CN105139395A - SAR image segmentation method based on wavelet pooling convolutional neural networks - Google Patents

SAR image segmentation method based on wavelet pooling convolutional neural networks Download PDF

Info

Publication number
CN105139395A
CN105139395A CN201510512535.1A CN201510512535A CN105139395A CN 105139395 A CN105139395 A CN 105139395A CN 201510512535 A CN201510512535 A CN 201510512535A CN 105139395 A CN105139395 A CN 105139395A
Authority
CN
China
Prior art keywords
sar image
pixel
convolutional neural
neural networks
image
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
Application number
CN201510512535.1A
Other languages
Chinese (zh)
Other versions
CN105139395B (en
Inventor
刘芳
段一平
郝红侠
焦李成
李玲玲
尚荣华
马文萍
杨淑媛
马晶晶
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN201510512535.1A priority Critical patent/CN105139395B/en
Publication of CN105139395A publication Critical patent/CN105139395A/en
Application granted granted Critical
Publication of CN105139395B publication Critical patent/CN105139395B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

The invention discloses an SAR image segmentation method based on wavelet pooling convolutional neural networks. The SAR image segmentation method comprises 1. constructing a wavelet pooling layer and forming wavelet pooling convolutional neural networks; 2. selecting image blocks and inputting the image blocks into the wavelet pooling convolutional neural networks, and training the image blocks; 3. inputting all the image blocks into the trained networks, and testing the image blocks to obtain a first class mark of an SAR image; 4. performing superpixel segmentation of the SAR image, and blending the superpixel segmentation result with the first class mark of the SAR image to obtain a second class mark of the SAR image; 5. obtaining a third class mark of the SAR image according to a Markov random field model, and blending the third class mark of the SAR image with the superpixel segmentation result to obtain a fourth class mark of the SAR image; and 6. blending the second class mark of the SAR image with the fourth class mark of the SAR image according to an SAR image gradient map to obtain the eventual segmentation result. The SAR image segmentation method based on wavelet pooling convolutional neural networks improves the segmentation effect of the SAR image and can be used for target detection and identification.

Description

Based on the SAR image segmentation method of small echo pond convolutional neural networks
Technical field
The invention belongs to technical field of image processing, relate to SAR image segmentation method, can be used for Target detection and identification.
Background technology
Along with the fast development of remote sensing technology, SAR image decipher plays an increasingly important role in the energy, environment, archaeology etc.SAR image segmentation is a ring basic and important in SAR image decipher, and it can be offered help for follow-up classification, detection, identification and tracking.The main target of SAR image segmentation SAR image is divided into the connected region not have common factor, and completing this target needs to mark each pixel of SAR image, is therefore a very difficult task.
Existing SAR image segmentation method is mainly divided into the method for feature based and the method for Corpus--based Method model.The feature that the method for feature based mainly extracts some SAR image is split, such as textural characteristics, limit characteristic sum composite character etc.These methods achieve good segmentation effect in the homogenous region of SAR image.But SAR image is earth observation, some complex scenes such as city, forest always can be observed.The light and shade that the scene of these complexity invariably accompanies sharp-pointed changes, and this characteristic makes the method for feature based be challenged.The mode of SAR image segmentation problem probability is expressed by the method for Corpus--based Method model, because these class methods consider the spatial context relation of image, receives increasing concern recent years.The feature interpretation of image is the distribution of some experiences by the method for Corpus--based Method model, such as Nakagami distribution, Gamma distribution, K distribution, G distribution etc.Spatial context model is described as Gibbs distribution, polynomial expression logistic regression function etc.In fact, SAR image segmentation problem transforms into the parameter problem asking these distributions or function.Parametric solution solves by the method for Combinatorial Optimization usually, and this is a process consuming time, and many times always can not find satisfactory solution.Although the method based on model achieves good segmentation effect, for the region of the grand texture of some complexity, still they can not be divided into conforming region.These two class methods are all under artificially given feature, carry out the segmentation of SAR image, and convolutional neural networks from the architectural feature of data study image successively itself, can improve the consistance of heterogeneous areas in cutting procedure.But, pond method traditional in convolutional neural networks, i.e. maximum pond, average pond method are simply in the neighborhood window of characteristic pattern, get maximal value or mean value, certain destruction can be caused to the structure of the feature learnt, be unfavorable for the segmentation of SAR image.
Summary of the invention
The object of the invention is to for above-mentioned existing methods not enough, propose a kind of SAR image segmentation method based on small echo pond convolutional neural networks, to promote the effect of SAR image segmentation.
For achieving the above object, technical scheme of the present invention is as follows:
(1) in convolutional neural networks, carry out wavelet transformation to characteristic pattern and get its approximation subband forming small echo pond layer, utilize small echo pond layer to form the convolutional neural networks in small echo pond;
(2) according to SAR image signature, every class chooses the pixel of 50%, and is input to by the image block centered by this pixel in the convolutional neural networks in small echo pond and trains;
(3) to all pixels of SAR image, the image block chosen centered by this pixel is input in the network trained and tests, and obtains the first kind mark of SAR image;
(4) super-pixel segmentation is carried out to SAR image, obtain the super-pixel segmentation result of SAR image, and the first kind mark of itself and SAR image is merged, obtain the Equations of The Second Kind mark of SAR image;
(5) obtain the 3rd class mark of SAR image according to Markov random field models, and the super-pixel segmentation result of itself and SAR image is merged, obtain the 4th class mark of SAR image;
(6) according to SAR image gradient map, the Equations of The Second Kind mark of SAR image and the 4th class mark are merged, obtains final segmentation result.
The present invention compared with prior art tool has the following advantages:
First, traditional tank method in convolutional neural networks, i.e. maximum pond, average pond method, just simply in the neighborhood of characteristic pattern, get maximal value or mean value, certain destruction can be caused to the architectural feature learnt, the small echo pondization that the present invention adopts not only has the function of noise reduction, and can better keep the architectural feature that learns, and this segmentation for image is vital.
Second, the convolutional neural networks that the present invention is based on small echo pond can from raw data automatic learning feature, and not needing the type of artificial given feature, this abundant architectural feature learnt from raw data can improve heterogeneous areas, the such as consistance of the segmentation result such as city, forest.
3rd, the present invention is based on the convergence strategy of SAR image gradient, merge based on the convolutional neural networks in small echo pond and the advantage of Markov random field models, again can the consistance of retaining zone while making to be segmented in accurately positioning boundary.
Accompanying drawing explanation
Fig. 1 is realization flow figure of the present invention;
Fig. 2 is small echo pond convolutional neural networks schematic diagram in the present invention;
Fig. 3 is the characteristic pattern that in the present invention, small echo pond convolutional neural networks learns;
Fig. 4 is the SAR image signature that the present invention emulates use;
Fig. 5 is the segmentation result figure of the NoerdlingerRies image being 1 meter to X-band resolution by the present invention and existing method;
Fig. 6 is the segmentation result figure of the Piperiver image being 1 meter to Ku wave band resolution by the present invention and existing method.
Embodiment
With reference to Fig. 1, implementation step of the present invention is as follows:
Step 1, builds small echo pond layer and forms the convolutional neural networks in small echo pond.
(1.1) according to the following formula to characteristic pattern CF kcarry out wavelet transformation,
{ SF k ϵ ( x , y ) } ϵ = a , h , v , d = d o w n ( CF k η ( x , y ) * ψ )
Wherein k=1 ..., N, N are the number of characteristic pattern: characteristic pattern CF kin neighborhood block centered by point (x, y), * represents convolution operation, and ψ is wavelet basis function, lower 2 sampling operations of down () representative, be point in each subband that wavelet transformation obtains, a, h, v, d represent approximation subband, horizontal subband, vertical subband and diagonal angle subband respectively;
(1.2) approximation subband characteristic pattern is used form small echo pond layer, wherein k=1 ..., N, N are the number of characteristic pattern;
(1.3) small echo pond layer is utilized to form the convolutional neural networks in small echo pond, as shown in Figure 2, this neural network is the eight layers of neural network comprising input layer, convolutional layer, pond layer, full articulamentum and output layer, wherein, ground floor is input layer, and the second layer and the 4th layer are convolutional layer, and third layer and layer 5 are pond layer, layer 6 and layer 7 are full articulamentum, and the 8th layer is output layer.
Step 2, chooses image block and is input in the convolutional neural networks in small echo pond and trains.
(2.1) in SAR image signature as shown in Figure 4, every class chooses the pixel of 50%, and to each pixel choose centered by it 32 × 32 image block as training sample, wherein Fig. 4 (a) is NoerdlingerRies image, Fig. 4 (b) is the signature of NoerdlingerRies image, and classification number is 5, Fig. 4 (c) is Piperiver image, Fig. 4 (b) is the signature of Piperiver image, and classification number is 5;
(2.2) training sample is input to the convolutional neural networks in the small echo pond that step (1) builds, and by the parameter of Back Propagation Algorithm learning network, until all training sample training are complete.
Step 3, is input to all image blocks in the network trained and tests.
(3.1) to pixels all in SAR image, the image block of 32 × 32 centered by this pixel is chosen as test sample book;
(3.2) test sample book is input in the neural network trained in step (2), study is to characteristic pattern, as shown in Figure 3, wherein Fig. 3 (a) for test sample book type be forest time, the second layer that e-learning arrives and the 4th layer of characteristic pattern, Fig. 3 (b) for test sample book type be city time, the second layer that e-learning arrives and the 4th layer of characteristic pattern, Fig. 3 (c) for test sample book type be farmland 1 time, the second layer that e-learning arrives and the 4th layer of characteristic pattern, Fig. 3 (d) for test sample book type be farmland 2 time, the second layer that e-learning arrives and the 4th layer of characteristic pattern,
(3.3) utilize propagated forward algorithm, prediction obtains the class mark of each image block, namely obtains the first kind mark of each pixel in SAR image.
Step 4, carries out super-pixel segmentation to SAR image, obtains the Equations of The Second Kind mark of this SAR image.
The common method of image superpixel segmentation has: watershed divide, MeanShift and TurboPixels etc.Adopt in the present invention but be not limited to the method for TurboPixels, obtaining the segmentation result of SAR image super-pixel;
According to the first kind mark of the SAR image obtained in step (3), calculate number of pixels of all categories in each super-pixel;
Each super-pixel is labeled as that classification that in this super-pixel, number of pixels is maximum, obtains the Equations of The Second Kind mark of SAR image.
Step 5, obtains the 3rd class mark of SAR image according to Markov random field models.
(5.1) the prior probability formula of Markov random field models is as follows:
P ( x s ) = exp ( - u ( x s ) ) Σ x s ∈ L exp ( - u ( x s ) )
Wherein, x sthe class mark of representative image pixel, L={1 ..., m} represents the classification of image, u (x s) be defined as follows:
u ( x s ) = - β Σ i ∈ N s [ δ ( x s , x i ) - 1 ]
Wherein, N srepresent x scentered by neighborhood territory pixel, β is model parameter, δ ( x s , x i ) = 1 , x s = x i 0 , x s ≠ x i ;
(5.2) the likelihood probability formula of Markov random field models is as follows:
p ( y s | x s ) = 2 Γ ( v l ) ( v l μ l ) v l y s 2 v l - 1 e ( - v l y s 2 μ l )
Wherein, y sthe pixel value of representative image, v l, μ lfor model parameter, Γ () is Gamma distribution.
(5.3) posterior probability formula of Markov random field models is as follows:
x ^ s = arg max x s ∈ L { p ( y s | x s ) · p ( x s | x N s ) }
(5.4) the 3rd class mark of SAR image is obtained according to Markov random field models wherein L={1 ..., m} represents the classification of image.
Step 6, merges the 3rd class mark of SAR image and super-pixel segmentation result, obtains the 4th class mark of this SAR image.
(6.1) according to the 3rd class mark of the SAR image obtained in step (5), number of pixels of all categories in each super-pixel is calculated;
(6.2) each super-pixel is labeled as that maximum classification of number of pixels in this super-pixel, obtains the 4th class mark of this SAR image.
Step 7, obtains SAR image gradient map, the Equations of The Second Kind mark of SAR image and the 4th class mark is merged, obtain final class mark.
(7.1) ratio operator r, right-angled intersection operator c and first difference operator y are merged, obtain the operation operator f after merging:
f = x y 1 - x - y + 2 x y
Wherein, x = r 2 + c 2 2 ;
(7.2) the operation operator f after fusion and SAR image are carried out convolution, obtain SAR image gradient map, and the gradient of pixel (x, y) is designated as G (x, y);
(7.3) for pixel (x, y) gradient G (x, y) threshold value T=185 is set, and G (x, y) and T are compared, if G (x, y) >=T, then using the Equations of The Second Kind mark of this pixel as final class mark, otherwise, using the 4th class mark of this pixel as final class mark.
Effect of the present invention can be described by following emulation experiment:
One. simulated conditions
First group of parameter: image sources to be X-band resolution the be NoerdlingerRies image of 1 meter, the size based on the convolution kernel of the convolutional neural networks in small echo pond is parameter beta=1 in 5*5, Markov random field models, each neighborhood of pixels N ssize be 3*3;
Second group of parameter: image sources to be Ku wave band resolution the be Piperiver image of 1 meter, the size based on the convolution kernel of the convolutional neural networks in small echo pond is parameter beta=1 in 5*5, Markov random field models, each neighborhood of pixels N ssize be 3*3.
Two. emulation content:
Emulation 1, with the inventive method and existing Markov random field models, convolutional neural networks and based on the convolutional neural networks in small echo pond respectively to the Image Segmentation Using of first group of parameter, result is as Fig. 5.Wherein Fig. 5 (a) is former figure, Fig. 5 (b) is the segmentation result of Markov random field models, the segmentation result that Fig. 5 (c) is convolutional neural networks, Fig. 5 (d) is the segmentation result of the convolutional neural networks based on small echo pond, Fig. 5 (e) is result of the present invention, Fig. 5 (f) for authentic signature figure, Fig. 5 (g) be the corresponding optical imagery of this SAR image.
Emulation 2, with the inventive method and existing Markov random field models, convolutional neural networks and based on the convolutional neural networks in small echo pond respectively to the Image Segmentation Using of second group of parameter, result is as Fig. 6, wherein Fig. 6 (a) is former figure, Fig. 6 (b) is the segmentation result of Markov random field models, the segmentation result that Fig. 6 (c) is convolutional neural networks, Fig. 6 (d) is the segmentation result of the convolutional neural networks based on small echo pond, Fig. 6 (e) is result of the present invention, Fig. 6 (f) is authentic signature figure, Fig. 6 (g) is the corresponding optical imagery of this SAR image.
Three. analysis of simulation result:
Maintain the boundary information of image as can be seen from Fig. 5 and Fig. 6, Markov random field models preferably, but easily produce the result of over-segmentation, this causes due to predefined spatial context model; Convolutional neural networks consistance in the segmentation result of heterogeneous areas is better, this is because convolutional neural networks can learn the architectural feature of image preferably, but due to the structure of homogenous region more weak, so convolutional neural networks creates wrong point in homogenous region; Convolutional neural networks based on small echo pond can make study to feature keep more structure, therefore its consistance in the segmentation result of heterogeneous areas is better than convolutional neural networks; Present invention incorporates the advantage based on small echo pond convolutional neural networks and Markov two algorithms, therefore result not only has good region consistency but also can positioning boundary accurately.
In sum, present invention achieves the balance between region consistency and accurate boundary alignment in SAR image segmentation, obtain the segmentation effect that SAR image is good.
The present embodiment does not have specifically described part all to belong to common practise and the known technology of the art; and above exemplifying is only illustrate of the present invention; do not form the restriction to protection scope of the present invention, everyly all to belong within protection scope of the present invention with the same or analogous design of the present invention.

Claims (4)

1., based on the SAR image segmentation method of small echo pond convolutional neural networks, comprising:
(1) in convolutional neural networks, carry out wavelet transformation to characteristic pattern and get its approximation subband forming small echo pond layer, utilize small echo pond layer to form the convolutional neural networks in small echo pond;
(2) according to SAR image signature, every class chooses the pixel of 50%, and is input to by the image block centered by this pixel in the convolutional neural networks in small echo pond and trains;
(3) to all pixels of SAR image, the image block chosen centered by this pixel is input in the network trained and tests, and obtains the first kind mark of SAR image;
(4) super-pixel segmentation is carried out to SAR image, obtain the super-pixel segmentation result of SAR image, and the first kind mark of itself and SAR image is merged, obtain the Equations of The Second Kind mark of SAR image;
(5) obtain the 3rd class mark of SAR image according to Markov random field models, and the super-pixel segmentation result of itself and SAR image is merged, obtain the 4th class mark of SAR image;
(6) according to SAR image gradient map, the Equations of The Second Kind mark of SAR image and the 4th class mark are merged, obtains final segmentation result.
2. the SAR image segmentation method of convolutional neural networks according to claim 1, carry out wavelet transformation to characteristic pattern in wherein said step (1) and get its approximation subband forming small echo pond layer, carry out as follows:
(1.1) according to the following formula to characteristic pattern CF kcarry out wavelet transformation,
{ SF k ϵ ( x , y ) } ϵ = a , h , v , d = d o w n ( CF k η ( x , y ) * ψ )
Wherein k=1 ..., N, N are the number of characteristic pattern, characteristic pattern CF kin neighborhood block centered by point (x, y), * represents convolution operation, and ψ is wavelet basis function, lower 2 sampling operations of down () representative, be point in each subband that wavelet transformation obtains, a, h, v, d represent approximation subband, horizontal subband, vertical subband and diagonal angle subband respectively;
(1.2) approximation subband characteristic pattern is used form small echo pond layer, wherein k=1 ..., N, N are the number of characteristic pattern.
3. SAR image segmentation method according to claim 1, in wherein said step (4), the super-pixel segmentation result of SAR image and the first kind mark of SAR image are merged, be first according to the first kind mark of the SAR image obtained in step (3), calculate number of pixels of all categories in each super-pixel; Each super-pixel is labeled as that maximum classification of number of pixels in this super-pixel again, obtains the Equations of The Second Kind mark of SAR image.
4. the SAR image segmentation method based on small echo pond convolutional neural networks according to claim 1, according to SAR image gradient map in wherein said step (6), the Equations of The Second Kind mark of SAR image and the 4th class mark are merged, obtain final segmentation result, carry out as follows:
(6.1) ratio operator r, right-angled intersection operator c and first difference operator y are merged, obtain the operation operator f after merging:
f = x y 1 - x - y + 2 x y
Wherein, x = r 2 + c 2 2 ;
(6.2) the operation operator f after fusion and SAR image are carried out convolution, obtain SAR image gradient map, and the gradient of pixel (x, y) is designated as G (x, y).
(6.3) for pixel (x, y) gradient G (x, y) threshold value T=185 is set, and G (x, y) and T are compared, if G (x, y) >=T, then using the Equations of The Second Kind mark of this pixel as final class mark, otherwise, using the 4th class mark of this pixel as final class mark.
CN201510512535.1A 2015-08-19 2015-08-19 SAR image segmentation method based on small echo pond convolutional neural networks Active CN105139395B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510512535.1A CN105139395B (en) 2015-08-19 2015-08-19 SAR image segmentation method based on small echo pond convolutional neural networks

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510512535.1A CN105139395B (en) 2015-08-19 2015-08-19 SAR image segmentation method based on small echo pond convolutional neural networks

Publications (2)

Publication Number Publication Date
CN105139395A true CN105139395A (en) 2015-12-09
CN105139395B CN105139395B (en) 2018-03-06

Family

ID=54724727

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510512535.1A Active CN105139395B (en) 2015-08-19 2015-08-19 SAR image segmentation method based on small echo pond convolutional neural networks

Country Status (1)

Country Link
CN (1) CN105139395B (en)

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105447190A (en) * 2015-12-18 2016-03-30 小米科技有限责任公司 Image retrieval method, device and server based on convolutional neural network
CN105718957A (en) * 2016-01-26 2016-06-29 西安电子科技大学 Polarized SAR image classification method based on nonsubsampled contourlet convolutional neural network
CN105740894A (en) * 2016-01-28 2016-07-06 北京航空航天大学 Semantic annotation method for hyperspectral remote sensing image
CN106023145A (en) * 2016-05-06 2016-10-12 哈尔滨工程大学 Remote sensing image segmentation and identification method based on superpixel marking
CN106407986A (en) * 2016-08-29 2017-02-15 电子科技大学 Synthetic aperture radar image target identification method based on depth model
CN106846278A (en) * 2017-02-17 2017-06-13 深圳市唯特视科技有限公司 A kind of image pixel labeling method based on depth convolutional neural networks
CN107203999A (en) * 2017-04-28 2017-09-26 北京航空航天大学 A kind of skin lens image automatic division method based on full convolutional neural networks
CN107292256A (en) * 2017-06-14 2017-10-24 西安电子科技大学 Depth convolved wavelets neutral net expression recognition method based on secondary task
CN107341488A (en) * 2017-06-16 2017-11-10 电子科技大学 A kind of SAR image target detection identifies integral method
CN107480707A (en) * 2017-07-26 2017-12-15 天津大学 A kind of deep neural network method based on information lossless pond
CN107578067A (en) * 2017-09-15 2018-01-12 国家烟草质量监督检验中心 A kind of recognition methods of pipe tobacco component and device based on convolutional neural networks
CN107832794A (en) * 2017-11-09 2018-03-23 车智互联（北京）科技有限公司 A kind of convolutional neural networks generation method, the recognition methods of car system and computing device
CN108648161A (en) * 2018-05-16 2018-10-12 江苏科技大学 The binocular vision obstacle detection system and method for asymmetric nuclear convolutional neural networks
CN108931771A (en) * 2018-06-06 2018-12-04 电子科技大学 A kind of method for tracking target based on synthetic aperture radar image-forming technology
CN109583333A (en) * 2018-11-16 2019-04-05 中证信用增进股份有限公司 Image-recognizing method based on water logging method and convolutional neural networks
CN109993050A (en) * 2018-11-28 2019-07-09 中国人民解放军国防科技大学 Synthetic aperture radar image identification method
CN111553421A (en) * 2020-04-28 2020-08-18 电子科技大学 SAR equipment task failure cause reasoning method based on double-layer nested structure

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101350099A (en) * 2008-09-12 2009-01-21 西安电子科技大学 Method for segmenting image based on wavelet domain concealed Markov tree model
CN101587587A (en) * 2009-07-14 2009-11-25 武汉大学 The segmentation method for synthetic aperture radar images of consideration of multi-scale Markov field
US20120321186A1 (en) * 2011-06-20 2012-12-20 Massachusetts Institute Of Technology Efficient mcmc sampling with implicit shape representations
CN103955702A (en) * 2014-04-18 2014-07-30 西安电子科技大学 SAR image terrain classification method based on depth RBF network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101350099A (en) * 2008-09-12 2009-01-21 西安电子科技大学 Method for segmenting image based on wavelet domain concealed Markov tree model
CN101587587A (en) * 2009-07-14 2009-11-25 武汉大学 The segmentation method for synthetic aperture radar images of consideration of multi-scale Markov field
US20120321186A1 (en) * 2011-06-20 2012-12-20 Massachusetts Institute Of Technology Efficient mcmc sampling with implicit shape representations
CN103955702A (en) * 2014-04-18 2014-07-30 西安电子科技大学 SAR image terrain classification method based on depth RBF network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
WU J ET AL: "Local maximal homogeneous region search for SAR speckle reduction with skech-based geometrical kernel function", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 *
权金娟: "基于神经网络的SAR图像多尺度分割的研究", 《万方在线学位论文》 *

Cited By (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105447190B (en) * 2015-12-18 2019-03-15 小米科技有限责任公司 Picture retrieval method, device and server based on convolutional neural networks
CN105447190A (en) * 2015-12-18 2016-03-30 小米科技有限责任公司 Image retrieval method, device and server based on convolutional neural network
CN105718957A (en) * 2016-01-26 2016-06-29 西安电子科技大学 Polarized SAR image classification method based on nonsubsampled contourlet convolutional neural network
CN105740894A (en) * 2016-01-28 2016-07-06 北京航空航天大学 Semantic annotation method for hyperspectral remote sensing image
CN105740894B (en) * 2016-01-28 2020-05-29 北京航空航天大学 Semantic annotation method for hyperspectral remote sensing image
CN106023145A (en) * 2016-05-06 2016-10-12 哈尔滨工程大学 Remote sensing image segmentation and identification method based on superpixel marking
CN106407986A (en) * 2016-08-29 2017-02-15 电子科技大学 Synthetic aperture radar image target identification method based on depth model
CN106407986B (en) * 2016-08-29 2019-07-19 电子科技大学 A kind of identification method of image target of synthetic aperture radar based on depth model
CN106846278A (en) * 2017-02-17 2017-06-13 深圳市唯特视科技有限公司 A kind of image pixel labeling method based on depth convolutional neural networks
CN107203999A (en) * 2017-04-28 2017-09-26 北京航空航天大学 A kind of skin lens image automatic division method based on full convolutional neural networks
CN107203999B (en) * 2017-04-28 2020-01-24 北京航空航天大学 Dermatoscope image automatic segmentation method based on full convolution neural network
CN107292256B (en) * 2017-06-14 2019-12-24 西安电子科技大学 Auxiliary task-based deep convolution wavelet neural network expression recognition method
CN107292256A (en) * 2017-06-14 2017-10-24 西安电子科技大学 Depth convolved wavelets neutral net expression recognition method based on secondary task
CN107341488A (en) * 2017-06-16 2017-11-10 电子科技大学 A kind of SAR image target detection identifies integral method
CN107480707B (en) * 2017-07-26 2020-08-07 天津大学 Deep neural network method based on information lossless pooling
CN107480707A (en) * 2017-07-26 2017-12-15 天津大学 A kind of deep neural network method based on information lossless pond
CN107578067A (en) * 2017-09-15 2018-01-12 国家烟草质量监督检验中心 A kind of recognition methods of pipe tobacco component and device based on convolutional neural networks
CN107832794B (en) * 2017-11-09 2020-07-14 车智互联（北京）科技有限公司 Convolutional neural network generation method, vehicle system identification method and computing device
CN107832794A (en) * 2017-11-09 2018-03-23 车智互联（北京）科技有限公司 A kind of convolutional neural networks generation method, the recognition methods of car system and computing device
CN108648161A (en) * 2018-05-16 2018-10-12 江苏科技大学 The binocular vision obstacle detection system and method for asymmetric nuclear convolutional neural networks
CN108931771A (en) * 2018-06-06 2018-12-04 电子科技大学 A kind of method for tracking target based on synthetic aperture radar image-forming technology
CN109583333A (en) * 2018-11-16 2019-04-05 中证信用增进股份有限公司 Image-recognizing method based on water logging method and convolutional neural networks
CN109993050A (en) * 2018-11-28 2019-07-09 中国人民解放军国防科技大学 Synthetic aperture radar image identification method
CN111553421A (en) * 2020-04-28 2020-08-18 电子科技大学 SAR equipment task failure cause reasoning method based on double-layer nested structure
CN111553421B (en) * 2020-04-28 2022-03-08 电子科技大学 SAR equipment task failure cause reasoning method based on double-layer nested structure

Also Published As

Publication number Publication date
CN105139395B (en) 2018-03-06

Similar Documents

Publication Publication Date Title
CN105139395A (en) SAR image segmentation method based on wavelet pooling convolutional neural networks
Sony et al. A systematic review of convolutional neural network-based structural condition assessment techniques
CN110705457B (en) Remote sensing image building change detection method
CN109493346B (en) Stomach cancer pathological section image segmentation method and device based on multiple losses
CN108596055B (en) Airport target detection method of high-resolution remote sensing image under complex background
CN109784283B (en) Remote sensing image target extraction method based on scene recognition task
CN105608698B (en) A kind of method for detecting change of remote sensing image based on SAE
CN110516539A (en) Remote sensing image building extracting method, system, storage medium and equipment based on confrontation network
CN106127204A (en) A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks
CN107451997A (en) A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning
CN103294792B (en) Based on the polarization SAR terrain classification method of semantic information and polarization decomposing
CN105930868A (en) Low-resolution airport target detection method based on hierarchical reinforcement learning
CN110009095A (en) Road driving area efficient dividing method based on depth characteristic compression convolutional network
CN101350099A (en) Method for segmenting image based on wavelet domain concealed Markov tree model
CN107423747A (en) A kind of conspicuousness object detection method based on depth convolutional network
CN113312993B (en) Remote sensing data land cover classification method based on PSPNet
CN106683102A (en) SAR image segmentation method based on ridgelet filters and convolution structure model
CN109829507B (en) Aerial high-voltage transmission line environment detection method
CN104850677A (en) Deep learning cellular automaton simulation analysis method based on geographic information system
CN104680167A (en) Aurora oval position determining method based on deep learning
CN103226825B (en) Based on the method for detecting change of remote sensing image of low-rank sparse model
CN106504219B (en) Constrained path morphology high-resolution remote sensing image road Enhancement Method
CN105550712A (en) Optimized convolution automatic encoding network-based auroral image sorting method
CN112818777B (en) Remote sensing image target detection method based on dense connection and feature enhancement
CN104036491A (en) SAR image segmentation method based on area division and self-adaptive polynomial implicit model

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant