CN114662567A - Convolutional neural network image classification method based on different model feature fusion - Google Patents
Convolutional neural network image classification method based on different model feature fusion Download PDFInfo
- Publication number
- CN114662567A CN114662567A CN202210204664.4A CN202210204664A CN114662567A CN 114662567 A CN114662567 A CN 114662567A CN 202210204664 A CN202210204664 A CN 202210204664A CN 114662567 A CN114662567 A CN 114662567A
- Authority
- CN
- China
- Prior art keywords
- neural network
- feature
- crnet
- model
- image classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000004927 fusion Effects 0.000 title claims abstract description 31
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 19
- 238000013528 artificial neural network Methods 0.000 claims abstract description 49
- 238000012549 training Methods 0.000 claims abstract description 36
- 238000000605 extraction Methods 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000012546 transfer Methods 0.000 claims abstract description 7
- 238000012795 verification Methods 0.000 claims abstract description 6
- 238000012360 testing method Methods 0.000 claims abstract description 5
- 238000003062 neural network model Methods 0.000 claims description 47
- 239000011159 matrix material Substances 0.000 claims description 38
- 230000006870 function Effects 0.000 claims description 19
- 238000011176 pooling Methods 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 8
- 230000005540 biological transmission Effects 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 6
- 238000007500 overflow downdraw method Methods 0.000 claims description 3
- 230000006835 compression Effects 0.000 claims description 2
- 238000007906 compression Methods 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a convolutional neural network image classification method based on different model feature fusion, which comprises the following steps: preparing a data set, dividing the data set into a training set and a verification set, preprocessing the training set, constructing a cross neural network CRNet, setting a loss function and a network hyper-parameter, training the cross neural network CRNet, inputting the preprocessed test set into the trained cross neural network CRNet, and judging an image classification result; the cross neural network CRNet comprises a shallow layer feature extraction network, a different model feature transfer network and a prediction network. According to the method, the characteristic information is transmitted among the models, so that the characteristic reuse is realized, the loss of the key information of the characteristic is reduced, and the image classification precision is improved; meanwhile, redundant layers in the network are reduced, the number of parameters is reduced, the training speed is improved, and the practicability is high.
Description
Technical Field
The invention belongs to the technical field of image classification, and particularly relates to a convolutional neural network image classification method based on different model feature fusion.
Background
Computer vision technology is widely applied and mainly used for detecting and classifying images, and contents in the images are various objects on the ground. The image identification classification is a reference problem of image processing, and the improvement of the identification precision has great practical value in military application, medicine, life and other scenes.
With the great success of deep learning algorithms in machine vision in recent years, they have been considered as the first method of image processing. In the deep learning field, the performance of the neural network has a direct relation with the structure of the neural network model, the structure comprises the depth of the neural network and an internal path connection mode, namely a characteristic transmission path, theoretically, the depth of the neural network is increased, the complexity of the network is higher, and more accurate characteristics can be extracted, so that the performance of the neural network is improved. However, with the increase of the depth of the neural network, the problems of the gradient disappearance phenomenon, the neural network function degradation, the low training speed and the like become more obvious, and in order to solve the problems, the improvement of the feature transfer mode is particularly important.
The existing widely used deep neural network still has a lot of redundant information when performing feature extraction on input information, and when deep network features are transmitted, there is a possibility that part of detail information is discarded, and the performance of the algorithm is not sufficiently limited by feature extraction in the neural network training process. Therefore, in order to improve the efficiency of neural network image classification, it is urgent and necessary to find a convolutional neural network image classification method based on different model feature fusion to improve the image classification accuracy and speed from the aspect of feature reuse.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a convolutional neural network image classification method based on different model feature fusion. Preparing a data set, dividing the data set into a training set and a verification set, preprocessing the training set and the verification set, constructing a cross neural network CRNet, setting a loss function and a network hyper-parameter, training the cross neural network CRNet, inputting the preprocessed test set into the trained cross neural network CRNet, and judging an image classification result; the cross neural network CRNet comprises a shallow layer feature extraction network, a different model feature transfer network and a prediction network. According to the method, the characteristic information is transmitted among the models, so that the characteristic reuse is realized, the loss of the key information of the characteristic is reduced, and the image classification precision is improved; meanwhile, redundant layers in the network are reduced, the number of parameters is reduced, the training speed is improved, and the practicability is high.
The invention provides a convolutional neural network image classification method based on different model feature fusion, which comprises the following steps:
s1, preparing a data set: dividing a data set into a training set and a verification set, and carrying out preprocessing operation including normalization and order disorder on the data set;
s2, constructing a cross neural network CRNet: when different models are used for forward transmission of feature information, a feature fusion method is used for feature reuse; the cross neural network CRNet comprises a shallow layer feature extraction network, a different model feature transmission network and a prediction network; the shallow feature extraction network specifically comprises the following steps:
s211, inputting the preprocessed training set into a first convolution layer for shallow feature map extraction to obtain a first feature matrix;
s212, obtaining a second feature matrix by the first feature matrix through batch normalization and a first activation function;
s213, inputting the second feature matrix into the maximum pooling layer compression feature size to obtain a third feature matrix;
the different model feature transfer network specifically comprises the following steps:
s221, constructing a first neural network model and a second neural network model, wherein the input of the first neural network model and the input of the second neural network model are both the third feature matrix and both comprise a plurality of basic modules, and each basic module comprises a plurality of second convolution layers;
s222, establishing connection between the first neural network model and the second neural network model, and performing feature fusion;
and S223, performing feature fusion on the outputs of the first neural network model and the second neural network model to obtain a fourth feature matrix.
The prediction network specifically comprises the following steps:
s231, inputting the fourth feature matrix into a pooling layer for global weighted average pooling to obtain a fifth feature matrix, wherein the expression of the global weighted average pooling is as follows:
wherein,
represents a weighted average pixel value; p is a radical ofiRepresenting the probability occupied by the ith pixel value; a isiRepresents the ith pixel value; a isjRepresents traversing the jth pixel value; r is a real number;
s232, classifying the fifth feature matrix through a full connection layer by using a second activation function softmax, and outputting a prediction target image result;
s3, setting a loss function and a network hyper-parameter: setting a loss function, a learning rate, a minimum value of a batch and training times in a cross neural network CRNet;
s4, training a cross neural network CRNet: training the training set subjected to the preprocessing operation in the step S1 based on the cross neural network CRNet constructed in the step S2, and outputting a trained cross neural network CRNet;
s41, initializing neural network parameters and weights;
s42, performing forward propagation and backward propagation, and updating the weight matrix;
s43, judging whether the training times reach the maximum training times, if so, outputting a trained cross neural network CRNet, otherwise, executing a step S42;
and S5, inputting the test set preprocessed in the step S1 into a trained cross neural network CRNet, and judging the image classification result.
Further, in step S221, each of the first neural network model and the second neural network model has 8 basic modules, a structure of the first neural network model is {3,3,2}, a structure of the second neural network model is {2,2,2,2}, and each basic module has 2 second convolutional layers; in the step S222, the 3 rd basic module of the first neural network model and the 4 th basic module of the second neural network model perform feature matrix fusion, and the 2 nd basic module of the second neural network model and the 6 th basic module of the first neural network model perform feature matrix fusion.
Preferably, dropout operation is added to the second convolution layer in step S221 to prevent overfitting, the prediction target image result in step S232 is output in a one-hot manner, and initialization of the weights in step S41 is penalized by using L2 regularization to improve generalization capability of the model.
Preferably, the activation function in step S212 selects the modified linear unit ReLu, and the loss function in step S3 selects the cross entropy loss.
Preferably, the basic module in step S221 adopts a residual structure.
Preferably, the convolution kernel of the first convolution layer in step S211 is set to 7, the convolution kernel of the maximum pooling layer in step S213 is set to 3, and the convolution kernel of the second convolution layer in step S221 is set to 3.
Compared with the prior art, the invention has the technical effects that:
1. the convolutional neural network image classification method based on different model feature fusion is designed, feature reuse is realized through feature information transmission among models, loss of feature key information is reduced, and image classification precision is improved.
2. The convolutional neural network image classification method based on different model feature fusion uses relatively low network depth, reduces redundant layers in the network, accordingly reduces parameters, improves the speed of training the neural network, and has strong practicability.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, made with reference to the accompanying drawings.
FIG. 1 is a flow chart of the convolutional neural network image classification method based on different model feature fusion according to the present invention;
FIG. 2 is an exemplary diagram of a cross neural network CRNet of the present invention;
FIG. 3 is an exemplary diagram of the basic modules of the present invention;
FIG. 4 is a CIFAR10 image classification accuracy chart based on the cross neural network CRNet of the invention.
Detailed Description
The present application will be described in further detail with reference to the following drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not to be construed as limiting the invention. It should be noted that, for convenience of description, only the portions related to the related invention are shown in the drawings.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.
FIG. 1 shows a convolutional neural network image classification method based on different model feature fusion, which comprises the following steps:
s1, preparing a data set: and dividing the data set into a training set and a verification set, and carrying out preprocessing operation including normalization and disordering of the Image in the data set. In a particular embodiment, a CIFAR10 image classification dataset is selected as the study object.
S2, constructing a cross neural network CRNet, as shown in figure 2: when different models are used for forward transmission of feature information, a feature fusion method is used for feature reuse; the cross neural network CRNet comprises a shallow layer feature extraction network, a different model feature transfer network and a prediction network.
The shallow feature extraction network specifically comprises the following steps:
s211, inputting the preprocessed training set into a first convolution layer with convolution kernel of 7 to perform shallow feature map extraction, and obtaining a first feature matrix.
S212, the first feature matrix is subjected to batch normalization and first activation function correction linear unit ReLu, and a second feature matrix is obtained.
S213, inputting the second feature matrix into the largest pooling layer Pool with convolution kernel of 3 to compress the feature size, and obtaining a third feature matrix.
The different model feature transfer network specifically comprises the following steps:
s221, a first neural network model and a second neural network model are constructed, the input of the first neural network model and the input of the second neural network model are both third feature matrices and both comprise 8 basic modules blocks, the structure of the first neural network model is {3,3,2}, the structure of the second neural network model is {2,2,2,2}, each basic module Block comprises 2 second convolutional layers with convolution kernels of 3, a residual error structure is adopted, and as shown in FIG. 3, a dropout operation is added in each second convolutional layer to prevent overfitting.
S222, establishing connection between the first neural network model and the second neural network model, and performing feature fusion. And the 3 rd basic module Block of the first neural network model and the 4 th basic module Block of the second neural network model are subjected to feature matrix fusion, and the 2 nd basic module Block of the second neural network model and the 6 th basic module Block of the first neural network model are subjected to feature matrix fusion.
And S223, performing feature fusion on the outputs of the first neural network model and the second neural network model to obtain a fourth feature matrix.
The prediction network specifically comprises the following steps:
s231, inputting the fourth feature matrix into the pooling layer for global weighted average pooling wavg pool to obtain a fifth feature matrix, wherein an expression of the global weighted average pooling wavg pool is as follows:
wherein,
represents a weighted average pixel value; p is a radical ofiRepresenting the probability occupied by the ith pixel value; a isiRepresents the ith pixel value; a isjRepresents traversing the jth pixel value; r is a real number.
And S232, classifying the fifth feature matrix through the full connection layer Fc by using a second activation function softmax, outputting a predicted target image result, and outputting the result in a one-hot mode.
In a particular embodiment, the cross neural network CRNet is written based on the TensorFlow framework.
S3, setting a loss function and a network hyper-parameter: and setting a loss function, a learning rate, a minimum value of a batch and training times in the cross neural network CRNet, wherein the loss function selects cross entropy loss.
S4, training a cross neural network CRNet: training the training set subjected to the preprocessing operation in the step S1 based on the cross neural network CRNet constructed in the step S2, and outputting the trained cross neural network CRNet.
And S41, initializing neural network parameters and weights, and performing punishment on the initialization of the weights by utilizing L2 regularization so as to improve the generalization capability of the model.
And S42, carrying out forward propagation and backward propagation, and updating the weight matrix.
And S43, judging whether the training times reach the maximum training times, if so, outputting the trained cross neural network CRNet, otherwise, executing the step S42.
And S5, inputting the test set preprocessed in the step S1 into a trained cross neural network CRNet, and judging the image classification result.
In a specific embodiment, the accuracy of the trained CIFAR10 image classification is shown in fig. 4, which shows that the convolutional neural network image classification method based on different model feature fusion can improve the image classification accuracy.
According to the convolutional neural network image classification method based on different model feature fusion, feature reuse is realized through feature information transmission among models, loss of feature key information is reduced, and image classification precision is improved; the relatively low network depth is used, the redundant layers in the network are reduced, the parameter quantity is reduced, the speed of training the neural network is improved, and the practicability is high.
Finally, it should be noted that: although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that: modifications and equivalents may be made thereto without departing from the spirit and scope of the invention and it is intended to cover in the claims the invention as defined in the appended claims.
Claims (6)
1. A convolutional neural network image classification method based on different model feature fusion is characterized by comprising the following steps:
s1, preparing a data set: dividing a data set into a training set and a verification set, and carrying out preprocessing operations including normalization and disordering on the data set;
s2, constructing a cross neural network CRNet: when different models are used for forward transmission of feature information, a feature fusion method is used for feature reuse; the cross neural network CRNet comprises a shallow layer feature extraction network, a different model feature transmission network and a prediction network; the shallow feature extraction network specifically comprises the following steps:
s211, inputting the preprocessed training set into a first convolution layer for shallow feature map extraction to obtain a first feature matrix;
s212, obtaining a second feature matrix by the first feature matrix through batch normalization and an activation function;
s213, inputting the second feature matrix into the maximum pooling layer compression feature size to obtain a third feature matrix;
the different model feature transfer network specifically comprises the following steps:
s221, constructing a first neural network model and a second neural network model, wherein the input of the first neural network model and the input of the second neural network model are both the third feature matrix and both comprise a plurality of basic modules, and each basic module comprises a plurality of second convolution layers;
s222, establishing connection between the first neural network model and the second neural network model, and performing feature fusion;
and S223, performing feature fusion on the outputs of the first neural network model and the second neural network model to obtain a fourth feature matrix.
The prediction network specifically comprises the following steps:
s231, inputting the fourth feature matrix into a pooling layer for global weighted average pooling to obtain a fifth feature matrix, wherein the expression of the global weighted average pooling is as follows:
wherein,
represents a weighted average pixel value; p is a radical ofiRepresenting the probability occupied by the ith pixel value; a isiRepresents the ith pixel value; a isjRepresents traversing the jth pixel value; r is a real number;
s232, classifying the fifth feature matrix through a full connection layer by using a second activation function softmax, and outputting a prediction target image result;
s3, setting a loss function and a network hyper-parameter: setting a loss function, a learning rate, a minimum value of a batch and training times in a cross neural network CRNet;
s4, training a cross neural network CRNet: training the training set subjected to the preprocessing operation in the step S1 based on the cross neural network CRNet constructed in the step S2, and outputting a trained cross neural network CRNet;
s41, initializing neural network parameters and weights;
s42, performing forward propagation and backward propagation, and updating the weight matrix;
s43, judging whether the training times reach the maximum training times, if so, outputting a trained cross neural network CRNet, otherwise, executing a step S42;
and S5, inputting the test set preprocessed in the step S1 into a trained cross neural network CRNet, and judging the image classification result.
2. The convolutional neural network image classification method based on different model feature fusion of claim 1, wherein in step S221, each of the first neural network model and the second neural network model has 8 basic modules, the structure of the first neural network model is {3,3,2}, the structure of the second neural network model is {2,2,2,2}, and each basic module has 2 second convolutional layers; in the step S222, the 3 rd basic module of the first neural network model and the 4 th basic module of the second neural network model perform feature matrix fusion, and the 2 nd basic module of the second neural network model and the 6 th basic module of the first neural network model perform feature matrix fusion.
3. The convolutional neural network image classification method based on different model feature fusion as claimed in claim 1, wherein dropout operation is added to the second convolutional layer in step S221 to prevent overfitting, the prediction target image result in step S232 is output in a one-hot manner, and initialization of the weights in step S41 is punished by using L2 regularization to improve generalization capability of the model.
4. The method for classifying the image of the convolutional neural network based on fusion of different model features of claim 1, wherein the activation function in step S212 selects a modified linear unit ReLu, and the loss function in step S3 selects cross entropy loss.
5. The convolutional neural network image classification method based on different model feature fusion of claims 1 and 2, wherein the basic module in the step S221 adopts a residual structure.
6. The convolutional neural network image classification method based on different model feature fusion as claimed in claims 1 and 2, wherein the convolution kernel of the first convolutional layer in step S211 is set to 7, the convolution kernel of the maximum pooling layer in step S213 is set to 3, and the convolution kernel of the second convolutional layer in step S221 is set to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210204664.4A CN114662567A (en) | 2022-03-03 | 2022-03-03 | Convolutional neural network image classification method based on different model feature fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210204664.4A CN114662567A (en) | 2022-03-03 | 2022-03-03 | Convolutional neural network image classification method based on different model feature fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114662567A true CN114662567A (en) | 2022-06-24 |
Family
ID=82027206
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210204664.4A Pending CN114662567A (en) | 2022-03-03 | 2022-03-03 | Convolutional neural network image classification method based on different model feature fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114662567A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115049885A (en) * | 2022-08-16 | 2022-09-13 | 之江实验室 | Storage and calculation integrated convolutional neural network image classification device and method |
| CN117853926A (en) * | 2024-01-17 | 2024-04-09 | 南京北斗创新应用科技研究院有限公司 | Building detection method and system based on artificial neural network classification |
-
2022
- 2022-03-03 CN CN202210204664.4A patent/CN114662567A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115049885A (en) * | 2022-08-16 | 2022-09-13 | 之江实验室 | Storage and calculation integrated convolutional neural network image classification device and method |
| CN115049885B (en) * | 2022-08-16 | 2022-12-27 | 之江实验室 | Storage and calculation integrated convolutional neural network image classification device and method |
| CN117853926A (en) * | 2024-01-17 | 2024-04-09 | 南京北斗创新应用科技研究院有限公司 | Building detection method and system based on artificial neural network classification |
| CN117853926B (en) * | 2024-01-17 | 2024-07-02 | 南京北斗创新应用科技研究院有限公司 | Building detection method and system based on artificial neural network classification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111126488B (en) | Dual-attention-based image recognition method | |
| CN112541409B (en) | Attention-integrated residual network expression recognition method | |
| CN114897779B (en) | Cervical cytology image abnormal region positioning method and device based on fusion attention | |
| CN109063719B (en) | Image classification method combining structure similarity and class information | |
| CN114662567A (en) | Convolutional neural network image classification method based on different model feature fusion | |
| US12039440B2 (en) | Image classification method and apparatus, and image classification model training method and apparatus | |
| CN111612008A (en) | Image segmentation method based on convolution network | |
| CN111008639B (en) | License plate character recognition method based on attention mechanism | |
| CN114358188A (en) | Feature extraction model processing method, feature extraction model processing device, sample retrieval method, sample retrieval device and computer equipment | |
| CN116468740A (en) | Image semantic segmentation model and segmentation method | |
| CN116580322A (en) | Unmanned aerial vehicle infrared small target detection method under ground background | |
| Luan et al. | Sunflower seed sorting based on convolutional neural network | |
| WO2022127333A1 (en) | Training method and apparatus for image segmentation model, image segmentation method and apparatus, and device | |
| CN117237704A (en) | Multi-label image classification method based on two-dimensional dependence | |
| CN112132207A (en) | Target detection neural network construction method based on multi-branch feature mapping | |
| CN116758610A (en) | Attention mechanism and feature fusion-based light-weight human ear recognition method and system | |
| CN115861595A (en) | Multi-scale domain self-adaptive heterogeneous image matching method based on deep learning | |
| CN115527253A (en) | Attention mechanism-based lightweight facial expression recognition method and system | |
| CN111652079B (en) | Expression recognition method and system applied to mobile crowd and storage medium | |
| CN114648760A (en) | Image segmentation method, image segmentation device, electronic device, and storage medium | |
| CN114120406A (en) | Face feature extraction and classification method based on convolutional neural network | |
| CN113283530A (en) | Image classification system based on cascade characteristic blocks | |
| CN112364193A (en) | Image retrieval-oriented method for fusing multilayer characteristic deep neural network model | |
| Xu et al. | Research and verification of convolutional neural network lightweight in BCI | |
| Song et al. | Lightweight small target detection algorithm based on YOLOv5 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |