CN109711413A - Image, semantic dividing method based on deep learning - Google Patents
Image, semantic dividing method based on deep learning Download PDFInfo
- Publication number
- CN109711413A CN109711413A CN201811646148.7A CN201811646148A CN109711413A CN 109711413 A CN109711413 A CN 109711413A CN 201811646148 A CN201811646148 A CN 201811646148A CN 109711413 A CN109711413 A CN 109711413A
- Authority
- CN
- China
- Prior art keywords
- module
- image
- neural network
- deep neural
- deep
- 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.)
- Granted
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
A kind of image, semantic dividing method based on deep learning is handled by data set, building deep semantic segmentation network, deep semantic divide network training and parameter learning, form to test image progress four part of semantic segmentation.The present invention makes full use of the marginal information of gray level image using the RGB image of input picture and gray level image as the input of network model, effectively increases the abundant degree of input feature vector；Convolutional neural networks and two-way thresholding recursive unit are combined, on the basis of learning image local feature, capture more context dependencies and global characteristics information；Coordinate information is added to characteristic pattern by the first coordinate channels module and the second coordinate channels module, enriches the translation specifications of model, the generalization ability of lift scheme generates high resolution, the accurate semantic segmentation result in boundary.
Description
Technical field
The invention belongs to computer visions and depth learning technology field, and in particular to a kind of image based on deep learning
Semantic segmentation method.
Background technique
Image, semantic segmentation be from pixel level, understand, identification picture content, the purpose is to establish each pixel and
Mapping relations one by one between semantic classes, are split according to semantic information, are widely used in scene understanding, drive automatically
It sails, the fields such as medical image analysis, robot vision.Image, semantic segmentation be image understanding foundation stone, segmentation result it is good
To badly the processing to subsequent image content be directly affected, therefore, had to the research of image, semantic cutting techniques very important
Realistic meaning.
Traditional image, semantic dividing method depends on manual feature extraction and probability graph model mostly, such as: random forest,
Condition random field (CRF), markov random file (MRF) etc., these methods can only learn the characterization information of shallow-layer, cannot generate
Accurate fine segmentation result.Since 2012, with the fast development of deep learning, the image language based on convolutional neural networks
Adopted dividing method becomes research hotspot.2014 years, Hariharan et al. proposed collaboration target detection and semantic segmentation
SDS (simultaneous detection and segmentation) method, it is every that this method uses MCG method to extract first
Then multiple candidate regions in width image divide two-way to extract bounding box (bounding-box) feature and foreground area using CNN
Feature, and the information fusion of two-way feature is completed, finally, utilizing non-maximum constrained NMS (non-maximum suppression)
Method generative semantics segmentation result.In addition to SDS method, similar also the methods of R-CNN, SPP are all based on candidate region
Semantic segmentation method, but such methods are suggested dependent on a large amount of region, cause memory consumption very big, the training time compares
Long, obtained semantic segmentation result precision is low.
In order to further decrease memory overhead, semantic segmentation precision is promoted.2015, Long et al. proposed full convolution net
Network model FCN (fully convolutional networks), the model full connection that depth convolutional Neural network is last
Layer is wholly converted into convolutional layer, forms end-to-end, pixel to pixel full convolutional network frame, enters image, semantic segmentation
One completely new epoch.Kendall et al. proposes a kind of depth convolutional encoding-decoding architecture SegNet, and the model is by one
Convolutional encoding network and a deconvolution decoding network composition, each encoder layer corresponds to a decoder layer, final to compile
The output of code device is admitted to soft-max classifier and is classified pixel-by-pixel.Chen et al. proposes one on the basis of FCN
More mature semantic segmentation model Deeplab-CRF, the model are obtained using the DCNN (depth convolutional neural networks) after optimization
It is upsampled to original image size to coarse shot chart and by bilinear interpolation, then uses full condition of contact random field (CRF)
It is iterated optimization, obtains fine semantic segmentation result.
The shortcomings that above-mentioned semantic segmentation method: first, mode input is generally RGB image, input is excessively single, may
Local feature is caused to lack；Second, these methods are all based on convolutional neural networks to carry out feature extraction, without sufficiently benefit
With the local feature information and global context dependence of image, cause the segmenting edge of image very coarse, segmentation precision
It is very low.
Summary of the invention
Technical problem to be solved by the present invention lies in the defects for overcoming existing method, and it is high, general to provide a kind of segmentation precision
The strong image, semantic dividing method based on deep learning of change ability.
Technical solution used by above-mentioned technical problem is solved to include the following steps:
S1, data set processing
Image data set is divided into training image collection and test chart image set, and data enhancing behaviour is carried out to training image collection
Make, the quantity of training image is increased into ten thousand grades of units；
S2, building deep semantic divide network
Deep semantic divides network by parallel deep neural network module, Fusion Features module, Softmax classification layer structure
At the parallel deep neural network module is used to carry out feature extraction to input picture, and the Fusion Features module will simultaneously
The output characteristic pattern of row deep neural network is weighted fusion and obtains new characteristic pattern, and the Softmax classification layer is by pixel
Class label prediction score value is converted into pixel class label prediction probability distribution map；
The parallel deep neural network module is by the first deep neural network module and the second deep neural network mould
Block composition, and the first deep neural network module is identical with the second deep neural network module network structure, the first depth nerve
The input of network module is the RGB image of input picture, and the input of the second deep neural network module is the gray scale of input picture
Image；
The first deep neural network module is by full convolutional network module, the first coordinate channels module, first circulation
Layer module, the second coordinate channels module, second circulation layer module, spatial pyramid pond module composition, the first coordinate channels mould
Block is identical as the structure of the second coordinate channels module, and first circulation layer module is identical as the structure of second circulation layer module, described
Full convolutional network module carries out local shape factor to input picture, and the first circulation layer module is used to capture the upper and lower of image
Literary dependence and global characteristics information, the first coordinate channels module connect the characteristic pattern that full convolutional network module exports
I, j, r coordinate channels constitute new characteristic pattern, described to learn more translation specifications information and improve the generalization ability of model
The characteristic pattern that spatial pyramid pond module exports second circulation layer module carries out convolution operation in multiple sample rates, extracts
The characteristic information in different scale region；
S3, deep semantic segmentation network training and parameter learning
S31, network model parameter initialization: pre-training model pair of the ResNet101 on ImageNet data set is used
Full convolutional network module carries out parameter initialization, is uniformly distributed using standard to first circulation layer module and second circulation layer module
Parameter initialization is carried out, carries out parameter initialization using convolutional layer of the standard gaussian distribution to spatial pyramid pond module；
S32, divide network using the enhanced training image collection training deep semantic of data, generate the pre- mark of pixel class
Probability distribution graph is signed, using prediction label probability and original tag probability calculation prediction loss, specifically uses losses by mixture function L
(θ) is used as objective function,
L (θ)=L1(θ)+L2(θ)
L in formula1(θ) is cross entropy loss function, L2(θ) is L2 regularization term, and θ is the parameter of deep semantic segmentation network；
S33, using stochastic gradient descent algorithm optimization object function, update network model ginseng with back-propagation algorithm
Number, terminates to train when the value of objective function no longer declines；
S4, semantic segmentation is carried out to test image
S41, the trained deep semantic of test chart image set input step S3 is divided into network；
S42, parallel deep neural network module carry out feature extraction to the test chart image set of input
The gray level image of input of the RGB image of test image as the first deep neural network module, test image is made
For the input of the second deep neural network module；
First deep neural network modular character extraction process are as follows: full convolutional network module passes through empty convolution, maximum pond
Change, convolution operation carries out local shape factor to the RGB image of test image；The characteristic pattern that full convolutional network module exports is led to
It crosses the first coordinate channels module and obtains the new horizontal and vertical scanning of characteristic pattern feeding first circulation layer module progress, learn image
Global characteristics information；The characteristic pattern that first circulation layer module exports is obtained into new characteristic pattern by the second coordinate channels module
It is re-fed into second circulation layer module and carries out horizontal and vertical scanning, capture the global characteristics information of image；By second circulation layer mould
The characteristic pattern input space pyramid pond module of block output, carries out convolution operation in multiple sample rates, extracts different scale
The characteristic information in region；
Second deep neural network modular character extraction process and the first deep neural network modular character extraction process phase
Together；
S43, the spy for exporting the characteristic pattern of the first deep neural network module output and the second deep neural network module
Sign figure is weighted fusion and obtains new characteristic pattern；
S44, the result of step S43 is sent into Softmax classification layer progress pixel class Tag Estimation, obtains every in image
Object category belonging to a pixel, and do bilinear interpolation operation and be upsampled to original image size, obtain fine semantic segmentation
Figure.
As a kind of perferred technical scheme, the first circulation layer module is by two two-way thresholding recursive unit structures
At the neuron number of two-way thresholding recursive unit is 150.
As a kind of perferred technical scheme, the spatial pyramid pond module by 4 different sample rates cavity
Convolution is constituted, and the convolution kernel size of empty convolution is 3 × 3, and spreading rate is respectively 4,6,8,12.
As a kind of perferred technical scheme, i, j, r coordinate channels are led to by i coordinate channels, j coordinate in the step S2
Road, r coordinate channels are constituted, and i coordinate channels, j coordinate channels and r coordinate channels are the coordinates matrix of e × f, i coordinate channels the
1 row~e row element is followed successively by 0,1 ..., e-1, the element that j coordinate channels the 1st arrange the~the f column is followed successively by 0,1 ..., f-
1, e, f takes positive integer, and r coordinate channels areM is the arbitrary element in i coordinate channels, n
For element identical with m coordinate position in j coordinate channels, the element in i coordinate channels and j coordinate channels is linearly zoomed to [-
1,1] in range.
As a kind of perferred technical scheme, the learning rate of parameter learning is carried out according to following formula in the step S3
Decaying:
T is the number of iterations, l in formula0It is initial learning rate, ltIt is the learning rate of the t times iteration, power is that momentum is 0.9.
Beneficial effects of the present invention are as follows:
The present invention makes full use of grayscale image using the RGB image of input picture and gray level image as the input of network model
The marginal information of picture effectively increases the abundant degree of input feature vector；Convolutional neural networks are mutually tied with two-way thresholding recursive unit
It closes, on the basis of learning image local feature, captures more context dependencies and global characteristics information；Pass through first
Coordinate information is added to characteristic pattern in coordinate channels module and the second coordinate channels module, enriches the translation specifications of model, Lifting Modules
The generalization ability of type generates high resolution, the accurate semantic segmentation result in boundary.
Detailed description of the invention
Fig. 1 is the image, semantic dividing method flow chart based on deep learning.
Fig. 2 is the first deep neural network module.
Fig. 3 is the semantic segmentation figure of WeizmannHorse data concentrated part test image.
Fig. 4 is the semantic segmentation figure of StanfordBackground data concentrated part test image.
Specific embodiment
The present invention is described in more detail with reference to the accompanying drawings and examples, but the present invention is not limited to these Examples.
Embodiment 1
WeizmannHorse data set is the image segmentation data set being made of 328 width images, data concentrated part
For image as shown in figure 3, the training of network model uses Pytorch platform, code writes completion, the present embodiment base on python
In the image, semantic dividing method of deep learning, as shown in Figure 1, its step are as follows:
S1, data set processing
200 are randomly selected from WeizmannHorse data set as training image collection, remaining 128 as survey
Attempt image set, and data enhancement operations are carried out to training image collection, the quantity of training image is made to increase to 11000；
S2, building deep semantic divide network
Deep semantic divides network by parallel deep neural network module, Fusion Features module, Softmax classification layer structure
It is used to carry out input picture feature extraction at, parallel deep neural network module, Fusion Features module is by two parallel depth
The output characteristic pattern of neural network is weighted fusion and obtains new characteristic pattern, and Softmax classification layer is pre- by pixel class label
It surveys score value and is converted into pixel class label prediction probability distribution map；
Parallel deep neural network module is made of the first deep neural network module and the second deep neural network module,
And first deep neural network module it is identical with the second deep neural network modular structure, the first deep neural network module it is defeated
Enter the RGB image for input picture, the input of the second deep neural network module is the gray level image of input picture；
In Fig. 2, the first deep neural network module is followed by full convolutional network module, the first coordinate channels module, first
Circular layer module, the second coordinate channels module, second circulation layer module, spatial pyramid pond module composition, the first coordinate channels
Module is identical as the structure of the second coordinate channels module, and first circulation layer module is identical as the structure of second circulation layer module；
Full convolutional network module carries out local shape factor to input picture, and full convolutional network module is by Deeplab_
The first convolution group~five convolution group of Resnet101 network is constituted in largeFOV model, wherein the first convolution group~the
Three convolution groups are operated using convolution operation, maximum pondization, and Volume Four product group and the 5th convolution group use convolution operation, cavity
Convolution operation；
First circulation layer module is made of two two-way thresholding recursive units, the neuron number of two-way thresholding recursive unit
It is 150, for capturing the context dependency and global characteristics information of image；1 × 1 big wisp feature of piecemeal is used first
Figure X is divided into G × K nonoverlapping region units, wherein G, K are respectively equal to the length and width of characteristic pattern X；Then a bidirectional gate is used
Each column progress vertical scanning of the limit recursive unit along characteristic pattern X, top-down scanning, a bottom-up scanning, every time
A region unit is read, and connects the output prediction that scanning obtains to obtain a compound characteristics figure by coordinated indexing
Similarly, using another two-way thresholding recursive unit along compound characteristics figureEvery row carry out horizontal sweep, one from
From left to right scanning, a right-to-left scanning read a region unit every time, and output prediction are connected by coordinated indexing
Obtain a new compound characteristics figureNew compound characteristics figureWith the contextual information from whole image；
First coordinate channels module constitutes newly characteristic pattern connection i, j, r coordinate channels that full convolutional network module exports
Characteristic pattern, to learn more translation specifications information and improve the generalization ability of model；I, j, r coordinate channels by i coordinate channels,
J coordinate channels, r coordinate channels are constituted, and i coordinate channels, j coordinate channels and r coordinate channels are the coordinates matrix of e × f, and i is sat
Mark the 1st row of channel~e row element is followed successively by 0,1 ..., e-1, the element that j coordinate channels the 1st arrange the~the f column is followed successively by 0,
1 ..., f-1, e, f take positive integer, and r coordinate channels areM is appointing in i coordinate channels
Meaning element, n is element identical with m coordinate position in j coordinate channels, and the element in i coordinate channels and j coordinate channels is linear
It zooms in [- 1,1] range；
The characteristic pattern that spatial pyramid pond module exports second circulation layer module carries out convolution in multiple sample rates
The characteristic information in different scale region is extracted in operation, which is made of the empty convolution of 4 different sample rates, empty convolution
Convolution kernel size be 3 × 3, spreading rate is respectively 4,6,8,12；
S3, deep semantic segmentation network training and parameter learning
S31, network model parameter initialization: pre-training model pair of the ResNet101 on ImageNet data set is used
Full convolutional network module carries out parameter initialization, is uniformly distributed using standard to first circulation layer module and second circulation layer module
Parameter initialization is carried out, carries out parameter initialization using convolutional layer of the standard gaussian distribution to spatial pyramid pond module；
S32, the picture size that the enhanced training image of data is concentrated is cut to 330 × 330, uses the instruction after cutting
Practice image training deep semantic and divide network, generate pixel class prediction label probability distribution graph, using prediction label probability and
Original tag probability calculation prediction loss, is specifically used as objective function using losses by mixture function L (θ),
L (θ)=L1(θ)+L2(θ)
L in formula1(θ) is cross entropy loss function, L2(θ) is L2 regularization term, and θ is the parameter of deep semantic segmentation network；
The cross entropy loss function L of the present embodiment1(θ) are as follows:
Y in formulapqIt is prediction label probability vector,It is original tag probability vector, N is the number of pixels of every picture,
N is 330 × 330=108900, and B is batch size, and B 10, C are pixel class numbers, and C 2, ln () are to seek natural logrithm；
The L2 regularization term L of the present embodiment2(θ) are as follows:
λ is regularization coefficient and is positive number in formula, and N is the number of pixels of every image, and N is 330 × 330=108900, B
It is batch size, B 10, S are the number of parameters of w and S takes positive integer, and w is weight parameter；
S33, using stochastic gradient descent algorithm optimization object function, update network model ginseng with back-propagation algorithm
Number, terminates to train when the value of objective function no longer declines, and for acceleration model convergence, introduces the learning rate of parameter learning,
Learning rate is decayed according to following formula:
T is the number of iterations and t≤35000, l in formula0It is initial learning rate, l0It is 0.003, ltIt is the study of the t times iteration
Rate, gradient decay to 0.0001, power be momentum be 0.9；
S4, semantic segmentation is carried out to test image
S41, the trained deep semantic of test chart image set input step S3 is divided into network；
S42, parallel deep neural network module carry out feature extraction to the test chart image set of input
Input of the RGB image of test image as the first deep neural network module, the corresponding grayscale image of test image
As the input as the second deep neural network module；
First deep neural network modular character extraction process are as follows: full convolutional network module passes through empty convolution, maximum pond
Change, convolution operation carries out local shape factor to the RGB image of test image；The characteristic pattern that full convolutional network module exports is led to
It crosses the first coordinate channels module and obtains the new horizontal and vertical scanning of characteristic pattern feeding first circulation layer module progress, learn image
Global characteristics information；The characteristic pattern that first circulation layer module exports is obtained into new characteristic pattern by the second coordinate channels module
It is re-fed into second circulation layer module and carries out horizontal and vertical scanning, capture the global characteristics information of image；By second circulation layer mould
The characteristic pattern input space pyramid pond module of block output, carries out convolution operation in multiple sample rates, extracts different scale
The characteristic information in region；
Second deep neural network modular character extraction process and the first deep neural network modular character extraction process phase
Together；
S43, the spy for exporting the characteristic pattern of the first deep neural network module output and the second deep neural network module
Sign figure is weighted fusion and obtains new characteristic pattern；
S44, the result of step S43 is sent into Softmax classification layer progress pixel class Tag Estimation, obtains every in image
Object category belonging to a pixel, and do bilinear interpolation operation and be upsampled to original image size, obtain fine semantic segmentation
Figure.
Semantic segmentation, part are carried out to 128 test images in WeizmannHorse data set using the present embodiment method
The semantic segmentation figure of test image is as shown in Figure 3, wherein it is the corresponding coloured silk of input picture that the first row, which is input picture, the second row,
Color label image, the third line are its corresponding semantic segmentation figures.
Embodiment 2
StanfordBackground data set is the image segmentation data set being made of 715 width images, data set
For middle parts of images as shown in figure 4, the training of network model uses Pytorch platform, code writes completion on python.
Image, semantic dividing method of the present embodiment based on deep learning, in step sl from StanfordBackground
573 are randomly selected in data set as training image collection, remaining 142 are used as test chart image set, and to training image collection
Data enhancement operations are carried out, the quantity of training image is made to increase to 13752；The enhanced training of data is schemed in step S32
Picture size in image set is cut to 421 × 421, divides network using the training image training deep semantic after cutting, generates
Pixel class prediction label probability distribution graph is specifically adopted using prediction label probability and original tag probability calculation prediction loss
Use losses by mixture function L (θ) as objective function,
L (θ)=L1(θ)+L2(θ)
L in formula1(θ) is cross entropy loss function, L2(θ) is L2 regularization term, and θ is the parameter of deep semantic segmentation network；
The cross entropy loss function L of the present embodiment1(θ) are as follows:
Y in formulapqIt is prediction label probability vector,It is original tag probability vector, N is the number of pixels of every picture,
N is 421 × 421=177241, and B is batch size, and B 6, C are pixel class numbers, and C 8, ln () are to seek natural logrithm；
The L2 regularization term L of the present embodiment2(θ) are as follows:
λ is regularization coefficient and is positive number in formula, and N is the number of pixels of every image, and N is 421 × 421=177241, B
It is batch size, B 6, S are the number of parameters of w and S takes positive integer, and w is weight parameter；Stochastic gradient descent is used in step S33
Algorithm optimization objective function updates network model parameter with back-propagation algorithm, when the value of objective function no longer declines
Terminate training, for acceleration model convergence, introduce the learning rate of parameter learning, learning rate is decayed according to following formula:
T is the number of iterations and t≤35000, l in formula0It is initial learning rate, l0It is 0.001, ltIt is the study of the t times iteration
Rate, gradient decay to 0.0001, power be momentum be 0.9；
Other operating procedures and parameter are same as Example 1.
Semantic point is carried out to 142 test images in StanfordBackground data set using the present embodiment method
It cuts, the semantic segmentation figure of partial test image is as shown in Figure 4, wherein it is input picture pair that the first row, which is input picture, the second row,
Color label image, the third line answered are its corresponding semantic segmentation figures.
Claims (5)
1. a kind of image, semantic dividing method based on deep learning, it is characterised in that be made of following step:
S1, data set processing
Image data set is divided into training image collection and test chart image set, and data enhancement operations are carried out to training image collection, it will
The quantity of training image increases to ten thousand grades of units；
S2, building deep semantic divide network
Deep semantic is divided network and is made of parallel deep neural network module, Fusion Features module, Softmax classification layer, institute
The parallel deep neural network module stated is used to carry out feature extraction to input picture, and the Fusion Features module is by parallel depth
The output characteristic pattern of neural network is weighted fusion and obtains new characteristic pattern, and the Softmax classification layer is by pixel class mark
Label prediction score value is converted into pixel class label prediction probability distribution map；
The parallel deep neural network module is by the first deep neural network module and the second deep neural network module group
At, and the first deep neural network module is identical with the second deep neural network module network structure, the first deep neural network
The input of module is the RGB image of input picture, and the input of the second deep neural network module is the gray level image of input picture；
The first deep neural network module is by full convolutional network module, the first coordinate channels module, first circulation layer mould
Block, the second coordinate channels module, second circulation layer module, spatial pyramid pond module composition, the first coordinate channels module with
The structure of second coordinate channels module is identical, and first circulation layer module is identical as the structure of second circulation layer module, the full volume
Product network module carries out local shape factor to input picture, the first circulation layer module be used to capture the context of image according to
Rely relationship and global characteristics information, the first coordinate channels module to the characteristic pattern that full convolutional network module export connect i, j,
R coordinate channels constitute new characteristic pattern, to learn more translation specifications information and improve the generalization ability of model, the space
The characteristic pattern that pyramid pond module exports second circulation layer module carries out convolution operation in multiple sample rates, extracts different
The characteristic information of dimensional area；
S3, deep semantic segmentation network training and parameter learning
S31, network model parameter initialization: it is rolled up using pre-training model of the ResNet101 on ImageNet data set to complete
Product network module carries out parameter initialization, is uniformly distributed using standard to first circulation layer module and the progress of second circulation layer module
Parameter initialization carries out parameter initialization using convolutional layer of the standard gaussian distribution to spatial pyramid pond module；
S32, divide network using the enhanced training image collection training deep semantic of data, it is general to generate pixel class prediction label
Rate distribution map specifically uses losses by mixture function L (θ) using prediction label probability and original tag probability calculation prediction loss
As objective function,
L (θ)=L1(θ)+L2(θ)
L in formula1(θ) is cross entropy loss function, L2(θ) is L2 regularization term, and θ is the parameter of deep semantic segmentation network；
S33, using stochastic gradient descent algorithm optimization object function, update network model parameter with back-propagation algorithm, directly
Terminate to train when no longer declining to the value of objective function；
S4, semantic segmentation is carried out to test image
S41, the trained deep semantic of test chart image set input step S3 is divided into network；
The RGB image that S42, parallel deep neural network module carry out feature extraction test image to the test chart image set of input is made
For the input of the first deep neural network module, the gray level image of test image is as the defeated of the second deep neural network module
Enter；
First deep neural network modular character extraction process are as follows: full convolutional network module passes through empty convolution, maximum pond, volume
Product operation carries out local shape factor to the RGB image of test image；The characteristic pattern of full convolutional network module output is passed through the
One coordinate channels module obtains new characteristic pattern and is sent into the horizontal and vertical scanning of first circulation layer module progress, learns the complete of image
Office's characteristic information；The characteristic pattern that first circulation layer module exports is obtained new characteristic pattern by the second coordinate channels module to send again
Enter second circulation layer module and carry out horizontal and vertical scanning, captures the global characteristics information of image；Second circulation layer module is defeated
Characteristic pattern input space pyramid pond module out carries out convolution operation in multiple sample rates, extracts different scale region
Characteristic information；
Second deep neural network modular character extraction process is identical as the first deep neural network modular character extraction process；
S43, the characteristic pattern for exporting the characteristic pattern of the first deep neural network module output and the second deep neural network module
It is weighted fusion and obtains new characteristic pattern；
S44, the result of step S43 is sent into Softmax classification layer progress pixel class Tag Estimation, obtains each picture in image
Object category belonging to element, and do bilinear interpolation operation and be upsampled to original image size, obtain fine semantic segmentation figure.
2. the image, semantic dividing method according to claim 1 based on deep learning, it is characterised in that: described first
Circulation layer module is made of two two-way thresholding recursive units, and the neuron number of two-way thresholding recursive unit is 150.
3. the image, semantic dividing method according to claim 1 based on deep learning, it is characterised in that: the space
Pyramid pond module is made of the empty convolution of 4 different sample rates, and the convolution kernel size of empty convolution is 3 × 3, spreading rate
Respectively 4,6,8,12.
4. the image, semantic dividing method according to claim 1 based on deep learning, it is characterised in that: the step
I, j, r coordinate channels are made of i coordinate channels, j coordinate channels, r coordinate channels in S2, and i coordinate channels, j coordinate channels and r are sat
Mark channel is the coordinates matrix of e × f, the 1st row of i coordinate channels~e row element is followed successively by 0,1 ..., e-1, j coordinate is logical
The element that road the 1st arranges the~the f column is followed successively by 0,1 ..., f-1, e, f take positive integer, and r coordinate channels areM be i coordinate channels in arbitrary element, n be j coordinate channels in m coordinate position phase
Same element, the element in i coordinate channels and j coordinate channels is linearly zoomed in [- 1,1] range.
5. the image, semantic dividing method according to claim 1 based on deep learning, it is characterised in that: the step
The learning rate of parameter learning is decayed according to following formula in S3:
T is the number of iterations, l in formula0It is initial learning rate, ltIt is the learning rate of the t times iteration, power is that momentum is 0.9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811646148.7A CN109711413B (en) | 2018-12-30 | 2018-12-30 | Image semantic segmentation method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811646148.7A CN109711413B (en) | 2018-12-30 | 2018-12-30 | Image semantic segmentation method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109711413A true CN109711413A (en) | 2019-05-03 |
| CN109711413B CN109711413B (en) | 2023-04-07 |
Family
ID=66260443
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811646148.7A Active CN109711413B (en) | 2018-12-30 | 2018-12-30 | Image semantic segmentation method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109711413B (en) |
Cited By (87)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163878A (en) * | 2019-05-28 | 2019-08-23 | 四川智盈科技有限公司 | A kind of image, semantic dividing method based on dual multiple dimensioned attention mechanism |
| CN110175613A (en) * | 2019-06-03 | 2019-08-27 | 常熟理工学院 | Street view image semantic segmentation method based on Analysis On Multi-scale Features and codec models |
| CN110189337A (en) * | 2019-05-31 | 2019-08-30 | 广东工业大学 | A kind of automatic Pilot image, semantic dividing method |
| CN110232418A (en) * | 2019-06-19 | 2019-09-13 | 深圳前海达闼云端智能科技有限公司 | Semantic recognition method, terminal and computer readable storage medium |
| CN110245665A (en) * | 2019-05-13 | 2019-09-17 | 天津大学 | Image, semantic dividing method based on attention mechanism |
| CN110264483A (en) * | 2019-06-19 | 2019-09-20 | 东北大学 | A kind of semantic image dividing method based on deep learning |
| CN110276402A (en) * | 2019-06-25 | 2019-09-24 | 北京工业大学 | A kind of salt body recognition methods based on the enhancing of deep learning semanteme boundary |
| CN110289081A (en) * | 2019-05-14 | 2019-09-27 | 杭州电子科技大学 | The epilepsy detection method of depth network stack model adaptation weighted feature fusion |
| CN110298849A (en) * | 2019-07-02 | 2019-10-01 | 电子科技大学 | Hard exudate dividing method based on eye fundus image |
| CN110310289A (en) * | 2019-06-17 | 2019-10-08 | 北京交通大学 | Lung tissue's image partition method based on deep learning |
| CN110348537A (en) * | 2019-07-18 | 2019-10-18 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN110390314A (en) * | 2019-07-29 | 2019-10-29 | 深兰科技（上海）有限公司 | A kind of visual perception method and apparatus |
| CN110472483A (en) * | 2019-07-02 | 2019-11-19 | 五邑大学 | A kind of method and device of the small sample semantic feature enhancing towards SAR image |
| CN110517329A (en) * | 2019-08-12 | 2019-11-29 | 北京邮电大学 | A kind of deep learning method for compressing image based on semantic analysis |
| CN110619639A (en) * | 2019-08-26 | 2019-12-27 | 苏州同调医学科技有限公司 | Method for segmenting radiotherapy image by combining deep neural network and probability map model |
| CN110619633A (en) * | 2019-09-10 | 2019-12-27 | 武汉科技大学 | Liver image segmentation method based on multi-path filtering strategy |
| CN110675421A (en) * | 2019-08-30 | 2020-01-10 | 电子科技大学 | Depth image collaborative segmentation method based on few labeling frames |
| CN110717921A (en) * | 2019-09-26 | 2020-01-21 | 哈尔滨工程大学 | Full convolution neural network semantic segmentation method of improved coding and decoding structure |
| CN110729045A (en) * | 2019-10-12 | 2020-01-24 | 闽江学院 | Tongue image segmentation method based on context-aware residual error network |
| CN110728683A (en) * | 2019-09-29 | 2020-01-24 | 吉林大学 | Image semantic segmentation method based on dense connection |
| CN110807462A (en) * | 2019-09-11 | 2020-02-18 | 浙江大学 | Training method insensitive to context of semantic segmentation model |
| CN110866922A (en) * | 2019-11-19 | 2020-03-06 | 中山大学 | Image semantic segmentation model and modeling method based on reinforcement learning and transfer learning |
| CN110880182A (en) * | 2019-11-18 | 2020-03-13 | 东声(苏州)智能科技有限公司 | Image segmentation model training method, image segmentation device and electronic equipment |
| CN110889854A (en) * | 2019-10-16 | 2020-03-17 | 深圳信息职业技术学院 | Sketch component segmentation method, system and device based on multi-scale deep learning and storage medium |
| CN110890155A (en) * | 2019-11-25 | 2020-03-17 | 中国科学技术大学 | Multi-class arrhythmia detection method based on lead attention mechanism |
| CN111062947A (en) * | 2019-08-14 | 2020-04-24 | 深圳市智影医疗科技有限公司 | Deep learning-based X-ray chest radiography focus positioning method and system |
| CN111079744A (en) * | 2019-12-06 | 2020-04-28 | 鲁东大学 | Intelligent vehicle license plate identification method and device suitable for complex illumination environment |
| CN111161273A (en) * | 2019-12-31 | 2020-05-15 | 电子科技大学 | Medical ultrasonic image segmentation method based on deep learning |
| CN111160311A (en) * | 2020-01-02 | 2020-05-15 | 西北工业大学 | Yellow river ice semantic segmentation method based on multi-attention machine system double-flow fusion network |
| CN111160109A (en) * | 2019-12-06 | 2020-05-15 | 北京联合大学 | Road segmentation method and system based on deep neural network |
| CN111310509A (en) * | 2020-03-12 | 2020-06-19 | 北京大学 | Real-time bar code detection system and method based on logistics waybill |
| CN111325259A (en) * | 2020-02-14 | 2020-06-23 | 武汉大学 | Remote sensing image classification method based on deep learning and binary coding |
| CN111400492A (en) * | 2020-02-17 | 2020-07-10 | 合肥工业大学 | Hierarchical feature text classification method and system based on SFM-DCNN |
| CN111507423A (en) * | 2020-04-24 | 2020-08-07 | 国网湖南省电力有限公司 | Engineering quantity calculation method for cleaning transmission line channel |
| CN111539412A (en) * | 2020-04-21 | 2020-08-14 | 上海云从企业发展有限公司 | Image analysis method, system, device and medium based on OCR |
| CN111583390A (en) * | 2020-04-28 | 2020-08-25 | 西安交通大学 | Three-dimensional semantic graph reconstruction method of convolutional neural network based on deep semantic fusion |
| CN111612803A (en) * | 2020-04-30 | 2020-09-01 | 杭州电子科技大学 | Vehicle image semantic segmentation method based on image definition |
| CN111666842A (en) * | 2020-05-25 | 2020-09-15 | 东华大学 | Shadow detection method based on double-current-cavity convolution neural network |
| CN111738265A (en) * | 2020-05-20 | 2020-10-02 | 山东大学 | Semantic segmentation method, system, medium, and electronic device for RGB-D image |
| CN111754520A (en) * | 2020-06-09 | 2020-10-09 | 江苏师范大学 | Deep learning-based cerebral hematoma segmentation method and system |
| CN111783811A (en) * | 2019-10-30 | 2020-10-16 | 北京京东尚科信息技术有限公司 | Pseudo label generation method and device |
| CN111860827A (en) * | 2020-06-04 | 2020-10-30 | 西安电子科技大学 | Multi-target positioning method and device of direction-finding system based on neural network model |
| CN111870279A (en) * | 2020-07-31 | 2020-11-03 | 西安电子科技大学 | Method, system and application for segmenting left ventricular myocardium of ultrasonic image |
| CN111899274A (en) * | 2020-08-05 | 2020-11-06 | 大连交通大学 | Particle size analysis method based on deep learning TEM image segmentation |
| CN111915619A (en) * | 2020-06-05 | 2020-11-10 | 华南理工大学 | Full convolution network semantic segmentation method for dual-feature extraction and fusion |
| CN111932501A (en) * | 2020-07-13 | 2020-11-13 | 太仓中科信息技术研究院 | Seal ring surface defect detection method based on semantic segmentation |
| CN112149547A (en) * | 2020-09-17 | 2020-12-29 | 南京信息工程大学 | Remote sensing image water body identification based on image pyramid guidance and pixel pair matching |
| CN112163111A (en) * | 2020-09-28 | 2021-01-01 | 杭州电子科技大学 | Rotation-invariant semantic information mining method |
| CN112164077A (en) * | 2020-09-25 | 2021-01-01 | 陕西师范大学 | Cell example segmentation method based on bottom-up path enhancement |
| CN112184714A (en) * | 2020-11-10 | 2021-01-05 | 平安科技（深圳）有限公司 | Image segmentation method, image segmentation device, electronic device, and medium |
| CN112465840A (en) * | 2020-12-10 | 2021-03-09 | 重庆紫光华山智安科技有限公司 | Semantic segmentation model training method, semantic segmentation method and related device |
| CN112465826A (en) * | 2019-09-06 | 2021-03-09 | 上海高德威智能交通系统有限公司 | Video semantic segmentation method and device |
| CN112508030A (en) * | 2020-12-18 | 2021-03-16 | 山西省信息产业技术研究院有限公司 | Tunnel crack detection and measurement method based on double-depth learning model |
| CN112507338A (en) * | 2020-12-21 | 2021-03-16 | 华南理工大学 | Improved system based on deep learning semantic segmentation algorithm |
| CN112508032A (en) * | 2021-01-29 | 2021-03-16 | 成都东方天呈智能科技有限公司 | Face image segmentation method and segmentation network for context information of association |
| CN112541916A (en) * | 2020-12-11 | 2021-03-23 | 华南理工大学 | Waste plastic image segmentation method based on dense connection |
| CN112571425A (en) * | 2020-11-30 | 2021-03-30 | 汕头大学 | Leakage point positioning autonomous control method and system of pressurized leakage stoppage robot |
| CN112580509A (en) * | 2020-12-18 | 2021-03-30 | 中国民用航空总局第二研究所 | Logical reasoning type road surface detection method and system |
| CN112614131A (en) * | 2021-01-10 | 2021-04-06 | 复旦大学 | Pathological image analysis method based on deformation representation learning |
| CN112651440A (en) * | 2020-12-25 | 2021-04-13 | 陕西地建土地工程技术研究院有限责任公司 | Soil effective aggregate classification and identification method based on deep convolutional neural network |
| CN112669325A (en) * | 2021-01-06 | 2021-04-16 | 大连理工大学 | Video semantic segmentation method based on active learning |
| CN112766195A (en) * | 2021-01-26 | 2021-05-07 | 西南交通大学 | Electrified railway bow net arcing visual detection method |
| CN112837326A (en) * | 2021-01-27 | 2021-05-25 | 南京中兴力维软件有限公司 | Remnant detection method, device and equipment |
| CN112884788A (en) * | 2021-03-08 | 2021-06-01 | 中南大学 | Cup optic disk segmentation method and imaging method based on rich context network |
| CN112950645A (en) * | 2021-03-24 | 2021-06-11 | 中国人民解放军国防科技大学 | Image semantic segmentation method based on multitask deep learning |
| CN112990304A (en) * | 2021-03-12 | 2021-06-18 | 国网智能科技股份有限公司 | Semantic analysis method and system suitable for power scene |
| CN112991266A (en) * | 2021-02-07 | 2021-06-18 | 复旦大学 | Semantic segmentation method and system for small sample medical image |
| CN112989919A (en) * | 2020-12-25 | 2021-06-18 | 首都师范大学 | Method and system for extracting target object from image |
| CN113033570A (en) * | 2021-03-29 | 2021-06-25 | 同济大学 | Image semantic segmentation method for improving fusion of void volume and multilevel characteristic information |
| CN113191367A (en) * | 2021-05-25 | 2021-07-30 | 华东师范大学 | Semantic segmentation method based on dense scale dynamic network |
| CN113205523A (en) * | 2021-04-29 | 2021-08-03 | 浙江大学 | Medical image segmentation and identification system, terminal and storage medium with multi-scale representation optimization |
| CN113222033A (en) * | 2021-05-19 | 2021-08-06 | 北京数研科技发展有限公司 | Monocular image estimation method based on multi-classification regression model and self-attention mechanism |
| CN113269786A (en) * | 2021-05-19 | 2021-08-17 | 青岛理工大学 | Assembly image segmentation method and device based on deep learning and guided filtering |
| CN113269197A (en) * | 2021-04-25 | 2021-08-17 | 南京三百云信息科技有限公司 | Certificate image vertex coordinate regression system and identification method based on semantic segmentation |
| CN113280820A (en) * | 2021-06-09 | 2021-08-20 | 华南农业大学 | Orchard visual navigation path extraction method and system based on neural network |
| CN113421269A (en) * | 2021-06-09 | 2021-09-21 | 南京瑞易智能科技有限公司 | Real-time semantic segmentation method based on double-branch deep convolutional neural network |
| CN113450311A (en) * | 2021-06-01 | 2021-09-28 | 国网河南省电力公司漯河供电公司 | Pin screw defect detection method and system based on semantic segmentation and spatial relationship |
| CN113468969A (en) * | 2021-06-03 | 2021-10-01 | 江苏大学 | Aliasing electronic component space expression method based on improved monocular depth estimation |
| CN113487622A (en) * | 2021-05-25 | 2021-10-08 | 中国科学院自动化研究所 | Head and neck organ image segmentation method and device, electronic equipment and storage medium |
| CN113496442A (en) * | 2020-03-19 | 2021-10-12 | 荷盛崧钜智财顾问股份有限公司 | Graph representation generation system, graph representation generation method and graph representation intelligent module thereof |
| CN113592894A (en) * | 2021-08-29 | 2021-11-02 | 浙江工业大学 | Image segmentation method based on bounding box and co-occurrence feature prediction |
| WO2022141669A1 (en) * | 2020-12-31 | 2022-07-07 | 苏州大学 | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions |
| CN115423810A (en) * | 2022-11-04 | 2022-12-02 | 国网江西省电力有限公司电力科学研究院 | Blade icing form analysis method for wind generating set |
| CN115631127A (en) * | 2022-08-15 | 2023-01-20 | 无锡东如科技有限公司 | Image segmentation method for industrial defect detection |
| CN115861323A (en) * | 2023-02-28 | 2023-03-28 | 泉州装备制造研究所 | Leather defect detection method based on refined segmentation network |
| CN116188479A (en) * | 2023-02-21 | 2023-05-30 | 北京长木谷医疗科技有限公司 | Hip joint image segmentation method and system based on deep learning |
| CN117351520A (en) * | 2023-10-31 | 2024-01-05 | 广州恒沙数字科技有限公司 | Front background image mixed generation method and system based on generation network |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184226A (en) * | 2015-08-11 | 2015-12-23 | 北京新晨阳光科技有限公司 | Digital identification method, digital identification device, neural network training method and neural network training device |
| CN106295139A (en) * | 2016-07-29 | 2017-01-04 | 姹ゅ钩 | A kind of tongue body autodiagnosis health cloud service system based on degree of depth convolutional neural networks |
| CN106548159A (en) * | 2016-11-08 | 2017-03-29 | 中国科学院自动化研究所 | Reticulate pattern facial image recognition method and device based on full convolutional neural networks |
| CN106709568A (en) * | 2016-12-16 | 2017-05-24 | 北京工业大学 | RGB-D image object detection and semantic segmentation method based on deep convolution network |
| US20170200067A1 (en) * | 2016-01-08 | 2017-07-13 | Siemens Healthcare Gmbh | Deep Image-to-Image Network Learning for Medical Image Analysis |
| US20180150684A1 (en) * | 2016-11-30 | 2018-05-31 | Shenzhen AltumView Technology Co., Ltd. | Age and gender estimation using small-scale convolutional neural network (cnn) modules for embedded systems |
| US20180165810A1 (en) * | 2016-12-13 | 2018-06-14 | Shanghai Sixth People's Hospital | Method of automatically detecting microaneurysm based on multi-sieving convolutional neural network |
| CN108268870A (en) * | 2018-01-29 | 2018-07-10 | 重庆理工大学 | Multi-scale feature fusion ultrasonoscopy semantic segmentation method based on confrontation study |
| CN108319972A (en) * | 2018-01-18 | 2018-07-24 | 南京师范大学 | A kind of end-to-end difference online learning methods for image, semantic segmentation |
| CN108427961A (en) * | 2018-02-11 | 2018-08-21 | 陕西师范大学 | Synthetic aperture focusing imaging depth appraisal procedure based on convolutional neural networks |
| CN108564587A (en) * | 2018-03-07 | 2018-09-21 | 浙江大学 | A kind of a wide range of remote sensing image semantic segmentation method based on full convolutional neural networks |
| CN108898145A (en) * | 2018-06-15 | 2018-11-27 | 西南交通大学 | A kind of image well-marked target detection method of combination deep learning |
| WO2018217828A1 (en) * | 2017-05-23 | 2018-11-29 | Intel Corporation | Methods and apparatus for discriminative semantic transfer and physics-inspired optimization of features in deep learning |
| CN109035263A (en) * | 2018-08-14 | 2018-12-18 | 电子科技大学 | Brain tumor image automatic segmentation method based on convolutional neural networks |
-
2018
- 2018-12-30 CN CN201811646148.7A patent/CN109711413B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184226A (en) * | 2015-08-11 | 2015-12-23 | 北京新晨阳光科技有限公司 | Digital identification method, digital identification device, neural network training method and neural network training device |
| US20170200067A1 (en) * | 2016-01-08 | 2017-07-13 | Siemens Healthcare Gmbh | Deep Image-to-Image Network Learning for Medical Image Analysis |
| CN106295139A (en) * | 2016-07-29 | 2017-01-04 | 姹ゅ钩 | A kind of tongue body autodiagnosis health cloud service system based on degree of depth convolutional neural networks |
| CN106548159A (en) * | 2016-11-08 | 2017-03-29 | 中国科学院自动化研究所 | Reticulate pattern facial image recognition method and device based on full convolutional neural networks |
| US20180150684A1 (en) * | 2016-11-30 | 2018-05-31 | Shenzhen AltumView Technology Co., Ltd. | Age and gender estimation using small-scale convolutional neural network (cnn) modules for embedded systems |
| US20180165810A1 (en) * | 2016-12-13 | 2018-06-14 | Shanghai Sixth People's Hospital | Method of automatically detecting microaneurysm based on multi-sieving convolutional neural network |
| CN106709568A (en) * | 2016-12-16 | 2017-05-24 | 北京工业大学 | RGB-D image object detection and semantic segmentation method based on deep convolution network |
| WO2018217828A1 (en) * | 2017-05-23 | 2018-11-29 | Intel Corporation | Methods and apparatus for discriminative semantic transfer and physics-inspired optimization of features in deep learning |
| CN108319972A (en) * | 2018-01-18 | 2018-07-24 | 南京师范大学 | A kind of end-to-end difference online learning methods for image, semantic segmentation |
| CN108268870A (en) * | 2018-01-29 | 2018-07-10 | 重庆理工大学 | Multi-scale feature fusion ultrasonoscopy semantic segmentation method based on confrontation study |
| CN108427961A (en) * | 2018-02-11 | 2018-08-21 | 陕西师范大学 | Synthetic aperture focusing imaging depth appraisal procedure based on convolutional neural networks |
| CN108564587A (en) * | 2018-03-07 | 2018-09-21 | 浙江大学 | A kind of a wide range of remote sensing image semantic segmentation method based on full convolutional neural networks |
| CN108898145A (en) * | 2018-06-15 | 2018-11-27 | 西南交通大学 | A kind of image well-marked target detection method of combination deep learning |
| CN109035263A (en) * | 2018-08-14 | 2018-12-18 | 电子科技大学 | Brain tumor image automatic segmentation method based on convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| JIZHI ZHANG: ""A Weakly Supervised Deep Learning Semantic Segmentation Framework"", 《2017 IEEE INTERNATIONAL CONFERENCE ON SMART CLOUD (SMARTCLOUD)》 * |
| 马苗: ""一种耦合深度信念网络的图像识别方法"", 《西安电子科技大学学报》 * |
Cited By (134)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245665A (en) * | 2019-05-13 | 2019-09-17 | 天津大学 | Image, semantic dividing method based on attention mechanism |
| CN110245665B (en) * | 2019-05-13 | 2023-06-06 | 天津大学 | Image semantic segmentation method based on attention mechanism |
| CN110289081A (en) * | 2019-05-14 | 2019-09-27 | 杭州电子科技大学 | The epilepsy detection method of depth network stack model adaptation weighted feature fusion |
| CN110289081B (en) * | 2019-05-14 | 2021-11-02 | 杭州电子科技大学 | Epilepsia detection method based on deep network stack model self-adaptive weighting feature fusion |
| CN110163878A (en) * | 2019-05-28 | 2019-08-23 | 四川智盈科技有限公司 | A kind of image, semantic dividing method based on dual multiple dimensioned attention mechanism |
| CN110189337A (en) * | 2019-05-31 | 2019-08-30 | 广东工业大学 | A kind of automatic Pilot image, semantic dividing method |
| CN110175613B (en) * | 2019-06-03 | 2021-08-10 | 常熟理工学院 | Streetscape image semantic segmentation method based on multi-scale features and codec model |
| CN110175613A (en) * | 2019-06-03 | 2019-08-27 | 常熟理工学院 | Street view image semantic segmentation method based on Analysis On Multi-scale Features and codec models |
| CN110310289A (en) * | 2019-06-17 | 2019-10-08 | 北京交通大学 | Lung tissue's image partition method based on deep learning |
| CN110232418A (en) * | 2019-06-19 | 2019-09-13 | 深圳前海达闼云端智能科技有限公司 | Semantic recognition method, terminal and computer readable storage medium |
| CN110264483A (en) * | 2019-06-19 | 2019-09-20 | 东北大学 | A kind of semantic image dividing method based on deep learning |
| CN110276402A (en) * | 2019-06-25 | 2019-09-24 | 北京工业大学 | A kind of salt body recognition methods based on the enhancing of deep learning semanteme boundary |
| CN110472483A (en) * | 2019-07-02 | 2019-11-19 | 五邑大学 | A kind of method and device of the small sample semantic feature enhancing towards SAR image |
| CN110472483B (en) * | 2019-07-02 | 2022-11-15 | 五邑大学 | SAR image-oriented small sample semantic feature enhancement method and device |
| CN110298849A (en) * | 2019-07-02 | 2019-10-01 | 电子科技大学 | Hard exudate dividing method based on eye fundus image |
| US11481574B2 (en) | 2019-07-18 | 2022-10-25 | Beijing Sensetime Technology Development Co., Ltd. | Image processing method and device, and storage medium |
| CN110348537A (en) * | 2019-07-18 | 2019-10-18 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN110390314A (en) * | 2019-07-29 | 2019-10-29 | 深兰科技（上海）有限公司 | A kind of visual perception method and apparatus |
| CN110517329A (en) * | 2019-08-12 | 2019-11-29 | 北京邮电大学 | A kind of deep learning method for compressing image based on semantic analysis |
| CN110517329B (en) * | 2019-08-12 | 2021-05-14 | 北京邮电大学 | Deep learning image compression method based on semantic analysis |
| CN111062947A (en) * | 2019-08-14 | 2020-04-24 | 深圳市智影医疗科技有限公司 | Deep learning-based X-ray chest radiography focus positioning method and system |
| CN111062947B (en) * | 2019-08-14 | 2023-04-25 | 深圳市智影医疗科技有限公司 | X-ray chest radiography focus positioning method and system based on deep learning |
| CN110619639A (en) * | 2019-08-26 | 2019-12-27 | 苏州同调医学科技有限公司 | Method for segmenting radiotherapy image by combining deep neural network and probability map model |
| CN110675421A (en) * | 2019-08-30 | 2020-01-10 | 电子科技大学 | Depth image collaborative segmentation method based on few labeling frames |
| CN110675421B (en) * | 2019-08-30 | 2022-03-15 | 电子科技大学 | Depth image collaborative segmentation method based on few labeling frames |
| CN112465826B (en) * | 2019-09-06 | 2023-05-16 | 上海高德威智能交通系统有限公司 | Video semantic segmentation method and device |
| CN112465826A (en) * | 2019-09-06 | 2021-03-09 | 上海高德威智能交通系统有限公司 | Video semantic segmentation method and device |
| CN110619633A (en) * | 2019-09-10 | 2019-12-27 | 武汉科技大学 | Liver image segmentation method based on multi-path filtering strategy |
| CN110807462B (en) * | 2019-09-11 | 2022-08-30 | 浙江大学 | Training method insensitive to context of semantic segmentation model |
| CN110807462A (en) * | 2019-09-11 | 2020-02-18 | 浙江大学 | Training method insensitive to context of semantic segmentation model |
| CN110717921B (en) * | 2019-09-26 | 2022-11-15 | 哈尔滨工程大学 | Full convolution neural network semantic segmentation method of improved coding and decoding structure |
| CN110717921A (en) * | 2019-09-26 | 2020-01-21 | 哈尔滨工程大学 | Full convolution neural network semantic segmentation method of improved coding and decoding structure |
| CN110728683A (en) * | 2019-09-29 | 2020-01-24 | 吉林大学 | Image semantic segmentation method based on dense connection |
| CN110728683B (en) * | 2019-09-29 | 2021-02-26 | 吉林大学 | Image semantic segmentation method based on dense connection |
| CN110729045A (en) * | 2019-10-12 | 2020-01-24 | 闽江学院 | Tongue image segmentation method based on context-aware residual error network |
| CN110889854B (en) * | 2019-10-16 | 2023-12-05 | 深圳信息职业技术学院 | Sketch part segmentation method, system, device and storage medium based on multi-scale deep learning |
| CN110889854A (en) * | 2019-10-16 | 2020-03-17 | 深圳信息职业技术学院 | Sketch component segmentation method, system and device based on multi-scale deep learning and storage medium |
| CN111783811A (en) * | 2019-10-30 | 2020-10-16 | 北京京东尚科信息技术有限公司 | Pseudo label generation method and device |
| CN110880182A (en) * | 2019-11-18 | 2020-03-13 | 东声(苏州)智能科技有限公司 | Image segmentation model training method, image segmentation device and electronic equipment |
| CN110866922A (en) * | 2019-11-19 | 2020-03-06 | 中山大学 | Image semantic segmentation model and modeling method based on reinforcement learning and transfer learning |
| CN110866922B (en) * | 2019-11-19 | 2023-05-16 | 中山大学 | Image semantic segmentation model and modeling method based on reinforcement learning and migration learning |
| CN110890155B (en) * | 2019-11-25 | 2022-10-28 | 中国科学技术大学 | Multi-class arrhythmia detection method based on lead attention mechanism |
| CN110890155A (en) * | 2019-11-25 | 2020-03-17 | 中国科学技术大学 | Multi-class arrhythmia detection method based on lead attention mechanism |
| CN111160109A (en) * | 2019-12-06 | 2020-05-15 | 北京联合大学 | Road segmentation method and system based on deep neural network |
| CN111079744B (en) * | 2019-12-06 | 2020-09-01 | 鲁东大学 | Intelligent vehicle license plate identification method and device suitable for complex illumination environment |
| CN111160109B (en) * | 2019-12-06 | 2023-08-18 | 北京联合大学 | Road segmentation method and system based on deep neural network |
| CN111079744A (en) * | 2019-12-06 | 2020-04-28 | 鲁东大学 | Intelligent vehicle license plate identification method and device suitable for complex illumination environment |
| CN111161273A (en) * | 2019-12-31 | 2020-05-15 | 电子科技大学 | Medical ultrasonic image segmentation method based on deep learning |
| CN111160311A (en) * | 2020-01-02 | 2020-05-15 | 西北工业大学 | Yellow river ice semantic segmentation method based on multi-attention machine system double-flow fusion network |
| CN111325259A (en) * | 2020-02-14 | 2020-06-23 | 武汉大学 | Remote sensing image classification method based on deep learning and binary coding |
| CN111400492A (en) * | 2020-02-17 | 2020-07-10 | 合肥工业大学 | Hierarchical feature text classification method and system based on SFM-DCNN |
| CN111400492B (en) * | 2020-02-17 | 2022-08-19 | 合肥工业大学 | Hierarchical feature text classification method and system based on SFM-DCNN |
| CN111310509A (en) * | 2020-03-12 | 2020-06-19 | 北京大学 | Real-time bar code detection system and method based on logistics waybill |
| CN113496442A (en) * | 2020-03-19 | 2021-10-12 | 荷盛崧钜智财顾问股份有限公司 | Graph representation generation system, graph representation generation method and graph representation intelligent module thereof |
| CN111539412A (en) * | 2020-04-21 | 2020-08-14 | 上海云从企业发展有限公司 | Image analysis method, system, device and medium based on OCR |
| CN111507423A (en) * | 2020-04-24 | 2020-08-07 | 国网湖南省电力有限公司 | Engineering quantity calculation method for cleaning transmission line channel |
| CN111507423B (en) * | 2020-04-24 | 2023-06-09 | 国网湖南省电力有限公司 | Engineering quantity measuring method for cleaning transmission line channel |
| CN111583390A (en) * | 2020-04-28 | 2020-08-25 | 西安交通大学 | Three-dimensional semantic graph reconstruction method of convolutional neural network based on deep semantic fusion |
| CN111612803B (en) * | 2020-04-30 | 2023-10-17 | 杭州电子科技大学 | Vehicle image semantic segmentation method based on image definition |
| CN111612803A (en) * | 2020-04-30 | 2020-09-01 | 杭州电子科技大学 | Vehicle image semantic segmentation method based on image definition |
| CN111738265A (en) * | 2020-05-20 | 2020-10-02 | 山东大学 | Semantic segmentation method, system, medium, and electronic device for RGB-D image |
| CN111666842B (en) * | 2020-05-25 | 2022-08-26 | 东华大学 | Shadow detection method based on double-current-cavity convolution neural network |
| CN111666842A (en) * | 2020-05-25 | 2020-09-15 | 东华大学 | Shadow detection method based on double-current-cavity convolution neural network |
| CN111860827A (en) * | 2020-06-04 | 2020-10-30 | 西安电子科技大学 | Multi-target positioning method and device of direction-finding system based on neural network model |
| CN111915619A (en) * | 2020-06-05 | 2020-11-10 | 华南理工大学 | Full convolution network semantic segmentation method for dual-feature extraction and fusion |
| CN111754520B (en) * | 2020-06-09 | 2023-09-15 | 江苏师范大学 | Deep learning-based cerebral hematoma segmentation method and system |
| CN111754520A (en) * | 2020-06-09 | 2020-10-09 | 江苏师范大学 | Deep learning-based cerebral hematoma segmentation method and system |
| CN111932501A (en) * | 2020-07-13 | 2020-11-13 | 太仓中科信息技术研究院 | Seal ring surface defect detection method based on semantic segmentation |
| CN111870279A (en) * | 2020-07-31 | 2020-11-03 | 西安电子科技大学 | Method, system and application for segmenting left ventricular myocardium of ultrasonic image |
| CN111870279B (en) * | 2020-07-31 | 2022-01-28 | 西安电子科技大学 | Method, system and application for segmenting left ventricular myocardium of ultrasonic image |
| CN111899274A (en) * | 2020-08-05 | 2020-11-06 | 大连交通大学 | Particle size analysis method based on deep learning TEM image segmentation |
| CN111899274B (en) * | 2020-08-05 | 2024-03-29 | 大连交通大学 | Particle size analysis method based on deep learning TEM image segmentation |
| CN112149547A (en) * | 2020-09-17 | 2020-12-29 | 南京信息工程大学 | Remote sensing image water body identification based on image pyramid guidance and pixel pair matching |
| CN112149547B (en) * | 2020-09-17 | 2023-06-02 | 南京信息工程大学 | Remote sensing image water body identification method based on image pyramid guidance and pixel pair matching |
| CN112164077B (en) * | 2020-09-25 | 2023-12-29 | 陕西师范大学 | Cell instance segmentation method based on bottom-up path enhancement |
| CN112164077A (en) * | 2020-09-25 | 2021-01-01 | 陕西师范大学 | Cell example segmentation method based on bottom-up path enhancement |
| CN112163111A (en) * | 2020-09-28 | 2021-01-01 | 杭州电子科技大学 | Rotation-invariant semantic information mining method |
| CN112184714A (en) * | 2020-11-10 | 2021-01-05 | 平安科技（深圳）有限公司 | Image segmentation method, image segmentation device, electronic device, and medium |
| CN112184714B (en) * | 2020-11-10 | 2023-08-22 | 平安科技（深圳）有限公司 | Image segmentation method, device, electronic equipment and medium |
| CN112571425A (en) * | 2020-11-30 | 2021-03-30 | 汕头大学 | Leakage point positioning autonomous control method and system of pressurized leakage stoppage robot |
| CN112571425B (en) * | 2020-11-30 | 2022-04-01 | 汕头大学 | Leakage point positioning autonomous control method and system of pressurized leakage stoppage robot |
| CN112465840A (en) * | 2020-12-10 | 2021-03-09 | 重庆紫光华山智安科技有限公司 | Semantic segmentation model training method, semantic segmentation method and related device |
| CN112465840B (en) * | 2020-12-10 | 2023-02-17 | 重庆紫光华山智安科技有限公司 | Semantic segmentation model training method, semantic segmentation method and related device |
| CN112541916B (en) * | 2020-12-11 | 2023-06-23 | 华南理工大学 | Waste plastic image segmentation method based on dense connection |
| CN112541916A (en) * | 2020-12-11 | 2021-03-23 | 华南理工大学 | Waste plastic image segmentation method based on dense connection |
| CN112508030A (en) * | 2020-12-18 | 2021-03-16 | 山西省信息产业技术研究院有限公司 | Tunnel crack detection and measurement method based on double-depth learning model |
| CN112580509B (en) * | 2020-12-18 | 2022-04-15 | 中国民用航空总局第二研究所 | Logical reasoning type road surface detection method and system |
| CN112580509A (en) * | 2020-12-18 | 2021-03-30 | 中国民用航空总局第二研究所 | Logical reasoning type road surface detection method and system |
| CN112507338B (en) * | 2020-12-21 | 2023-02-14 | 华南理工大学 | Improved system based on deep learning semantic segmentation algorithm |
| CN112507338A (en) * | 2020-12-21 | 2021-03-16 | 华南理工大学 | Improved system based on deep learning semantic segmentation algorithm |
| CN112989919B (en) * | 2020-12-25 | 2024-04-19 | 首都师范大学 | Method and system for extracting target object from image |
| CN112651440B (en) * | 2020-12-25 | 2023-02-14 | 陕西地建土地工程技术研究院有限责任公司 | Soil effective aggregate classification and identification method based on deep convolutional neural network |
| CN112651440A (en) * | 2020-12-25 | 2021-04-13 | 陕西地建土地工程技术研究院有限责任公司 | Soil effective aggregate classification and identification method based on deep convolutional neural network |
| CN112989919A (en) * | 2020-12-25 | 2021-06-18 | 首都师范大学 | Method and system for extracting target object from image |
| WO2022141669A1 (en) * | 2020-12-31 | 2022-07-07 | 苏州大学 | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions |
| CN112669325A (en) * | 2021-01-06 | 2021-04-16 | 大连理工大学 | Video semantic segmentation method based on active learning |
| CN112614131A (en) * | 2021-01-10 | 2021-04-06 | 复旦大学 | Pathological image analysis method based on deformation representation learning |
| CN112766195B (en) * | 2021-01-26 | 2022-03-29 | 西南交通大学 | Electrified railway bow net arcing visual detection method |
| CN112766195A (en) * | 2021-01-26 | 2021-05-07 | 西南交通大学 | Electrified railway bow net arcing visual detection method |
| CN112837326A (en) * | 2021-01-27 | 2021-05-25 | 南京中兴力维软件有限公司 | Remnant detection method, device and equipment |
| CN112837326B (en) * | 2021-01-27 | 2024-04-09 | 南京中兴力维软件有限公司 | Method, device and equipment for detecting carryover |
| CN112508032A (en) * | 2021-01-29 | 2021-03-16 | 成都东方天呈智能科技有限公司 | Face image segmentation method and segmentation network for context information of association |
| CN112991266A (en) * | 2021-02-07 | 2021-06-18 | 复旦大学 | Semantic segmentation method and system for small sample medical image |
| CN112884788B (en) * | 2021-03-08 | 2022-05-10 | 中南大学 | Cup optic disk segmentation method and imaging method based on rich context network |
| CN112884788A (en) * | 2021-03-08 | 2021-06-01 | 中南大学 | Cup optic disk segmentation method and imaging method based on rich context network |
| CN112990304B (en) * | 2021-03-12 | 2024-03-12 | 国网智能科技股份有限公司 | Semantic analysis method and system suitable for power scene |
| CN112990304A (en) * | 2021-03-12 | 2021-06-18 | 国网智能科技股份有限公司 | Semantic analysis method and system suitable for power scene |
| CN112950645B (en) * | 2021-03-24 | 2023-05-12 | 中国人民解放军国防科技大学 | Image semantic segmentation method based on multitask deep learning |
| CN112950645A (en) * | 2021-03-24 | 2021-06-11 | 中国人民解放军国防科技大学 | Image semantic segmentation method based on multitask deep learning |
| CN113033570A (en) * | 2021-03-29 | 2021-06-25 | 同济大学 | Image semantic segmentation method for improving fusion of void volume and multilevel characteristic information |
| CN113269197A (en) * | 2021-04-25 | 2021-08-17 | 南京三百云信息科技有限公司 | Certificate image vertex coordinate regression system and identification method based on semantic segmentation |
| CN113269197B (en) * | 2021-04-25 | 2024-03-08 | 南京三百云信息科技有限公司 | Certificate image vertex coordinate regression system and identification method based on semantic segmentation |
| CN113205523A (en) * | 2021-04-29 | 2021-08-03 | 浙江大学 | Medical image segmentation and identification system, terminal and storage medium with multi-scale representation optimization |
| CN113222033A (en) * | 2021-05-19 | 2021-08-06 | 北京数研科技发展有限公司 | Monocular image estimation method based on multi-classification regression model and self-attention mechanism |
| CN113269786A (en) * | 2021-05-19 | 2021-08-17 | 青岛理工大学 | Assembly image segmentation method and device based on deep learning and guided filtering |
| CN113269786B (en) * | 2021-05-19 | 2022-12-27 | 青岛理工大学 | Assembly image segmentation method and device based on deep learning and guided filtering |
| CN113487622A (en) * | 2021-05-25 | 2021-10-08 | 中国科学院自动化研究所 | Head and neck organ image segmentation method and device, electronic equipment and storage medium |
| CN113191367A (en) * | 2021-05-25 | 2021-07-30 | 华东师范大学 | Semantic segmentation method based on dense scale dynamic network |
| CN113487622B (en) * | 2021-05-25 | 2023-10-31 | 中国科学院自动化研究所 | Head-neck organ image segmentation method, device, electronic equipment and storage medium |
| CN113450311B (en) * | 2021-06-01 | 2023-01-13 | 国网河南省电力公司漯河供电公司 | Pin screw defect detection method and system based on semantic segmentation and spatial relationship |
| CN113450311A (en) * | 2021-06-01 | 2021-09-28 | 国网河南省电力公司漯河供电公司 | Pin screw defect detection method and system based on semantic segmentation and spatial relationship |
| CN113468969A (en) * | 2021-06-03 | 2021-10-01 | 江苏大学 | Aliasing electronic component space expression method based on improved monocular depth estimation |
| CN113280820A (en) * | 2021-06-09 | 2021-08-20 | 华南农业大学 | Orchard visual navigation path extraction method and system based on neural network |
| CN113280820B (en) * | 2021-06-09 | 2022-11-29 | 华南农业大学 | Orchard visual navigation path extraction method and system based on neural network |
| CN113421269A (en) * | 2021-06-09 | 2021-09-21 | 南京瑞易智能科技有限公司 | Real-time semantic segmentation method based on double-branch deep convolutional neural network |
| CN113592894B (en) * | 2021-08-29 | 2024-02-02 | 浙江工业大学 | Image segmentation method based on boundary box and co-occurrence feature prediction |
| CN113592894A (en) * | 2021-08-29 | 2021-11-02 | 浙江工业大学 | Image segmentation method based on bounding box and co-occurrence feature prediction |
| CN115631127B (en) * | 2022-08-15 | 2023-09-19 | 无锡东如科技有限公司 | Image segmentation method for industrial defect detection |
| CN115631127A (en) * | 2022-08-15 | 2023-01-20 | 无锡东如科技有限公司 | Image segmentation method for industrial defect detection |
| CN115423810A (en) * | 2022-11-04 | 2022-12-02 | 国网江西省电力有限公司电力科学研究院 | Blade icing form analysis method for wind generating set |
| CN116188479B (en) * | 2023-02-21 | 2024-04-02 | 北京长木谷医疗科技股份有限公司 | Hip joint image segmentation method and system based on deep learning |
| CN116188479A (en) * | 2023-02-21 | 2023-05-30 | 北京长木谷医疗科技有限公司 | Hip joint image segmentation method and system based on deep learning |
| CN115861323A (en) * | 2023-02-28 | 2023-03-28 | 泉州装备制造研究所 | Leather defect detection method based on refined segmentation network |
| CN117351520A (en) * | 2023-10-31 | 2024-01-05 | 广州恒沙数字科技有限公司 | Front background image mixed generation method and system based on generation network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109711413B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109711413A (en) | Image, semantic dividing method based on deep learning | |
| Remez et al. | Learning to segment via cut-and-paste | |
| CN110428428B (en) | Image semantic segmentation method, electronic equipment and readable storage medium | |
| CN106157319B (en) | The conspicuousness detection method in region and Pixel-level fusion based on convolutional neural networks | |
| CN112308860B (en) | Earth observation image semantic segmentation method based on self-supervision learning | |
| CN109446970A (en) | A kind of Intelligent Mobile Robot road scene recognition methods based on deep learning | |
| CN104992223B (en) | Intensive Population size estimation method based on deep learning | |
| CN106920243A (en) | The ceramic material part method for sequence image segmentation of improved full convolutional neural networks | |
| CN103530619B (en) | Gesture identification method based on a small amount of training sample that RGB-D data are constituted | |
| CN109446992A (en) | Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning | |
| CN107833183A (en) | A kind of satellite image based on multitask deep neural network while super-resolution and the method for coloring | |
| CN110348376A (en) | A kind of pedestrian's real-time detection method neural network based | |
| CN108334830A (en) | A kind of scene recognition method based on target semanteme and appearance of depth Fusion Features | |
| CN112784736B (en) | Character interaction behavior recognition method based on multi-modal feature fusion | |
| CN110276402A (en) | A kind of salt body recognition methods based on the enhancing of deep learning semanteme boundary | |
| CN113673338B (en) | Automatic labeling method, system and medium for weak supervision of natural scene text image character pixels | |
| CN109558862A (en) | The people counting method and system of attention refinement frame based on spatial perception | |
| CN109766856A (en) | A kind of method of double fluid RGB-D Faster R-CNN identification milking sow posture | |
| CN113379771B (en) | Hierarchical human body analysis semantic segmentation method with edge constraint | |
| CN114187450A (en) | Remote sensing image semantic segmentation method based on deep learning | |
| CN107944459A (en) | A kind of RGB D object identification methods | |
| CN108053420A (en) | A kind of dividing method based on the unrelated attribute dynamic scene of limited spatial and temporal resolution class | |
| CN109508686A (en) | A kind of Human bodys' response method based on the study of stratification proper subspace | |
| CN110458132A (en) | One kind is based on random length text recognition method end to end | |
| CN112329771A (en) | Building material sample identification method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |