CN113610144A - Vehicle classification method based on multi-branch local attention network - Google Patents

Vehicle classification method based on multi-branch local attention network Download PDF

Info

Publication number
CN113610144A
CN113610144A CN202110881344.8A CN202110881344A CN113610144A CN 113610144 A CN113610144 A CN 113610144A CN 202110881344 A CN202110881344 A CN 202110881344A CN 113610144 A CN113610144 A CN 113610144A
Authority
CN
China
Prior art keywords
vehicle
branch
attention
local attention
network
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
CN202110881344.8A
Other languages
Chinese (zh)
Inventor
周平
陈晨
闫如根
赵吉祥
胡昌隆
吕强
李涛
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.)
Zenmorn Hefei Technology Co ltd
Original Assignee
Zenmorn Hefei Technology Co ltd
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 Zenmorn Hefei Technology Co ltd filed Critical Zenmorn Hefei Technology Co ltd
Priority to CN202110881344.8A priority Critical patent/CN113610144A/en
Publication of CN113610144A publication Critical patent/CN113610144A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a vehicle classification method based on a multi-branch local attention network, which comprises the following specific steps of obtaining a vehicle picture to be classified and dividing the vehicle picture into a training set and a test set; inputting a training set of vehicle pictures into a vehicle classification model for constructing a multi-branch local attention network for training, wherein the multi-branch local attention network comprises a convolution attention module based on a channel and a local attention module based on a space; and carrying out classification prediction on the test set of the vehicle picture according to the trained vehicle classification model to obtain a classification result. According to the invention, a multi-branch local attention structure is added in the original ResNet-50 model, and the structure can more accurately acquire information of different neighborhoods in a characteristic diagram, so that the expressiveness of key characteristics is enhanced, and the vehicle classification accuracy is improved; meanwhile, the multi-branch attention structure has portability and can be embedded into other network models.

Description

Vehicle classification method based on multi-branch local attention network
Technical Field
The invention relates to the technical field of deep learning and computer vision, in particular to a vehicle classification method based on a multi-branch local attention network.
Background
In recent years, the development of an Intelligent Transportation System (ITS) is rapid, and along with the development of computer vision and deep learning, a chance is provided for more effective application of the intelligent transportation system. Computer vision is to simulate human visual ability by using a computer, extract information from an image of an objective object, process and understand the information, and finally apply the information to actual life, such as vehicle classification.
Conventional vision-based vehicle classification approaches typically rely on manual features such as Color Histograms (CH), Texture Descriptors (TD), GIST), or representations generated by encoding local features such as boww, IFK, SPM, etc., which are time consuming, labor intensive, have poor generalization capability, and are susceptible to environmental changes and occlusion. With the rapid development of deep learning theory and practice, target detection and classification based on deep learning enter a new stage. Different from the traditional feature extraction algorithm, the convolutional neural network has certain invariant geometric transformation, deformation and illumination, can overcome the difficulty of changing the appearance of the vehicle, can solve the problem of shielding, can adaptively describe the features constructed under the drive of training data, and has higher flexibility and comprehensive capability.
The existing convolutional neural network method still has the defects that key information and redundant information are not clearly distinguished, and the feature recognition capability of the model is still not strong enough. During the period of multi-layer convolution and pooling, a large amount of important information is lost, so that the extracted features cannot well represent the target.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and adopts a vehicle classification method based on a multi-branch local attention network to solve the problems in the background technology.
A vehicle classification method based on a multi-branch local attention network comprises the following specific steps:
obtaining a picture of a vehicle to be classified and dividing the picture into a training set and a test set;
inputting a training set of vehicle pictures into a vehicle classification model for constructing a multi-branch local attention network for training, wherein the multi-branch local attention network comprises a convolution attention module based on a channel and a local attention module based on a space;
and carrying out classification prediction on the test set of the vehicle picture according to the trained vehicle classification model to obtain a classification result.
As a further aspect of the invention: the specific steps of obtaining the picture of the vehicle to be classified and dividing the picture into a training set and a test set comprise:
acquiring a vehicle scene image, and dividing the vehicle scene image into a training set and a test set, wherein the ratio of the training set to the test set is 4: 1;
the training set comprises M training pictures which are X ═ X1,X2,…,Xm,…,XMThe training set comprises N training pictures which are Y ═ Y1,Y2,…,Yn,…,YNIn which X ismDenotes the m-th training picture, YnRepresenting the nth test picture;
and carrying out one-to-one labeling on the labels and the original pictures of the training set and the test set.
As a further aspect of the invention: the vehicle classification model for constructing the multi-branch local attention network is input into the training set of the vehicle pictures for training, the multi-branch local attention network comprises a convolution attention module based on a channel, and the specific steps of the local attention module based on a space comprise:
constructing a multi-branch local attention network structure based on a ResNet-50 model;
using a channel-based convolution attention module to be introduced as a first branch;
a local attention module based on space is used for introduction as a second branch;
the two branches are fused by a parallel method, and the fusion formula is as follows:
F′＝σ(MC(F)×MS(F))×F；
wherein M isc(F) Indicating the channel attention, Ms(F) Representing spatial attention, σ () is a sigmoid activation function;
and embedding a multi-branch local attention network structure into each bottleeck layer of the ResNet-50 to obtain a vehicle classification model based on the multi-branch local attention network.
As a further aspect of the invention: the specific steps introduced as the first branch using the channel-based convolution attention module include:
aggregating spatial information of feature maps using global average pooling and global maximum pooling operations, i.e.
Figure BDA0003192131760000021
And
Figure BDA0003192131760000022
by using 2 x 1 convolution method
Figure BDA0003192131760000023
And
Figure BDA0003192131760000024
combining the two parts;
finally, a multi-layer perceptron is added to learn the final channel attention feature map Mc(F) The formula is as follows:
Figure BDA0003192131760000025
wherein, W0∈R(C/r)×C，W1∈RC，W0And W1Is the weight of the multi-layer perceptron MLP, f2×1Representing a convolution operation with a filter size of 2 x 1.
As a further aspect of the invention: the specific steps of using the space-based local attention module as a second branch include:
applying local space maximum pooling and average pooling in parallel to all channels of the original feature map, wherein the kernel and step size are equal to epsilon;
generating two compressed spatial attention descriptors F by aggregating features within all ε -neighborhoods of FMAnd FA；
FA＝AvgPoolkernel,stride＝ε(F)；
FM＝MaxPoolkernel,stride＝ε(F)；
Tightening F along the channel direction using global max pooling and global average poolingMAnd FAGenerating descriptors
Figure BDA0003192131760000031
And
Figure BDA0003192131760000032
and are connected together;
sequentially increasing the convolution of the 3 multiplied by 3 holes and the nearest neighbor interpolation operation to obtain a characteristic diagram of space attention;
MS(F)＝σ(fnearest(f([Fmax(FM)；Favg(FA)])))；
wherein MaxPoint and AvgPool represent the kernel and the local maximal pooling and average pooling operation with step size ε in the spatial domain, Fmax(. about.) and Favg(x) is the maximum pooling and average pooling operation along the channel direction, f (x) represents the convolution of 3 × 3 holes, fnearest() is the nearest neighbor interpolation upsampling operator and σ () is the sigmoid activation function.
As a further aspect of the invention: the specific steps of carrying out classification prediction on the test set of the vehicle picture according to the trained vehicle classification model to obtain a classification result comprise:
setting a loss function, adjusting the size of the vehicle image in the training set, inputting the vehicle image into a multi-branch local attention network for training to obtain a trained network model;
and (4) utilizing the extraction network full-link layer of the trained network model to classify and predict the vehicle images in the test set by using a Softmax function to obtain a classification result.
Compared with the prior art, the invention has the following technical effects:
by adopting the technical scheme, the vehicle pictures are obtained, then the vehicle classification model of the multi-branch local attention network is constructed on the basis of the RestNet-50 model, wherein the vehicle classification model comprises a convolution attention module based on a channel and a local attention module based on a space, and then the vehicle pictures are predicted and classified by extracting the network full-connection layer and using a softmax function. The method can more accurately acquire the information of different neighborhoods in the feature map, enhance the expressive property of the features and improve the accuracy of vehicle classification. Meanwhile, the problem that the extracted features cannot well represent the target due to repeated pooling and convolution and loss of a large amount of important information is avoided.
Drawings
The following detailed description of embodiments of the invention refers to the accompanying drawings in which:
FIG. 1 is a schematic step diagram of a vehicle classification method according to some embodiments disclosed herein;
FIG. 2 is a schematic network model flow diagram of some embodiments disclosed herein;
FIG. 3 is a schematic diagram of a network architecture of some embodiments disclosed herein;
FIG. 4 is a schematic diagram of a channel-based convolution attention module of some embodiments disclosed herein;
FIG. 5 is a schematic view of a spatial-based local attention module of some embodiments disclosed herein;
fig. 6 is a schematic diagram of a multi-branch attention structure integrated CNN unit of some embodiments disclosed herein.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1 and fig. 2, in an embodiment of the present invention, a vehicle classification method based on a multi-branch local attention network includes:
s1, obtaining the picture of the vehicle to be classified and dividing the picture into a training set and a testing set, wherein the method comprises the following specific steps:
acquiring an image of a required vehicle in a specific scene, and dividing the image into a training set and a test set, wherein the ratio of the training set to the test set is 4: 1;
the training set comprises M training pictures which are X ═ X1,X2,…,Xm,…,XMThe training set comprises N training pictures which are Y ═ Y1,Y2,…,Yn,…,YNIn which X ismDenotes the m-th training picture, YnRepresenting the nth test picture;
and carrying out one-to-one labeling on the labels and the original pictures of the training set and the test set.
S2, inputting a training set of vehicle pictures into a vehicle classification model for constructing a multi-branch local attention network for training, wherein the multi-branch local attention network comprises a convolution attention module based on a channel and a local attention module based on a space, and the method specifically comprises the following steps:
s21, constructing a multi-branch local attention network structure based on the ResNet-50 model; specifically, as shown in fig. 3, a frame structure diagram of a vehicle image classification model is illustrated, a classic CNN model is combined with an attention mechanism, and an output dimension of the classification model is set as the total number of vehicle categories. And the convolutional neural network ResNet-50 is used as a backbone network of a classification model for extracting an original feature map from the vehicle image.
And S22, introducing a convolution attention module based on the channel as a first branch for calculating the weight of the feature map channel, and acquiring the feature map with updated channel attention based on the calculated weight of the feature map channel and the original feature map.
In an embodiment of step S22, as shown in fig. 4, the step of using the channel-based convolution attention module as the first branch includes:
the spatial information of the feature map is first aggregated using global average pooling and global maximum pooling operations, i.e. two 1 × 1 × c channels are obtained
Figure BDA0003192131760000051
And
Figure BDA0003192131760000052
mean pooling and maximum pooling, respectively.
Then using 2 x 1 convolution method to make
Figure BDA0003192131760000053
And
Figure BDA0003192131760000054
combining the two parts;
finally, a multi-layer perceptron is added to learn the final channel attention feature map Mc(F) In order to reduce the parameter overhead, the size of the hidden layer is set to C/r, where r is the compression ratio. Specifically, it has been found experimentally that better results can be obtained by setting the compression ratio r to 16, with the following formula:
Figure BDA0003192131760000055
wherein, W0∈R(C/r)×C，W1∈RC，W0And W1Is the weight of the multi-layer perceptron MLP, f2×1Representing a convolution operation with a filter size of 2 x 1.
And S23, introducing a local attention module based on space as a second branch for calculating the similarity of each feature in the feature map with other features to obtain the feature map after the spatial attention is updated.
In an embodiment of step S23, as shown in fig. 5, the step of using the space-based local attention module as the second branch includes:
spatial attention is achieved by using local similarity (spatial local pooling), and similarity calculation of non-adjacent image regions is achieved by calculating the similarity of each feature in the feature map to other features.
Applying local space maximum pooling and average pooling in parallel to all channels of the original feature map, wherein the kernel and step size are equal to epsilon;
generating two compressed spatial attention descriptors F by aggregating features within all ε -neighborhoods of FMAnd FA；
FA＝AvgPoolkernel,stride＝ε(F)；
FM＝MaxPoolkernel,stride＝ε(F)；
Tightening F along the channel direction using global max pooling and global average poolingMAnd FAGenerate a descriptor sum
Figure BDA0003192131760000061
And
Figure BDA0003192131760000062
each feature represents the maximum pooling feature and the average pooling feature across the channel. And connecting the feature descriptors to reduce the computation cost;
in order to reduce the feature loss caused by the multi-step pooling operation, the 3 x 3 hole convolution is added to further learn the features, and meanwhile, the nonlinear representation capability of the spatial attention descriptor is improved. Finally, upsampling is carried out through nearest neighbor interpolation, and a space attention feature map M with the same scale as the original input image is obtainedS(F)；
MS(F)＝σ(fnearest(f([Fmax(FM)；Favg(FA)」)))；
Wherein MaxPoint and AvgPool represent the kernel and the local maximal pooling and average pooling operation with step size ε in the spatial domain, Fmax(. about.) and Favg(x) is the maximum pooling and average pooling operation along the channel direction, f (x) represents the convolution of 3 × 3 holes, fnearest() is the nearest neighbor interpolation upsampling operator and σ () is the sigmoid activation function.
S24, merging the two branches by using a parallel method, as shown in fig. 6, connecting the feature diagram output from the above step after updating the attention of the channel and the feature diagram output from the above step after updating the attention of the space in parallel, and embedding the feature diagrams into each Layer of the backbone network;
the fusion formula is:
F′＝σ(MC(F)×MS(F))×F；
wherein M isc(F) Indicating the channel attention, Ms(F) Representing spatial attention, σ () is a sigmoid activation function; the parallel method has fewer activation functions than the sequential method, thereby having a larger characterization range and stronger feature extraction capability.
And embedding a multi-branch local attention network structure into each bottleeck layer of the ResNet-50 to obtain a vehicle classification model based on the multi-branch local attention network.
S3, carrying out classification prediction on the test set of the vehicle picture according to the trained vehicle classification model to obtain a classification result, and the specific steps comprise:
s31, setting a loss function, adjusting the size of the vehicle image in the training set, inputting the vehicle image into a multi-branch local attention network for training, and obtaining a trained network model;
specifically, the size of the vehicle image is adjusted to 224 × 224 pixels, and then the pixels of the image are normalized (normalization) to be used as the input of the classification model. The loss function is a cross entropy loss function.
When training, parameters in the CNN model pre-trained by ImageNet are loaded, but parameters of the fixed (freeze) classification model are used, and fine-tuning (fine-tune) is performed on the basis, for example, all parameters of the backbone network are fine-tuned by gradient descent.
And S32, carrying out classification prediction on the vehicle images in the test set by using a Softmax function through the trained network full-link layer to obtain a classification result.
In some embodiments, the backbone network in step S2 is ResNet-50, which can be selected from other networks, such as AlexNet, VGGNet, google lenet, or densneet.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents, which should be construed as being within the scope of the invention.

Claims (6)

1. A vehicle classification method based on a multi-branch local attention network is characterized by comprising the following specific steps:
obtaining a picture of a vehicle to be classified and dividing the picture into a training set and a test set;
inputting a training set of vehicle pictures into a vehicle classification model for constructing a multi-branch local attention network for training, wherein the multi-branch local attention network comprises a convolution attention module based on a channel and a local attention module based on a space;
and carrying out classification prediction on the test set of the vehicle picture according to the trained vehicle classification model to obtain a classification result.
2. The method for classifying vehicles based on the multi-branch local attention network according to claim 1, wherein the specific steps of obtaining the images of the vehicles to be classified and dividing the images into a training set and a test set comprise:
acquiring a vehicle scene image, and dividing the vehicle scene image into a training set and a test set, wherein the ratio of the training set to the test set is 4: 1;
the training set comprises M training pictures which are X ═ X1,X2,…,Xm,…,XMThe training set comprises N training pictures which are Y ═ Y1,Y2,…,Yn,…,YNIn which X ismDenotes the m-th training picture, YnRepresenting the nth test picture;
and carrying out one-to-one labeling on the labels and the original pictures of the training set and the test set.
3. The method for classifying vehicles based on the multi-branch local attention network as claimed in claim 1, wherein the inputting the training set of vehicle pictures into the vehicle classification model for constructing the multi-branch local attention network is trained, the multi-branch local attention network comprises a convolution attention module based on channels, and the specific steps of the local attention module based on space comprise:
constructing a multi-branch local attention network structure based on a ResNet-50 model;
using a channel-based convolution attention module to be introduced as a first branch;
a local attention module based on space is used for introduction as a second branch;
the two branches are fused by a parallel method, and the fusion formula is as follows:
F′＝σ(MC(F)×MS(F))×F；
wherein M isc(F) Indicating the channel attention, Ms(F) Representing spatial attention, σ () is a sigmoid activation function;
and embedding a multi-branch local attention network structure into each bottleeck layer of the ResNet-50 to obtain a vehicle classification model based on the multi-branch local attention network.
4. The method for classifying vehicles based on a multi-branch local attention network according to claim 3, wherein the step of introducing the first branch by using the channel-based convolution attention module comprises:
aggregating nulls for feature maps using global average pooling and global maximum pooling operationsInter information, i.e.
Figure FDA0003192131750000021
And
Figure FDA0003192131750000022
by using 2 x 1 convolution method
Figure FDA0003192131750000023
And
Figure FDA0003192131750000024
combining the two parts;
finally, a multi-layer perceptron is added to learn the final channel attention feature map Mc(F) The formula is as follows:
Figure FDA0003192131750000025
wherein, W0∈R(C/r)×C，W1∈RC，W0And W1Is the weight of the multi-layer perceptron MLP, f2×1Representing a convolution operation with a filter size of 2 x 1.
5. The method for classifying vehicles based on the multi-branch local attention network as claimed in claim 3, wherein the step of introducing the local attention module based on space as the second branch comprises:
applying local space maximum pooling and average pooling in parallel to all channels of the original feature map, wherein the kernel and step size are equal to epsilon;
generating two compressed spatial attention descriptors F by aggregating features within all ε -neighborhoods of FMAnd FA；
FA＝AvgPoolkernel,stride＝ε(F)；
FM＝MaxPoolkernel,stride＝ε(F)；
Tightening F along the channel direction using global max pooling and global average poolingMAnd FAGenerate a descriptor sum
Figure FDA0003192131750000026
And
Figure FDA0003192131750000027
and are connected together;
sequentially increasing the convolution of the 3 multiplied by 3 holes and the nearest neighbor interpolation operation to obtain a characteristic diagram of space attention;
Figure FDA0003192131750000028
wherein MaxPoint and AvgPool represent the kernel and the local maximal pooling and average pooling operation with step size ε in the spatial domain, Fmax(. about.) and Favg(x) is the maximum pooling and average pooling operation along the channel direction, f (x) represents the convolution of 3 × 3 holes, fnearest() is the nearest neighbor interpolation upsampling operator and σ () is the sigmoid activation function.
6. The method for classifying the vehicle based on the multi-branch local attention network according to claim 1, wherein the step of performing classification prediction on the test set of the vehicle image according to the trained vehicle classification model to obtain the classification result comprises:
setting a loss function, adjusting the size of the vehicle image in the training set, inputting the vehicle image into a multi-branch local attention network for training to obtain a trained network model;
and (4) utilizing the extraction network full-link layer of the trained network model to classify and predict the vehicle images in the test set by using a Softmax function to obtain a classification result.
CN202110881344.8A 2021-08-02 2021-08-02 Vehicle classification method based on multi-branch local attention network Pending CN113610144A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110881344.8A CN113610144A (en) 2021-08-02 2021-08-02 Vehicle classification method based on multi-branch local attention network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110881344.8A CN113610144A (en) 2021-08-02 2021-08-02 Vehicle classification method based on multi-branch local attention network

Publications (1)

Publication Number Publication Date
CN113610144A true CN113610144A (en) 2021-11-05

Family

ID=78339077

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110881344.8A Pending CN113610144A (en) 2021-08-02 2021-08-02 Vehicle classification method based on multi-branch local attention network

Country Status (1)

Country Link
CN (1) CN113610144A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114005078A (en) * 2021-12-31 2022-02-01 山东交通学院 Vehicle weight identification method based on double-relation attention mechanism
CN114299542A (en) * 2021-12-29 2022-04-08 北京航空航天大学 Video pedestrian re-identification method based on multi-scale feature fusion
CN116058852A (en) * 2023-03-09 2023-05-05 同心智医科技(北京)有限公司 Classification system, method, electronic device and storage medium for MI-EEG signals
CN116563615A (en) * 2023-04-21 2023-08-08 南京讯思雅信息科技有限公司 Bad picture classification method based on improved multi-scale attention mechanism
CN117421984A (en) * 2023-11-02 2024-01-19 中国人民解放军海军航空大学 Short-term track prediction method based on multi-modal learning
CN117636057A (en) * 2023-12-13 2024-03-01 石家庄铁道大学 Train bearing damage classification and identification method based on multi-branch cross-space attention model
CN117710755A (en) * 2024-02-04 2024-03-15 江苏未来网络集团有限公司 Vehicle attribute identification system and method based on deep learning

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111242127A (en) * 2020-01-15 2020-06-05 上海应用技术大学 Vehicle detection method with granularity level multi-scale characteristics based on asymmetric convolution
CN111639692A (en) * 2020-05-25 2020-09-08 南京邮电大学 Shadow detection method based on attention mechanism
CN112069868A (en) * 2020-06-28 2020-12-11 南京信息工程大学 Unmanned aerial vehicle real-time vehicle detection method based on convolutional neural network
CN112200161A (en) * 2020-12-03 2021-01-08 北京电信易通信息技术股份有限公司 Face recognition detection method based on mixed attention mechanism
CN112307982A (en) * 2020-11-02 2021-02-02 西安电子科技大学 Human behavior recognition method based on staggered attention-enhancing network
CN113192633A (en) * 2021-05-24 2021-07-30 山西大学 Stomach cancer fine-grained classification method based on attention mechanism

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111242127A (en) * 2020-01-15 2020-06-05 上海应用技术大学 Vehicle detection method with granularity level multi-scale characteristics based on asymmetric convolution
CN111639692A (en) * 2020-05-25 2020-09-08 南京邮电大学 Shadow detection method based on attention mechanism
CN112069868A (en) * 2020-06-28 2020-12-11 南京信息工程大学 Unmanned aerial vehicle real-time vehicle detection method based on convolutional neural network
CN112307982A (en) * 2020-11-02 2021-02-02 西安电子科技大学 Human behavior recognition method based on staggered attention-enhancing network
CN112200161A (en) * 2020-12-03 2021-01-08 北京电信易通信息技术股份有限公司 Face recognition detection method based on mixed attention mechanism
CN113192633A (en) * 2021-05-24 2021-07-30 山西大学 Stomach cancer fine-grained classification method based on attention mechanism

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114299542A (en) * 2021-12-29 2022-04-08 北京航空航天大学 Video pedestrian re-identification method based on multi-scale feature fusion
CN114005078A (en) * 2021-12-31 2022-02-01 山东交通学院 Vehicle weight identification method based on double-relation attention mechanism
CN114005078B (en) * 2021-12-31 2022-03-29 山东交通学院 Vehicle weight identification method based on double-relation attention mechanism
CN116058852B (en) * 2023-03-09 2023-12-22 同心智医科技(北京)有限公司 Classification system, method, electronic device and storage medium for MI-EEG signals
CN116058852A (en) * 2023-03-09 2023-05-05 同心智医科技(北京)有限公司 Classification system, method, electronic device and storage medium for MI-EEG signals
CN116563615A (en) * 2023-04-21 2023-08-08 南京讯思雅信息科技有限公司 Bad picture classification method based on improved multi-scale attention mechanism
CN116563615B (en) * 2023-04-21 2023-11-07 南京讯思雅信息科技有限公司 Bad picture classification method based on improved multi-scale attention mechanism
CN117421984A (en) * 2023-11-02 2024-01-19 中国人民解放军海军航空大学 Short-term track prediction method based on multi-modal learning
CN117421984B (en) * 2023-11-02 2024-06-25 中国人民解放军海军航空大学 Short-term track prediction method based on multi-modal learning
CN117636057A (en) * 2023-12-13 2024-03-01 石家庄铁道大学 Train bearing damage classification and identification method based on multi-branch cross-space attention model
CN117636057B (en) * 2023-12-13 2024-06-11 石家庄铁道大学 Train bearing damage classification and identification method based on multi-branch cross-space attention model
CN117710755A (en) * 2024-02-04 2024-03-15 江苏未来网络集团有限公司 Vehicle attribute identification system and method based on deep learning
CN117710755B (en) * 2024-02-04 2024-05-03 江苏未来网络集团有限公司 Vehicle attribute identification system and method based on deep learning

Similar Documents

Publication Publication Date Title
CN113610144A (en) Vehicle classification method based on multi-branch local attention network
CN109584248B (en) Infrared target instance segmentation method based on feature fusion and dense connection network
CN108197326B (en) Vehicle retrieval method and device, electronic equipment and storage medium
CN108304795B (en) Human skeleton behavior identification method and device based on deep reinforcement learning
CN107609601B (en) Ship target identification method based on multilayer convolutional neural network
CN113936339B (en) Fighting identification method and device based on double-channel cross attention mechanism
CN106778595B (en) Method for detecting abnormal behaviors in crowd based on Gaussian mixture model
CN109977757B (en) Multi-modal head posture estimation method based on mixed depth regression network
CN111046821B (en) Video behavior recognition method and system and electronic equipment
CN110717526A (en) Unsupervised transfer learning method based on graph convolution network
CN113269224A (en) Scene image classification method, system and storage medium
CN112749675A (en) Potato disease identification method based on convolutional neural network
CN106530330B (en) Video target tracking method based on low-rank sparse
CN115393690A (en) Light neural network air-to-ground observation multi-target identification method
CN114782798A (en) Underwater target detection method based on attention fusion
CN116704431A (en) On-line monitoring system and method for water pollution
CN115205196A (en) No-reference image quality evaluation method based on twin network and feature fusion
CN113870160A (en) Point cloud data processing method based on converter neural network
CN113850182B (en) DAMR _ DNet-based action recognition method
CN114743126A (en) Lane line sign segmentation method based on graph attention machine mechanism network
CN118552800A (en) Rail foreign matter semi-supervised anomaly detection method and system deployed at edge end
CN111242870A (en) Low-light image enhancement method based on deep learning knowledge distillation technology
CN113011506B (en) Texture image classification method based on deep fractal spectrum network
CN112418227B (en) Monitoring video truck segmentation method based on double self-attention mechanism
CN117115616A (en) Real-time low-illumination image target detection method based on convolutional neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination