CN109165675A - Image classification method based on periodically part connection convolutional neural networks - Google Patents
Image classification method based on periodically part connection convolutional neural networks Download PDFInfo
- Publication number
- CN109165675A CN109165675A CN201810831075.2A CN201810831075A CN109165675A CN 109165675 A CN109165675 A CN 109165675A CN 201810831075 A CN201810831075 A CN 201810831075A CN 109165675 A CN109165675 A CN 109165675A
- Authority
- CN
- China
- Prior art keywords
- convolution
- layer
- local connection
- convolutional neural
- periodic local
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000000605 extraction Methods 0.000 claims abstract description 30
- 230000000737 periodic effect Effects 0.000 claims description 75
- 230000004913 activation Effects 0.000 claims description 10
- 230000004927 fusion Effects 0.000 claims description 9
- 238000010586 diagram Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 101100153581 Bacillus anthracis topX gene Proteins 0.000 description 1
- 101150041570 TOP1 gene Proteins 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
-
- 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
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 present invention relates to a kind of image classification methods based on periodically part connection convolutional neural networks, comprising: multiple convolution kernel groups are configured to periodically part connection convolutional layer;Wherein, each convolution kernel group includes multiple convolution kernels, and each convolution kernel is used to carry out periodical convolution operation to the specified region of pre-set image;The periodically local connection convolutional neural networks according to the periodically part connection convolution layer building;According to periodically part connection convolutional neural networks the classifying to original image.Image-recognizing method provided in this embodiment improves image characteristics extraction efficiency using periodically part connection convolutional neural networks, effectively reduce the scale of network structure, so that image characteristics extraction is insensitive to the position of picture material and angle change, therefore, has stronger ability to express, to have higher image classification accuracy.
Description
Technical Field
The invention belongs to the field of digital image processing, and particularly relates to an image classification method based on a periodic local connection convolutional neural network.
Background
Image classification is one of the fundamental problems in the field of digital image processing, and is an image processing method that distinguishes different classes of objects according to different features reflected in image content. Specifically, the computer is used for automatically understanding the content of the image, determining the label of the image according to the content, and finally, automatically classifying the image according to the label. The image classification has wide application scenes, such as target detection and identification, scene classification and the like, which are the most representative extended applications in the image classification technology.
The image classification can be generally divided into two key steps of image feature extraction and feature classification, and the common image classification method is to classify images by constructing a feature classifier, and the image classifier is mainly specifically realized by a convolutional neural network. The convolutional neural network automatically learns and classifies potential features in the image data by utilizing an algorithm. At present, VGGNet, ResNet, Densenet, MobilenetV1 and MobilenetV2 are several representative neural network structures with optimal performance, wherein VGGNet, ResNet and Densenet well classify images through image feature extraction; and the network structure is improved by MobileneetV 1 and MobileneetV 2, so that the calculation amount of the algorithm is greatly reduced.
The network structures of VGGNet, ResNet and Densenet are large in scale, and the huge calculation amount makes the network structures unusable in many occasions. While the network structure is improved by MobilenetV1 and MobilenetV2, the amount of calculation is greatly reduced, but the classification accuracy is also reduced.
Disclosure of Invention
In order to solve the above problems in the prior art, the present invention provides an image classification method based on a periodic local connection convolutional neural network. The technical problem to be solved by the invention is realized by the following technical scheme:
the embodiment of the invention provides an image classification method based on a periodic local connection convolutional neural network, which is characterized by comprising the following steps:
constructing a plurality of convolution kernel groups into periodic local connection convolution layers; each convolution kernel group comprises a plurality of convolution kernels, and each convolution kernel is used for performing periodic convolution operation on a specified area of a preset image;
constructing the periodic local connection convolutional neural network according to the periodic local connection convolutional layer;
and classifying the original image according to the periodic local connection convolutional neural network.
In one embodiment of the present invention, the size of the convolution kernel is W × H, W being the width of the convolution kernel, and H being the height of the convolution kernel.
In one embodiment of the present invention, constructing the periodically locally connected convolutional neural network from the periodically locally connected convolutional layer comprises:
constructing a periodic local connection convolution module according to the periodic local connection convolution layer;
constructing a basic feature extraction unit according to the periodic local connection convolution module;
constructing a plurality of basic feature extraction units into a periodic local connection convolutional neural network.
In one embodiment of the present invention, constructing a periodically locally connected convolution module from the periodically locally connected convolution layer includes:
connecting the output end of the first coiling layer with the input end of the first active layer;
connecting an output end of the first active layer to an input end of the periodic local connection convolution layer;
connecting the output end of the periodic local connection convolution layer with the input end of a second active layer;
connecting the output end of the second active layer with the input end of a second convolution layer to construct the periodic local connection convolution module; wherein,
the first convolution layer is used for carrying out weighted fusion on the characteristic images of different channels in the first convolution layer;
the first activation layer is used for increasing sparsity of a feature map;
the periodic local connection convolution layer is used for carrying out weighted fusion on feature maps of the same channel in the periodic local connection convolution layer;
the second activation layer is used for increasing sparsity of the feature map;
the second convolution layer is used for carrying out weighted fusion on the characteristic images of different channels.
In an embodiment of the present invention, an input end of the periodic local connection convolution module is an input end of the first convolution layer, and an output end of the periodic local connection convolution module is an output end of the second convolution layer.
In one embodiment of the present invention, the first active layer and the second active layer are both ReLU active layers.
In an embodiment of the present invention, the number of cores of the first convolutional layer convolution kernel is O1Step value of S1Edge filling of P1;
The number of convolution kernels of the second convolution layer is O2Step value of S2Edge filling of P2。
In an embodiment of the present invention, constructing the basic feature extraction unit according to the periodically locally connected convolution module includes:
connecting the input end and the output end of the periodic local connection convolution module by using an addition bypass to construct the basic feature extraction unit; wherein the addition bypass is for point-to-point adding the input and the output.
In an embodiment of the present invention, after constructing the periodically locally connected convolutional neural network according to the periodically locally connected convolutional layer, the method further includes:
and training the periodic local connection convolutional neural network through a training sample set.
In one embodiment of the invention, the training sample set is a CIFAR-10 dataset.
Compared with the prior art, the invention has the beneficial effects that:
the method provided by the invention improves the image feature extraction efficiency by utilizing the periodic local connection convolutional neural network, can extract more effective image features from the convolutional neural network with the same scale, and effectively reduces the scale of the network structure; due to the periodic characteristic, the image feature extraction is insensitive to the position and angle change of the image content, so that the method has stronger expression capability and higher image classification accuracy.
Drawings
FIG. 1 is a schematic flowchart of an image classification method based on a periodically locally connected convolutional neural network according to the present invention;
FIG. 2 is a schematic structural diagram of a convolution module with periodic local connections according to the present invention;
FIG. 3 is a schematic diagram of a basic feature extraction unit according to the present invention;
fig. 4 is a schematic structural diagram of a periodic local connection convolutional neural network provided in the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
Example one
Referring to fig. 1, fig. 2, fig. 3 and fig. 4, fig. 1 is a schematic flowchart of an image classification method based on a periodically locally connected convolutional neural network according to the present invention; FIG. 2 is a schematic structural diagram of a convolution module with periodic local connections according to the present invention; FIG. 3 is a schematic diagram of a basic feature extraction unit according to the present invention; fig. 4 is a schematic structural diagram of a periodic local connection convolutional neural network provided in the present invention.
As shown in fig. 1, an image classification method based on a periodic local connection convolutional neural network includes:
step 1, a periodic local connection convolution layer is built according to a convolution kernel set.
Specifically, constructing a periodic locally connected convolutional layer from a convolutional kernel set includes:
establishing a convolution kernel periodically connected with the convolution layer, wherein the convolution kernel periodically performs convolution operation in a specified area of a preset image;
constructing a convolution kernel group through a convolution kernel, wherein the convolution kernel group periodically performs convolution operation in a preset image;
and copying the convolution kernel group to enable the convolution kernel group to be fully paved on the whole preset image (non-overlapping and fully paved) during convolution operation, so that the periodic local connection convolution layer is constructed.
In this embodiment, the convolution operation of periodically and locally connecting convolution layers is a completely new convolution process. Unlike conventional convolution operations (where the convolution kernel traverses the entire preset image for sliding convolution operations), each convolution kernel periodically and locally connected to a convolution layer only operates in a specific region of the image.
The convolution kernel group is a square formed by M convolution kernels, wherein M is a natural number which is larger than or equal to 1. Since the convolution kernel group is spread over the entire preset image, a convolution operation can be performed on each pixel of the preset pattern.
In this embodiment, the size W × H of convolution kernels of the periodically locally connected convolution layers is set as the number of convolution kernels of the periodically locally connected convolution layers is O, the advance value is S, the edge padding is P, and the period is L, where W is the width of the convolution kernel and H is the height of the convolution kernel.
Preferably, W × H is 3 × 3, O is 32, S is 1, P is 1, and L is 3.
In practical application, the size of the convolution kernel, the number of the convolution kernels, the step value and the period size may also be other values, and the specific values may be set according to actual requirements.
And 2, constructing the periodic local connection convolutional neural network according to the periodic local connection convolutional layer.
And step 21, constructing a periodic local connection convolution module according to the periodic local connection convolution layer.
As shown in fig. 2, the constructing the periodic local connection convolution module specifically includes:
connecting the output end of the first coiling layer with the input end of the first active layer;
connecting the output end of the first active layer with the input end of the periodic local connection convolution layer;
connecting the output end of the periodic local connection convolution layer with the input end of the second active layer;
the output end of the second active layer is connected with the input end of the second convolution layer.
In this embodiment, the first convolution layer is used to perform weighted fusion on the feature images of different channels of the convolution layer. The first activation layer is used for increasing the sparsity of the feature map, namely, the sparsity of the feature map input into the first activation layer is increased, so that the convergence process is accelerated. The periodically locally connected convolutional layers are used for performing weighted fusion on feature maps in the same channel in the convolutional layers. The second activation layer is used for increasing sparsity of the feature map; the second convolution layer is used for carrying out weighted fusion on the characteristic images of different channels.
The convolution kernel of the first convolution layer has a size of W1×H1With a number of convolution kernels of O1Step value of S1With an edge filling of P1。
The convolution kernel of the second convolution layer has a size of W2×H2With a number of convolution kernels of O2Step value of S2With an edge filling of P2。
Preferably, the size W of the convolution kernel of the first convolution layer1×H1=1×1,O1=32,S1=1,P11. Convolution kernel size W of second convolution layer2×H2=1×1,O2=32,S2=1,P2=1。
Preferably, the first active layer and the second active layer are both ReLU (Rectified Linear Unit, for short, ReLU) active layers, and the mathematical expression thereof is as follows:
f(x)=max(0,x),
where x is the output of the convolutional layer.
And step 22, constructing a basic feature extraction unit according to the periodic local connection convolution module.
In this embodiment, the input end of the periodic local connection convolution module is the input end of the first convolution layer, and the output end of the periodic local connection convolution module is the output end of the second convolution layer.
As shown in fig. 3, an addition bypass is constructed on the basis of a periodic local connection convolution module;
and carrying out point-to-point addition on the input end and the output end of the periodic local connection convolution module by using bypass addition, and taking the addition result as a final output result, thereby constructing the basic feature extraction unit.
Step 23, constructing a plurality of basic feature extraction units into a periodic local connection convolutional neural network;
as shown in fig. 4, according to the basic feature extraction unit constructed in step 22, n basic feature extraction units are constructed, and these basic feature extraction units are connected in sequence, so as to construct a periodic local connection convolutional neural network, where n is a natural number greater than or equal to 1.
The basic feature extraction units are connected in sequence, namely the output of the kth basic feature extraction unit is connected with the input of the (k + 1) th basic feature extraction unit, wherein the sum k is a natural number which is more than or equal to 1, and k is less than or equal to n.
In this embodiment, an input end of the periodically locally connected convolutional neural network is an input end of a first basic feature extraction unit, and an output end of the periodically locally connected convolutional neural network is an output end of a last basic feature extraction unit.
And 24, training the periodic local connection convolutional neural network through a training sample set.
Training the periodic local connection convolutional neural network through a training sample set refers to a process of adjusting parameters of the neural network by using the training sample set to enable the neural network to achieve required performance. I.e. to enable the convolutional neural network to accurately classify the original image.
Preferably, after the periodic local connection convolutional neural network is trained, the convolutional neural network can also be tested through a test sample set to evaluate the performance of the convolutional neural network.
Training a periodic locally connected convolutional neural network, comprising:
firstly, initializing the convolutional neural network, namely randomly initializing the convolutional core of each convolutional layer of the convolutional neural network;
then, selecting a training sample set;
preferably, the training sample set is a CIFAR-10 dataset.
After the training sample set is selected, the convolutional neural network is trained by utilizing the training sample set, so that the construction of the convolutional neural network is completed.
The specific training method comprises the following steps: using an existing SGD (Stochastic gradient descent) optimizer, the batch size is set to 64, and the training is performed for 60 rounds at a learning rate of 0.01, followed by 30 rounds at a learning rate of 0.001, for a total of 90 rounds.
And 3, classifying the original image according to the periodic local connection convolutional neural network.
The method for classifying the original image by using the periodic local connection convolutional neural network trained in the step 2 specifically comprises the following steps:
processing the pre-classified original image by using the convolutional neural network, and extracting the characteristics of the original image;
and according to the extracted features, obtaining a classification result of the original image according to a preset image classification rule.
The following is a specific comparison between the image recognition method provided by the present invention and the conventional image recognition method, by way of example.
Firstly, a mobilenetv2 convolutional neural network (the optimal method in the current lightweight image classification method) is adopted to test the data sets of the cifar-10, the imagenet-50(64) and the imagenet-50(128) respectively. Wherein imagenet-50(64) indicates that 50 categories of data sets with image resolution of 64 × 64 are randomly sampled from the imagenet data, and imagenet-50(128) indicates that 50 categories of data sets with image resolution of 128 × 128 are randomly sampled from the imagenet data.
Then, the data sets of cifar-10, imagenet-50(64), and imagenet-50(128) were tested using the periodic local connectivity convolutional neural network provided in this example (basic feature extraction unit was stacked 7 times, i.e., n ═ 7).
And finally, under the same experimental environment, the same data preprocessing method is adopted to obtain two methods for testing the image classification accuracy of the three data sets, so that the performance of the method is evaluated. Wherein, the accuracy of image classification is shown in the following table:
top1 in the above table indicates the classification accuracy, and Madds indicates the number of multiply-add required by the algorithm. As can be seen from the data in the table: according to the image identification method based on the periodic local connection convolutional neural network, on three data sets, better classification accuracy is obtained than that of the method adopting the Mobilenetv2 convolutional neural network, and accidental factors on different data are eliminated. Meanwhile, the calculated amount required by the method is consistent with that of the Mobilenetv2, and the image recognition method provided by the embodiment has a better effect.
The image recognition method provided by the embodiment improves the image feature extraction efficiency by utilizing the periodic local connection convolutional neural network, can extract more effective image features from the convolutional neural network with the same scale, and effectively reduces the scale of the network structure; due to the periodic characteristic, the image feature extraction is insensitive to the position and angle change of the image content, so that the image identification method provided by the embodiment has stronger expression capability and higher image classification accuracy.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (10)
1. An image classification method based on a periodic local connection convolutional neural network is characterized by comprising the following steps:
constructing a plurality of convolution kernel groups into periodic local connection convolution layers; each convolution kernel group comprises a plurality of convolution kernels, and each convolution kernel is used for performing periodic convolution operation on a specified area of a preset image;
constructing the periodic local connection convolutional neural network according to the periodic local connection convolutional layer;
and classifying the original image according to the periodic local connection convolutional neural network.
2. The method of claim 1, wherein the size of the convolution kernel is wxh, W being the width of the convolution kernel, and H being the height of the convolution kernel.
3. The method of claim 1, wherein constructing the periodically locally connected convolutional neural network from the periodically locally connected convolutional layers comprises:
constructing a periodic local connection convolution module according to the periodic local connection convolution layer;
constructing a basic feature extraction unit according to the periodic local connection convolution module;
constructing a plurality of basic feature extraction units into a periodic local connection convolutional neural network.
4. The method of claim 3, wherein constructing a periodically locally connected convolution module from the periodically locally connected convolution layer comprises:
connecting the output end of the first coiling layer with the input end of the first active layer;
connecting an output end of the first active layer to an input end of the periodic local connection convolution layer;
connecting the output end of the periodic local connection convolution layer with the input end of a second active layer;
connecting the output end of the second active layer with the input end of a second convolution layer to construct the periodic local connection convolution module; wherein,
the first convolution layer is used for carrying out weighted fusion on the characteristic images of different channels in the first convolution layer;
the first activation layer is used for increasing sparsity of a feature map;
the periodic local connection convolution layer is used for carrying out weighted fusion on feature maps of the same channel in the periodic local connection convolution layer;
the second activation layer is used for increasing sparsity of the feature map;
the second convolution layer is used for carrying out weighted fusion on the characteristic images of different channels.
5. The method of claim 4, wherein an input of the periodic local connection convolution module is an input of the first convolutional layer and an output of the periodic local connection convolution module is an output of the second convolutional layer.
6. The method of claim 4, wherein the first activation layer and the second activation layer are both ReLU activation layers.
7. The method of claim 4, wherein the first convolutional core has a core number of O1Step value of S1Edge filling of P1;
The number of convolution kernels of the second convolution layer is O2Step value of S2Edge filling of P2。
8. The method of claim 3, wherein constructing a basic feature extraction unit from the periodically locally connected convolution module comprises:
connecting the input end and the output end of the periodic local connection convolution module by using an addition bypass to construct the basic feature extraction unit; wherein the addition bypass is for point-to-point adding the input and the output.
9. The method of claim 1, further comprising, after constructing the periodically locally connected convolutional neural network from the periodically locally connected convolutional layer:
and training the periodic local connection convolutional neural network through a training sample set.
10. The method of claim 9, wherein the training sample set is a CIFAR-10 dataset.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810831075.2A CN109165675A (en) | 2018-07-26 | 2018-07-26 | Image classification method based on periodically part connection convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810831075.2A CN109165675A (en) | 2018-07-26 | 2018-07-26 | Image classification method based on periodically part connection convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109165675A true CN109165675A (en) | 2019-01-08 |
Family
ID=64898187
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810831075.2A Pending CN109165675A (en) | 2018-07-26 | 2018-07-26 | Image classification method based on periodically part connection convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109165675A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111950572A (en) * | 2019-05-14 | 2020-11-17 | 北京字节跳动网络技术有限公司 | Method, apparatus, electronic device and computer-readable storage medium for training classifier |
WO2021018251A1 (en) * | 2019-07-30 | 2021-02-04 | 华为技术有限公司 | Image classification method and device |
CN113052189A (en) * | 2021-03-30 | 2021-06-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105138993A (en) * | 2015-08-31 | 2015-12-09 | 小米科技有限责任公司 | Method and device for building face recognition model |
CN105224951A (en) * | 2015-09-30 | 2016-01-06 | 深圳市华尊科技股份有限公司 | A kind of vehicle type classification method and sorter |
CN105630882A (en) * | 2015-12-18 | 2016-06-01 | 哈尔滨工业大学深圳研究生院 | Remote sensing data deep learning based offshore pollutant identifying and tracking method |
CN106874898A (en) * | 2017-04-08 | 2017-06-20 | 复旦大学 | Extensive face identification method based on depth convolutional neural networks model |
CN107064845A (en) * | 2017-06-06 | 2017-08-18 | 深圳先进技术研究院 | One-dimensional division Fourier's parallel MR imaging method based on depth convolution net |
CN107909016A (en) * | 2017-11-03 | 2018-04-13 | 车智互联(北京)科技有限公司 | A kind of convolutional neural networks generation method and the recognition methods of car system |
CN108052884A (en) * | 2017-12-01 | 2018-05-18 | 华南理工大学 | A kind of gesture identification method based on improvement residual error neutral net |
-
2018
- 2018-07-26 CN CN201810831075.2A patent/CN109165675A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105138993A (en) * | 2015-08-31 | 2015-12-09 | 小米科技有限责任公司 | Method and device for building face recognition model |
CN105224951A (en) * | 2015-09-30 | 2016-01-06 | 深圳市华尊科技股份有限公司 | A kind of vehicle type classification method and sorter |
CN105630882A (en) * | 2015-12-18 | 2016-06-01 | 哈尔滨工业大学深圳研究生院 | Remote sensing data deep learning based offshore pollutant identifying and tracking method |
CN106874898A (en) * | 2017-04-08 | 2017-06-20 | 复旦大学 | Extensive face identification method based on depth convolutional neural networks model |
CN107064845A (en) * | 2017-06-06 | 2017-08-18 | 深圳先进技术研究院 | One-dimensional division Fourier's parallel MR imaging method based on depth convolution net |
CN107909016A (en) * | 2017-11-03 | 2018-04-13 | 车智互联(北京)科技有限公司 | A kind of convolutional neural networks generation method and the recognition methods of car system |
CN108052884A (en) * | 2017-12-01 | 2018-05-18 | 华南理工大学 | A kind of gesture identification method based on improvement residual error neutral net |
Non-Patent Citations (1)
Title |
---|
高志强 等: "《深度学习:从入门到实战》", 30 June 2018 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111950572A (en) * | 2019-05-14 | 2020-11-17 | 北京字节跳动网络技术有限公司 | Method, apparatus, electronic device and computer-readable storage medium for training classifier |
WO2021018251A1 (en) * | 2019-07-30 | 2021-02-04 | 华为技术有限公司 | Image classification method and device |
CN113052189A (en) * | 2021-03-30 | 2021-06-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
CN113052189B (en) * | 2021-03-30 | 2022-04-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
CN110334765B (en) | Remote sensing image classification method based on attention mechanism multi-scale deep learning | |
Fu et al. | Fast crowd density estimation with convolutional neural networks | |
Thai et al. | Image classification using support vector machine and artificial neural network | |
CN113065558A (en) | Lightweight small target detection method combined with attention mechanism | |
CN109614921B (en) | Cell segmentation method based on semi-supervised learning of confrontation generation network | |
CN108960404B (en) | Image-based crowd counting method and device | |
CN105354581B (en) | The color image feature extracting method of Fusion of Color feature and convolutional neural networks | |
CN113674334B (en) | Texture recognition method based on depth self-attention network and local feature coding | |
CN111160249A (en) | Multi-class target detection method of optical remote sensing image based on cross-scale feature fusion | |
CN111222519B (en) | Construction method, method and device of hierarchical colored drawing manuscript line extraction model | |
CN106096655B (en) | A kind of remote sensing image airplane detection method based on convolutional neural networks | |
CN109766873B (en) | Pedestrian re-identification method based on hybrid deformable convolution | |
CN104537647A (en) | Target detection method and device | |
CN111325165A (en) | Urban remote sensing image scene classification method considering spatial relationship information | |
CN105046272B (en) | A kind of image classification method based on succinct non-supervisory formula convolutional network | |
CN112132145B (en) | Image classification method and system based on model extended convolutional neural network | |
CN107679539B (en) | Single convolution neural network local information and global information integration method based on local perception field | |
CN112633382A (en) | Mutual-neighbor-based few-sample image classification method and system | |
CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
CN109165675A (en) | Image classification method based on periodically part connection convolutional neural networks | |
CN109002771A (en) | A kind of Classifying Method in Remote Sensing Image based on recurrent neural network | |
CN111898614B (en) | Neural network system and image signal and data processing method | |
WO2020119624A1 (en) | Class-sensitive edge detection method based on deep learning | |
CN107133579A (en) | Based on CSGF (2D)2The face identification method of PCANet convolutional networks |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190108 |