CN110378484A - A kind of empty spatial convolution pyramid pond context learning method based on attention mechanism - Google Patents
A kind of empty spatial convolution pyramid pond context learning method based on attention mechanism Download PDFInfo
- Publication number
- CN110378484A CN110378484A CN201910351669.8A CN201910351669A CN110378484A CN 110378484 A CN110378484 A CN 110378484A CN 201910351669 A CN201910351669 A CN 201910351669A CN 110378484 A CN110378484 A CN 110378484A
- Authority
- CN
- China
- Prior art keywords
- model
- empty
- pyramid pond
- spatial convolution
- attention
- 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
- 238000000034 method Methods 0.000 title claims abstract description 11
- 230000007246 mechanism Effects 0.000 title claims abstract description 10
- 230000007480 spreading Effects 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims abstract description 4
- 230000004913 activation Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims 1
- 238000001514 detection method Methods 0.000 abstract 1
- 238000003062 neural network model Methods 0.000 abstract 1
- 230000001737 promoting effect Effects 0.000 abstract 1
- 230000011218 segmentation Effects 0.000 abstract 1
- 238000013135 deep learning Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 239000010931 gold Substances 0.000 description 1
- 229910052737 gold Inorganic materials 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification 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
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Mathematical Optimization (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Medical Informatics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
Abstract
The empty spatial convolution pyramid pond context learning method based on attention mechanism that the invention discloses a kind of, be characterized in include: empty spatial convolution pyramid pond model and attention model.The cavity spatial convolution pyramid pond model is made of a plurality of empty convolution path in parallel with different spreading rates, for extracting multiple dimensioned contextual information.The attention model characterizes the relationship between different channel contextual informations by nonlinear function, and then distributes weight to the multi-scale information in each channel.The contextual feature learning ability of empty spatial convolution pyramid pond model can be enhanced in empty spatial convolution pyramid pond context learning method based on attention mechanism of the invention, and can flexibly it be embedded into neural network model, suitable for multiple-tasks such as image, semantic segmentation, target detection, image classifications, it is suitable for promoting and applying.
Description
Technical field
The invention belongs to deep learning field, in particular to a kind of empty spatial convolution pyramid based on attention mechanism
Pond context learning method.
Background technique
Empty spatial convolution pyramid pond model is extracted multiple dimensioned by the empty convolution of multiple and different spreading rates in parallel
Then contextual information carries out linear fusion to each channel using 1 × 1 convolution.However, since multiple dimensioned contextual information is logical
It is normally present in non-linearity manifold, only is not enough to portray the nonlinear dependence between multiple dimensioned contextual information using linear function
System, causes empty spatial convolution pyramid pond model that can not effectively extract multiple dimensioned contextual information.
Summary of the invention
In order to overcome above-mentioned empty spatial convolution pyramid pond model to be difficult to characterize asking for different channel non-linearities relationships
Topic, the present invention provide a kind of empty spatial convolution pyramid pond context learning method based on attention mechanism.
A kind of empty spatial convolution pyramid pond context learning method based on attention mechanism of the invention belongs to
Deep learning field, it is characterised in that include: empty spatial convolution pyramid pond model and attention model, the cavity volume
Product space pyramid pond model is made of a plurality of empty convolution path in parallel with different spreading rates, multiple dimensioned for extracting
Contextual information, the attention model characterize the relationship between different channel contextual informations by nonlinear function, in turn
Weight is distributed to multiple dimensioned contextual information, increases the multiple dimensioned contextual information of empty spatial convolution pyramid pond model
Habit ability.
Cavity spatial convolution pyramid pond model, it is characterised in that the cavity spatial convolution pyramid Chi Huamo
The single access of type can formalize are as follows:
Wherein p is the location index of the corresponding pixel in convolution kernel center, and c is the channel index of input, and d is the expansion
Rate, wC, (i, j)It is the convolution kernel weight of dedicated tunnel and position, xC, p+d (i, j)It is the pixel value of dedicated tunnel and position, G is to adopt
Sample grid, b are bias terms.Input feature vector figure x forms multiple dimensioned by empty spatial convolution pyramid pond model treatment
Characteristic pattern WhereinIt is d for the spreading ratenAccess corresponding to output, more rulers
Degree characteristic pattern is by being spliced to form the input x of attention modelASPP
The input of the attention model is complete by global pool and twice to be connected and obtains Bu Tong leading to after activation primitive
The weight z in road
Z=δ2(W2δ1(W1y))
WhereinIt is attention model input xASPPIn channel c, value corresponding to position (h, w), ycFor channel
The corresponding global pool value of c, y are the tensor obtained after each channel pool value is spliced, δ1And δ2For activation primitive, W1And W2It is complete
The weight of articulamentum.The input x of the attention modelASPPBy being multiplied to obtain multiple dimensioned contextual feature with the weight z
Scheme X
Detailed description of the invention
Fig. 1 is that a kind of empty spatial convolution pyramid pond context learning method based on attention mechanism of the present invention is shown
It is intended to
Specific embodiment
Using drawings and examples, the present invention will be further described below, and attached drawing described herein is used to provide to this
Further understanding for invention, constitutes part of this application, and do not constitute a limitation of the invention.
A kind of empty spatial convolution pyramid pond context learning method schematic diagram based on attention mechanism is shown in attached drawing
1, it is characterised in that include: empty spatial convolution pyramid pond model and attention model, the cavity spatial convolution gold word
Tower basin model is made of a plurality of empty convolution path in parallel with different spreading rates, for extracting multiple dimensioned context letter
Breath, the attention model characterize the relationship between different channel contextual informations by nonlinear function, and then to multiple dimensioned
Contextual information distributes weight, increases the multiple dimensioned contextual information learning ability of empty spatial convolution pyramid pond model.
Cavity spatial convolution pyramid pond model, it is characterised in that the cavity spatial convolution pyramid Chi Huamo
The single access of type can formalize are as follows:
Wherein p is the location index of the corresponding pixel in convolution kernel center, and c is the channel index of input, and d is the expansion
Rate, wC, (i, j)It is the convolution kernel weight of dedicated tunnel and position, xC, p+d (i, j)It is the pixel value of dedicated tunnel and position, G is to adopt
Sample grid, b are bias terms.Input feature vector figure x forms multiple dimensioned by empty spatial convolution pyramid pond model treatment
Characteristic pattern WhereinIt is d for the spreading ratenAccess corresponding to output, more rulers
Degree characteristic pattern is by being spliced to form the input x of attention modelASPP
The input of the attention model is complete by global pool and twice to be connected and obtains Bu Tong leading to after activation primitive
The weight z in road
Z=δ2(W2δ1(W1y))
WhereinIt is attention model input xASPPIn channel c, value corresponding to position (h, w), ycFor channel
The corresponding global pool value of c, y are the tensor obtained after each channel pool value is spliced, δ1And δ2For activation primitive, W1And W2It is complete
The weight of articulamentum.The input x of the attention modelASPPBy being multiplied to obtain multiple dimensioned contextual feature with the weight z
Scheme X
Above-described specific embodiment has carried out further the purpose of the present invention, technical scheme and beneficial effects
It is described in detail, it should be understood that being not intended to limit the present invention the foregoing is merely a specific embodiment of the invention
Protection scope, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should all include
Within protection scope of the present invention.
Claims (2)
1. a kind of empty spatial convolution pyramid pond context learning method based on attention mechanism, it is characterised in that packet
Contain: empty spatial convolution pyramid pond model and attention model, cavity spatial convolution pyramid pond model is by more
Empty convolution path in parallel of the item with different spreading rates forms, for extracting multiple dimensioned contextual information, the attention mould
Type characterizes the relationship between different channel contextual informations by nonlinear function, and then distributes multiple dimensioned contextual information and weigh
Weight increases the multiple dimensioned contextual information learning ability of empty spatial convolution pyramid pond model.
2. the pyramid pond of cavity spatial convolution described according to claim 1 model, it is characterised in that the cavity convolution is empty
Between the single access of pyramid pond model can formalize are as follows:
Wherein p is the location index of the corresponding pixel in convolution kernel center, and c is the channel index of input, and d is the spreading rate,
wE, (i, j)It is the convolution kernel weight of dedicated tunnel and position, xC, p+d (i, j)It is the pixel value of dedicated tunnel and position, G is sampling network
Lattice, b are bias terms.Input feature vector figure x forms Analysis On Multi-scale Features by the empty spatial convolution pyramid pond model treatment
Figure WhereinIt is d for the spreading ratenAccess corresponding to output, the multiple dimensioned spy
Sign figure is by being spliced to form the input x of attention modelASPP
The input of the attention model obtains different channels after full connection and activation primitive by global pool and twice
Weight z
Z=δ2(W2δ1(W1y))
WhereinIt is attention model input xASPPIn channel c, value corresponding to position (h, w), ycFor c pairs of channel
The global pool value answered, y are the tensor obtained after each channel pool value is spliced, δ1And δ2For activation primitive, W1And W2To connect entirely
The weight of layer.The input x of the attention modelASPPBy being multiplied to obtain multiple dimensioned contextual feature figure X with the weight z
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910351669.8A CN110378484A (en) | 2019-04-28 | 2019-04-28 | A kind of empty spatial convolution pyramid pond context learning method based on attention mechanism |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910351669.8A CN110378484A (en) | 2019-04-28 | 2019-04-28 | A kind of empty spatial convolution pyramid pond context learning method based on attention mechanism |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110378484A true CN110378484A (en) | 2019-10-25 |
Family
ID=68248538
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910351669.8A Pending CN110378484A (en) | 2019-04-28 | 2019-04-28 | A kind of empty spatial convolution pyramid pond context learning method based on attention mechanism |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110378484A (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110910405A (en) * | 2019-11-20 | 2020-03-24 | 湖南师范大学 | Brain tumor segmentation method and system based on multi-scale cavity convolutional neural network |
CN111046674A (en) * | 2019-12-20 | 2020-04-21 | 科大讯飞股份有限公司 | Semantic understanding method and device, electronic equipment and storage medium |
CN111159335A (en) * | 2019-12-12 | 2020-05-15 | 中国电子科技集团公司第七研究所 | Short text classification method based on pyramid pooling and LDA topic model |
CN111179283A (en) * | 2019-12-30 | 2020-05-19 | 深圳市商汤科技有限公司 | Image semantic segmentation method and device and storage medium |
CN111523546A (en) * | 2020-04-16 | 2020-08-11 | 湖南大学 | Image semantic segmentation method, system and computer storage medium |
CN111539524A (en) * | 2020-03-23 | 2020-08-14 | 字节跳动有限公司 | Lightweight self-attention module, neural network model and search method of neural network framework |
CN111723748A (en) * | 2020-06-22 | 2020-09-29 | 电子科技大学 | Infrared remote sensing image ship detection method |
CN111767799A (en) * | 2020-06-01 | 2020-10-13 | 重庆大学 | Improved down-going human target detection algorithm for fast R-CNN tunnel environment |
CN112541459A (en) * | 2020-12-21 | 2021-03-23 | 山东师范大学 | Crowd counting method and system based on multi-scale perception attention network |
CN112950640A (en) * | 2021-02-23 | 2021-06-11 | Oppo广东移动通信有限公司 | Video portrait segmentation method and device, electronic equipment and storage medium |
CN113205501A (en) * | 2021-05-10 | 2021-08-03 | 华中科技大学 | Weld defect multi-scale feature extraction module based on lightweight cavity convolution |
CN113222904A (en) * | 2021-04-21 | 2021-08-06 | 重庆邮电大学 | Concrete pavement crack detection method for improving PoolNet network structure |
CN113450366A (en) * | 2021-07-16 | 2021-09-28 | 桂林电子科技大学 | AdaptGAN-based low-illumination semantic segmentation method |
CN113486897A (en) * | 2021-07-29 | 2021-10-08 | 辽宁工程技术大学 | Semantic segmentation method for convolution attention mechanism up-sampling decoding |
CN113554156A (en) * | 2021-09-22 | 2021-10-26 | 中国海洋大学 | Multi-task learning model construction method based on attention mechanism and deformable convolution |
WO2021244621A1 (en) * | 2020-06-04 | 2021-12-09 | 华为技术有限公司 | Scenario semantic parsing method based on global guidance selective context network |
CN114782986A (en) * | 2022-03-28 | 2022-07-22 | 佳源科技股份有限公司 | Helmet wearing detection method, device, equipment and medium based on deep learning |
-
2019
- 2019-04-28 CN CN201910351669.8A patent/CN110378484A/en active Pending
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110910405A (en) * | 2019-11-20 | 2020-03-24 | 湖南师范大学 | Brain tumor segmentation method and system based on multi-scale cavity convolutional neural network |
CN110910405B (en) * | 2019-11-20 | 2023-04-18 | 湖南师范大学 | Brain tumor segmentation method and system based on multi-scale cavity convolutional neural network |
CN111159335A (en) * | 2019-12-12 | 2020-05-15 | 中国电子科技集团公司第七研究所 | Short text classification method based on pyramid pooling and LDA topic model |
CN111046674B (en) * | 2019-12-20 | 2024-05-31 | 科大讯飞股份有限公司 | Semantic understanding method and device, electronic equipment and storage medium |
CN111046674A (en) * | 2019-12-20 | 2020-04-21 | 科大讯飞股份有限公司 | Semantic understanding method and device, electronic equipment and storage medium |
CN111179283A (en) * | 2019-12-30 | 2020-05-19 | 深圳市商汤科技有限公司 | Image semantic segmentation method and device and storage medium |
CN111539524A (en) * | 2020-03-23 | 2020-08-14 | 字节跳动有限公司 | Lightweight self-attention module, neural network model and search method of neural network framework |
CN111539524B (en) * | 2020-03-23 | 2023-11-28 | 字节跳动有限公司 | Lightweight self-attention module and searching method of neural network framework |
CN111523546A (en) * | 2020-04-16 | 2020-08-11 | 湖南大学 | Image semantic segmentation method, system and computer storage medium |
CN111767799A (en) * | 2020-06-01 | 2020-10-13 | 重庆大学 | Improved down-going human target detection algorithm for fast R-CNN tunnel environment |
WO2021244621A1 (en) * | 2020-06-04 | 2021-12-09 | 华为技术有限公司 | Scenario semantic parsing method based on global guidance selective context network |
CN111723748A (en) * | 2020-06-22 | 2020-09-29 | 电子科技大学 | Infrared remote sensing image ship detection method |
CN111723748B (en) * | 2020-06-22 | 2022-04-29 | 电子科技大学 | Infrared remote sensing image ship detection method |
CN112541459A (en) * | 2020-12-21 | 2021-03-23 | 山东师范大学 | Crowd counting method and system based on multi-scale perception attention network |
CN112950640A (en) * | 2021-02-23 | 2021-06-11 | Oppo广东移动通信有限公司 | Video portrait segmentation method and device, electronic equipment and storage medium |
CN113222904A (en) * | 2021-04-21 | 2021-08-06 | 重庆邮电大学 | Concrete pavement crack detection method for improving PoolNet network structure |
CN113205501B (en) * | 2021-05-10 | 2022-06-17 | 华中科技大学 | Multi-scale feature extraction device and method for weld defects |
CN113205501A (en) * | 2021-05-10 | 2021-08-03 | 华中科技大学 | Weld defect multi-scale feature extraction module based on lightweight cavity convolution |
CN113450366B (en) * | 2021-07-16 | 2022-08-30 | 桂林电子科技大学 | AdaptGAN-based low-illumination semantic segmentation method |
CN113450366A (en) * | 2021-07-16 | 2021-09-28 | 桂林电子科技大学 | AdaptGAN-based low-illumination semantic segmentation method |
CN113486897A (en) * | 2021-07-29 | 2021-10-08 | 辽宁工程技术大学 | Semantic segmentation method for convolution attention mechanism up-sampling decoding |
CN113554156B (en) * | 2021-09-22 | 2022-01-11 | 中国海洋大学 | Multitask image processing method based on attention mechanism and deformable convolution |
CN113554156A (en) * | 2021-09-22 | 2021-10-26 | 中国海洋大学 | Multi-task learning model construction method based on attention mechanism and deformable convolution |
CN114782986A (en) * | 2022-03-28 | 2022-07-22 | 佳源科技股份有限公司 | Helmet wearing detection method, device, equipment and medium based on deep learning |
CN114782986B (en) * | 2022-03-28 | 2024-04-26 | 佳源科技股份有限公司 | Deep learning-based safety helmet wearing detection method, device, equipment and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110378484A (en) | A kind of empty spatial convolution pyramid pond context learning method based on attention mechanism | |
Huadong | Steps to the digital Silk Road | |
Xie et al. | Scale-free convolutional neural network for remote sensing scene classification | |
CN108921822A (en) | Image object method of counting based on convolutional neural networks | |
Wang et al. | FE-YOLOv5: Feature enhancement network based on YOLOv5 for small object detection | |
CN110765753B (en) | Document generation method, system, computer device and storage medium | |
CN108364023A (en) | Image-recognizing method based on attention model and system | |
Dai et al. | RADANet: Road augmented deformable attention network for road extraction from complex high-resolution remote-sensing images | |
CN109934260A (en) | Image, text and data fusion sensibility classification method and device based on random forest | |
CN109086722A (en) | Mix licence plate recognition method, device, electronic equipment | |
CN101430768A (en) | Two-dimension bar code system and its positioning method | |
Zhou et al. | Effective evaluation of infiltration and storage measures in sponge city construction: a case study of Fenghuang City | |
CN102831427A (en) | Texture feature extraction method fused with visual significance and gray level co-occurrence matrix (GLCM) | |
Vaz et al. | An application for regional coastal erosion processes in urban areas: A case study of the Golden Horseshoe in Canada | |
CN103678483A (en) | Video semantic analysis method based on self-adaption probability hypergraph and semi-supervised learning | |
Wang et al. | Flood risk assessment of metro system using improved trapezoidal fuzzy AHP: A case study of Guangzhou | |
CN103678552A (en) | Remote-sensing image retrieving method and system based on salient regional features | |
CN109543424A (en) | Data-privacy guard method, device, system and storage medium | |
CN105205491A (en) | Polarized SAR image classification method based on extreme learning machine | |
CN117113189A (en) | Comprehensive evaluation method for land classification, electronic equipment and storage medium | |
Cheng et al. | Multi-scale Feature Fusion and Transformer Network for urban green space segmentation from high-resolution remote sensing images | |
Zhao et al. | Changes of lake area, groundwater level and vegetation under the influence of ecological water conveyance—A case study of the tail lake of Tarim River in China | |
Song et al. | Spatial-temporal transformer network for multi-year ENSO prediction | |
Bettinelli et al. | Algorithm for parametric community detection in networks | |
Wirawan et al. | Semantic segmentation of rice field bund on unmanned aerial vehicle image using unet |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191025 |
|
WD01 | Invention patent application deemed withdrawn after publication |