. When the gamma value region is zero, the probability distribution is degraded into uniform distribution; s is(s) i,j Representing the structural similarity between the ith sample and the jth enhanced sample; the sum of probabilities of all samples is 1; wherein s is i,j Representing the structural similarity between the ith sample and the jth enhanced sample.
S24, optimizing parameters of the deep neural network according to the loss function.
In order to further implement the above technical solution, in S23, the i-th feature represents the structural similarity S between the j-th enhanced feature i,j The specific calculation method of (a) is as follows:
s i,j ＝-s(U i ,V j )
wherein H and W are the height and width of convolution feature maps U and V, respectively, c ij Representing the transmission cost between source supply side node i to destination node j, U i A convolution characteristic diagram representing the ith sample, V j A convolution signature representing the jth sample,the optimal transportation scheme is represented, and the optimal solution method is obtained by calculating the following formula:
subject to f ij ≥0,i＝1,…,m,j＝1,…,k
wherein,the local weights of the source node and the destination node are represented, and the calculation mode is solved by the following formula:
wherein G is avg Representing a global average pooling operation.
To further implementThe technical proposal, c ij The calculation mode of (a) is as follows:
wherein u is i ，v j Is a feature vector representation of the feed point and the destination point.
In order to further implement the above technical solution, in S23, the relative size p of the structural similarity between the ith sample and the kth enhanced sample i,k The specific calculation method of (a) is as follows:
when p is i,k The larger the value, the greater the learning weight given to the corresponding portion of the sample.
p i,k The relative magnitudes of the structural similarities between the ith sample and the kth enhanced sample are shown, reflecting the degree of similarity of the two view angle features. P is p i,k The larger the value, the more similar the pair of samples, the more difficult the enhanced sample is to distinguish, and the model should give more learning weight to that portion of the sample.
It should be noted that:
for sample x i The positive samples are defined as enhanced samples x i . The negative samples should maintain a larger structural similarity difference, and there may be samples originally labeled as the same class in one batch, and this part of samples may have a higher structural similarity, and regarding the enhanced samples of the same class as the negative samples may introduce more error information. For correctness, the invention regards negative samples as enhancement samples of other classes, the set being defined asI.e. enhanced samples of all other samples of a different class than the current sample. This ensures the correctness of the negative samples to some extent.
The loss function is calculated as follows, taking into account the contrast learning of the negative samples and adding adaptive weights to the difficult negative samples.
Difficulty in measuring samples. Each sample is considered as a separate class, and each negative sample is converted to a class probability by a softamx function according to the structural similarity score.
In this embodiment, β is a balance coefficient, and as a super parameter, adjustment is required, and β in this embodiment takes a value of 0.6.
The deep neural network model of this embodiment runs on an nvidia T itan 3090GPU server that mounts a Pytorch (v 1.7) deep learning framework. The present embodiment is a ResNet-18 based backbone network with all images scaled to 256 by 256 sizes. In the training phase, the input image is cut into 224×224 sizes through random positions, and in the testing phase, the input image is cut into 224×224 sizes through center positions for testing. There are weak and strong enhancements in the data enhancement phase: at weak enhancement, the image is flipped with a probability of 0.5; when the enhancement is strong, the color random dithering is performed with a probability of 0.2 in addition to the weak enhancement.
Since the display memory consumption in the structural similarity calculation process is considered to be too large, particularly in the step of calculating the structural similarity between every two in a batch, the display memory occupation ratio is large, and the batch size is set to be 32. The model was optimized using a random gradient descent method with momentum, a learning rate of 0.01, a momentum of 0.9, and a weight decay rate of 0.0001. The EMD distance calculation module needs to calculate the corresponding optimal transportation scheme, which is calculated by OpenCV library functions.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (4)
1. The expression recognition method based on positive and negative sample contrast learning is characterized by comprising the following steps of:
s1, collecting a face image;
s2, inputting the facial image into a trained machine learning model to identify the expression in the facial image;
s3, outputting expression categories of the facial images;
the training method of the machine learning model in the S2 specifically comprises the following steps:
s21, generating a plurality of new samples for each sample;
for a batch of samplesObtaining ∈10 after performing strong data enhancement>Obtaining ∈10 after weak data enhancement>Wherein x is i Is a face image, y i Is x i Corresponding expression labels; obtaining a current weak enhancement sample x' i And its strong enhancement sample x i Considered as a positive sample pair; will be identical to the current sample x' i Enhancement samples of all other samples of different classes are taken as negative sample pairs, the set of negative sample pairs of the ith sample is denoted +.>
S22, inputting a deep neural network, and extracting the characteristics of each sample;
for weak enhancement sample x' i And strong enhancement sample x i After feature extraction, feature representations are obtained respectively: u= { U i |i＝1,2,……HW},V＝{v j I j=1, 2, … … HW }, where u i ,v j A feature vector representation as a supply point and a destination point, respectively;
s23, calculating a loss function of the deep neural network;
defining the negative sample similarity set of the ith sample asThe aggregate size is K i ；
The total loss function is:
L＝L cls +βL hard
wherein beta is a hyper-parameter representing a balance coefficient; l (L) cls Classifying loss functions for softmax, L hard Adaptive weighting function for negative samples:
wherein N is the number of samples in the batch;converting probability for structural similarity, gamma is scaling coefficient, s i,j Representing the structural similarity between the ith sample and the jth enhanced sample; the sum of probabilities of all samples is 1;
s24, optimizing parameters of the deep neural network according to the loss function.
2. The expression recognition method based on positive and negative sample contrast learning as claimed in claim 1, wherein in S23, structural similarity S between the ith feature representation and the jth enhanced feature i,j The specific calculation method of (a) is as follows:
s i,j ＝-s(U i ,V j )
wherein H and W are the height and width of convolution feature maps U and V, respectively, c ij Representing the transmission cost between source supply side node i to destination node j, U i A convolution characteristic diagram representing the ith sample, V j A convolution signature representing the jth sample,the optimal transportation scheme is represented, and the optimal solution method is obtained by calculating the following formula:
subject to f ij ≥0,i＝1,…,m,j＝1,…,k
wherein,the local weights of the source node and the destination node are represented, and the calculation mode is solved by the following formula:
wherein G is avg Representing a global average pooling operation.
3. The expression recognition method based on positive and negative sample contrast learning as claimed in claim 2, wherein c ij The calculation mode of (a) is as follows:
wherein u is i ，v j Is a feature vector representation of the feed point and the destination point.
4. The method for recognizing expression based on positive and negative sample contrast learning according to claim 2, wherein in S23, the relative magnitude p of the structural similarity between the ith sample and the kth enhanced sample i,k The specific calculation method of (a) is as follows:
when p is i,k The larger the value, the greater the learning weight given to the corresponding portion of the sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210595007.7A CN114998960B (en) | 2022-05-28 | 2022-05-28 | Expression recognition method based on positive and negative sample contrast learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210595007.7A CN114998960B (en) | 2022-05-28 | 2022-05-28 | Expression recognition method based on positive and negative sample contrast learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114998960A CN114998960A (en) | 2022-09-02 |
| CN114998960B true CN114998960B (en) | 2024-03-26 |
Family
ID=83028670
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210595007.7A Active CN114998960B (en) | 2022-05-28 | 2022-05-28 | Expression recognition method based on positive and negative sample contrast learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114998960B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117079336B (en) * | 2023-10-16 | 2023-12-22 | 腾讯科技（深圳）有限公司 | Training method, device, equipment and storage medium for sample classification model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110851645A (en) * | 2019-11-08 | 2020-02-28 | 吉林大学 | Image retrieval method based on similarity maintenance under depth metric learning |
| CN110866134A (en) * | 2019-11-08 | 2020-03-06 | 吉林大学 | Image retrieval-oriented distribution consistency keeping metric learning method |
| CN111723675A (en) * | 2020-05-26 | 2020-09-29 | 河海大学 | Remote sensing image scene classification method based on multiple similarity measurement deep learning |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163234B (en) * | 2018-10-10 | 2023-04-18 | 腾讯科技（深圳）有限公司 | Model training method and device and storage medium |
| CN111814871B (en) * | 2020-06-13 | 2024-02-09 | 浙江大学 | Image classification method based on reliable weight optimal transmission |
-
2022
- 2022-05-28 CN CN202210595007.7A patent/CN114998960B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110851645A (en) * | 2019-11-08 | 2020-02-28 | 吉林大学 | Image retrieval method based on similarity maintenance under depth metric learning |
| CN110866134A (en) * | 2019-11-08 | 2020-03-06 | 吉林大学 | Image retrieval-oriented distribution consistency keeping metric learning method |
| CN111723675A (en) * | 2020-05-26 | 2020-09-29 | 河海大学 | Remote sensing image scene classification method based on multiple similarity measurement deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 少量样本下基于孪生CNN的SAR目标识别;王博威;潘宗序;胡玉新;马闻;;雷达科学与技术;20191215(06);第17-23页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114998960A (en) | 2022-09-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106803069B (en) | Crowd happiness degree identification method based on deep learning | |
| CN110399821B (en) | Customer satisfaction acquisition method based on facial expression recognition | |
| CN107895160A (en) | Human face detection and tracing device and method | |
| CN109522973A (en) | Medical big data classification method and system based on production confrontation network and semi-supervised learning | |
| CN112528928B (en) | Commodity identification method based on self-attention depth network | |
| CN112784763A (en) | Expression recognition method and system based on local and overall feature adaptive fusion | |
| CN109255289B (en) | Cross-aging face recognition method based on unified generation model | |
| CN109359608A (en) | A kind of face identification method based on deep learning model | |
| US20230162522A1 (en) | Person re-identification method of integrating global features and ladder-shaped local features and device thereof | |
| CN109710804B (en) | Teaching video image knowledge point dimension reduction analysis method | |
| CN110263174B (en) | Topic category analysis method based on focus attention | |
| Bu | Human motion gesture recognition algorithm in video based on convolutional neural features of training images | |
| He et al. | Open-vocabulary multi-label classification via multi-modal knowledge transfer | |
| CN110969073B (en) | Facial expression recognition method based on feature fusion and BP neural network | |
| CN108520213A (en) | A kind of face beauty prediction technique based on multiple dimensioned depth | |
| CN109377429A (en) | A kind of recognition of face quality-oriented education wisdom evaluation system | |
| CN113435335B (en) | Microscopic expression recognition method and device, electronic equipment and storage medium | |
| CN105404865A (en) | Probability state restricted Boltzmann machine cascade based face detection method | |
| CN112818850A (en) | Cross-posture face recognition method based on progressive neural network and attention mechanism | |
| CN114998960B (en) | Expression recognition method based on positive and negative sample contrast learning | |
| CN108345833A (en) | The recognition methods of mathematical formulae and system and computer equipment | |
| CN111368768A (en) | Human body key point-based employee gesture guidance detection method | |
| CN111461162B (en) | Zero-sample target detection model and establishing method thereof | |
| Zheng et al. | Attention assessment based on multi‐view classroom behaviour recognition | |
| CN111931630B (en) | Dynamic expression recognition method based on facial feature point data enhancement |
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 |