CN105139395B - SAR image segmentation method based on small echo pond convolutional neural networks - Google Patents
SAR image segmentation method based on small echo pond convolutional neural networks Download PDFInfo
- Publication number
- CN105139395B CN105139395B CN201510512535.1A CN201510512535A CN105139395B CN 105139395 B CN105139395 B CN 105139395B CN 201510512535 A CN201510512535 A CN 201510512535A CN 105139395 B CN105139395 B CN 105139395B
- Authority
- CN
- China
- Prior art keywords
- sar image
- category
- pixel
- small echo
- convolutional neural
- 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.)
- Active
Links
Landscapes
- Radar Systems Or Details Thereof (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of SAR image segmentation method based on small echo pond convolutional neural networks.Its scheme is:1. structure small echo pond layer simultaneously forms the convolutional neural networks in small echo pond;It is trained 2. choosing image block and being input in the convolutional neural networks in small echo pond;Tested 3. all image blocks are input in the network trained, obtain the category of SAR image first;4. a pair SAR image makees super-pixel segmentation, its result is merged with the category of SAR image first, obtains the category of SAR image second;5. obtaining the category of SAR image the 3rd according to Markov random field models, and it is merged with super-pixel segmentation result, obtain the category of SAR image the 4th;6. according to SAR image gradient map, the second category of SAR image and the 4th category are merged, obtain final segmentation result.The present invention improves the segmentation effect of SAR image, available for Target detection and identification.
Description
Technical field
The invention belongs to technical field of image processing, is related to SAR image segmentation method, available for Target detection and identification.
Background technology
With the fast development of remote sensing technology, SAR image interpretation plays increasingly heavier in the energy, environment, archaeology etc.
The effect wanted.SAR image segmentation be SAR image interpretation in basis and an important ring, it can be follow-up classification, detection,
Identification and tracking provide help.The main target of SAR image segmentation is that SAR image is divided into the connected region do not occured simultaneously,
Completing this target needs that each pixel of SAR image is marked, therefore is an extremely difficult task.
Existing SAR image segmentation method is broadly divided into the method for feature based and the method based on statistical model.It is based on
The method of feature is mainly that the feature for extracting some SAR images is split, such as textural characteristics, side feature and composite character
Deng.These methods achieve preferable segmentation effect in the homogenous region of SAR image.However, SAR image is earth observation, always
It is that can observe the complex scenes such as some cities, forest.The light and shade change that these complicated scenes invariably accompany sharp, it is this
Characteristic is challenged the method for feature based.Based on the method for statistical model by the mode of SAR image segmentation problem probability
Expression, because such method considers the spatial context relation of image, recent years is of increased attention.It is based on
The feature of image is described as the distribution of some experiences by the method for statistical model, such as Nakagami distributions, Gamma are distributed, K points
Cloth, G distributions etc..Spatial context model is described as Gibbs distributions, multinomial logistic regression function etc..In fact, SAR image
Segmentation problem transforms into the parameter problem for asking these distributions or function.Parametric solution is generally solved with the method for Combinatorial Optimization
Certainly, this is a time-consuming process, and many times can not always find satisfactory solution.Although the method based on model achieves
Preferable segmentation effect, for the region of some complicated grand textures, they still can not be divided into the region of uniformity.This two
Class method is all that the segmentation of SAR image is carried out under artificially given feature, and convolutional neural networks can from data in itself by
The architectural feature of the study image of layer, the uniformity of heterogeneous areas in cutting procedure can be improved.However, passed in convolutional neural networks
The pond method of system, i.e., maximum pond, average Chi Huafa, simply simply takes maximum or flat in the neighborhood window of characteristic pattern
Average, certain destruction can be caused to the structure of the feature learnt, is unfavorable for the segmentation of SAR image.
The content of the invention
It is an object of the invention to for above-mentioned existing methods deficiency, propose that one kind is based on small echo pond convolutional Neural net
The SAR image segmentation method of network, to lift the effect of SAR image segmentation.
To achieve the above object, technical scheme is as follows:
(1) in convolutional neural networks, wavelet transformation is carried out to characteristic pattern and takes its approximation subband to form small echo pond layer,
The convolutional neural networks in small echo pond are formed using small echo pond layer;
(2) marked and schemed according to SAR image, 50% pixel is chosen per class, and the image block centered on the pixel is defeated
Enter into the convolutional neural networks in small echo pond and be trained;
(3) to all pixels of SAR image, choose the image block centered on the pixel and be input in the network trained
Tested, obtain the first category of SAR image;
(4) super-pixel segmentation is carried out to SAR image, obtains the super-pixel segmentation result of SAR image, and itself and SAR are schemed
The first category fusion of picture, obtains the second category of SAR image;
(5) the 3rd category of SAR image is obtained according to Markov random field models, and by its super-pixel with SAR image
Segmentation result merges, and obtains the 4th category of SAR image;
(6) according to SAR image gradient map, the second category of SAR image and the 4th category is merged, obtained final
Segmentation result.
The present invention has the following advantages that compared with prior art:
First, the traditional tank method in convolutional neural networks, i.e., maximum pond, average Chi Huafa, simply simply exist
Maximum or average value are taken in the neighborhood of characteristic pattern, certain destruction can be caused to the architectural feature learnt, the present invention uses
Small echo pondization not only there is the function of noise reduction, and can preferably keep architectural feature learn, this divides for image
It is vital to cut.
Second, the convolutional neural networks based on small echo pond of the invention can from the automatic learning characteristic of initial data, without
The type of artificial given feature is needed, this abundant architectural feature learnt from initial data can improve heterogeneous areas,
Such as the uniformity of the segmentation result such as city, forest.
3rd, the convergence strategy based on SAR image gradient of the invention, by the convolutional neural networks based on small echo pond and
The advantages of Markov random field models, is merged, and making to be segmented in again can be with holding area while border is precisely located
Uniformity.
Brief description of the drawings
Fig. 1 is the implementation process 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 small echo pond convolutional neural networks learn in the present invention;
Fig. 4 is the SAR image mark figure that present invention emulation uses;
Fig. 5 is the segmentation with existing method to X-band resolution ratio for 1 meter of Noerdlinger Ries images with the present invention
Result figure;
Fig. 6 is the segmentation result with existing method to Ku wave bands resolution ratio for 1 meter of Piperiver images with the present invention
Figure.
Embodiment
Reference picture 1, implementation steps of the invention are as follows:
Step 1, build small echo pond layer and form the convolutional neural networks in small echo pond.
(1.1) according to the following formula to characteristic pattern CFkCarry out wavelet transformation,
Wherein k=1 ..., N, N are characterized the number of figure:CFk η(x,y)It is characteristic pattern CFkIn neighbour centered on point (x, y)
Domain block, * represent convolution operation, and ψ is wavelet basis function, and down () represents lower 2 sampling operations,It isPoint in each subband obtained by wavelet transformation, a, h, v, d represent approximation subband, horizontal subband, vertical son respectively
Band and diagonal subband;
(1.2) approximation subband characteristic pattern SF is usedk aSmall echo pond layer is formed, wherein k=1 ..., N, N are characterized of figure
Number;
(1.3) convolutional neural networks in small echo pond are formed using small echo pond layer, as shown in Fig. 2 the neutral net is
One eight layers of neutral net for including input layer, convolutional layer, pond layer, full articulamentum and output layer, wherein, first layer is input
Layer, the second layer be convolutional layer with the 4th layer, and third layer and layer 5 are pond layer, and layer 6 and layer 7 are full articulamentum, the
Eight layers are output layer.
Step 2, selection image block, which is input in the convolutional neural networks in small echo pond, is trained.
(2.1) in SAR image mark figure as shown in Figure 4,50% pixel is chosen per class, and each pixel is selected
Take using 32 × 32 image block centered on it as training sample, wherein Fig. 4 (a) is Noerdlinger Ries images, Fig. 4
(b) it is the mark figure of Noerdlinger Ries images, classification number is that 5, Fig. 4 (c) is Piperiver images, and Fig. 4 (d) is
The mark figure of Piperiver images, classification number are 5;
(2.2) training sample is input to the convolutional neural networks in the small echo pond of step (1) structure, and uses back-propagating
The parameter of Algorithm Learning network, until all training sample training finish.
Step 3, all image blocks are input in the network trained and tested.
(3.1) to all pixels in SAR image, choose 32 × 32 image block using centered on the pixel and be used as test specimens
This;
(3.2) test sample is input in the neutral net trained in step (2), characteristic pattern is arrived in study, such as Fig. 3 institutes
Show, wherein Fig. 3 (a) is test sample type when being forest, and the second layer and the 4th layer of characteristic pattern, Fig. 3 (b) that e-learning arrives are
When test sample type is city, the second layer and the 4th layer of characteristic pattern that e-learning arrives, Fig. 3 (c) is that test sample type is
During farmland 1, the second layer and the 4th layer of characteristic pattern that e-learning arrives, Fig. 3 (d) is test sample type when being farmland 2, network science
The second layer practised and the 4th layer of characteristic pattern;
(3.3) propagated forward algorithm is utilized, prediction obtains the category of each image block, that is, obtains each picture in SAR image
First category of element.
Step 4, super-pixel segmentation is carried out to SAR image, obtains the second category of the SAR image.
The common method of image superpixel segmentation has:Watershed, MeanShift and TurboPixels etc..Adopted in the present invention
With but the method that is not limited to TurboPixels, obtain the segmentation result of SAR image super-pixel;
According to the first category of the SAR image obtained in step (3), pixel of all categories in each super-pixel is calculated
Number;
By each super-pixel labeled as that most classification of number of pixels in the super-pixel, the second of SAR image is obtained
Category.
Step 5, the 3rd category of SAR image is obtained according to Markov random field models.
(5.1) the prior probability formula of Markov random field models is as follows:
Wherein, xsThe category of representative image pixel, L={ 1 ..., m } represent the classification of image, u (xs) be defined as follows:
Wherein, NsRepresent xsCentered on neighborhood territory pixel, β is model parameter,
(5.2) the likelihood probability formula of Markov random field models is as follows:
Wherein, ysThe pixel value of representative image, vl, μlFor model parameter, Γ () is distributed for Gamma.
(5.3) posterior probability formula of Markov random field models is as follows:
(5.4) the 3rd category of SAR image is obtained according to Markov random field modelsWherein L=1 ..., m }
Represent the classification of image.
Step 6, the 3rd category of SAR image is merged with super-pixel segmentation result, obtains the 4th class of the SAR image
Mark.
(6.1) according to the 3rd category of the SAR image obtained in step (5), picture of all categories in each super-pixel is calculated
Plain number;
(6.2) each super-pixel is obtained into the SAR image labeled as that most classification of number of pixels in the super-pixel
The 4th category.
Step 7, SAR image gradient map is obtained, the second category of SAR image and the 4th category are merged, obtained most
Whole category.
(7.1) ratio operator r, right-angled intersection operator c and first difference operator ξ are merged, the behaviour after being merged
Make operator f:
Wherein,
(7.2) the operation operator f after fusion and SAR image are subjected to convolution, obtain SAR image gradient map, and by pixel
The gradient of (x, y) is designated as G (x, y);
(7.3) gradient G (x, y) given threshold T=185 of pixel (x, y) is directed to, and by G (x, y) compared with T, such as
Fruit G (x, y) >=T, then using the second category of the pixel as final category, otherwise, using the 4th category of the pixel as final
Category.
The effect of the present invention can be illustrated by following emulation experiment:
One, simulated conditions
First group of parameter:Image sources are the Noerdlinger Ries images that X-band resolution ratio is 1 meter, based on small echo
The size of the convolution kernel of the convolutional neural networks in pond is 5*5, parameter beta=1 in Markov random field models, each neighborhood of pixels
NsSize be 3*3;
Second group of parameter:Image sources are the Piperiver images that Ku wave band resolution ratio is 1 meter, based on small echo pond
The size of the convolution kernel of convolutional neural networks is 5*5, parameter beta=1 in Markov random field models, each neighborhood of pixels NsIt is big
Small is 3*3.
Two, emulation contents:
Emulation 1, with the inventive method and existing Markov random field models, convolutional neural networks and based on small echo pond
Convolutional neural networks the image of first group of parameter is split respectively, as a result such as Fig. 5.Wherein Fig. 5 (a) is artwork, Fig. 5
(b) it is the segmentation result of Markov random field models, Fig. 5 (c) is the segmentation result of convolutional neural networks, and Fig. 5 (d) is based on small
The segmentation result of the convolutional neural networks in ripple pond, Fig. 5 (e) are the result of the present invention, and Fig. 5 (f) is authentic signature figure, Fig. 5 (g)
For the corresponding optical imagery of the SAR image.
Emulation 2, with the inventive method and existing Markov random field models, convolutional neural networks and based on small echo pond
Convolutional neural networks the image of second group of parameter is split respectively, as a result such as Fig. 6, wherein Fig. 6 (a) is artwork, Fig. 6
(b) it is the segmentation result of Markov random field models, Fig. 6 (c) is the segmentation result of convolutional neural networks, and Fig. 6 (d) is based on small
The segmentation result of the convolutional neural networks in ripple pond, Fig. 6 (e) are the result of the present invention, and Fig. 6 (f) is authentic signature figure, Fig. 6 (g)
For the corresponding optical imagery of the SAR image.
Three, analysiss of simulation result:
From figs. 5 and 6, it can be seen that Markov random field models preferably maintain the boundary information of image, but hold
The result of over-segmentation is also easy to produce, caused by this is due to pre-defined spatial context model;Convolutional neural networks are heterogeneous
Uniformity is preferable in the segmentation result in region, because convolutional neural networks can preferably learn the architectural feature of image,
It is weaker yet with the structure of homogenous region, so convolutional neural networks generate wrong point in homogenous region;Based on small echo pond
Convolutional neural networks the feature that study arrives can be made to keep more structures, therefore it is one in the segmentation result of heterogeneous areas
Cause property is more preferable than convolutional neural networks;Present invention incorporates based on two algorithms of small echo pond convolutional neural networks and Markov
Advantage, therefore result not only has preferable region consistency but also can accurately position border.
In summary, the present invention realizes flat between region consistency and accurate boundary alignment in SAR image segmentation
Weighing apparatus, obtains the good segmentation effect of SAR image.
The present embodiment belongs to the common knowledge and known technology of the art without the part specifically described, and more than
Enumerate only to the present invention for example, do not form the limitation to protection scope of the present invention, it is every with phase of the present invention
Same or similar design is belonged within protection scope of the present invention.
Claims (3)
1. based on the SAR image segmentation method of small echo pond convolutional neural networks, including:
(1) in convolutional neural networks, wavelet transformation is carried out to characteristic pattern and takes its approximation subband to form small echo pond layer, is utilized
Small echo pond layer forms the convolutional neural networks in small echo pond;
(2) marked and schemed according to SAR image, 50% pixel is chosen per class, and the image block centered on the pixel is input to
It is trained in the convolutional neural networks in small echo pond;
(3) to all pixels of SAR image, choose the image block centered on the pixel and be input in the network trained and carry out
Test, obtains the first category of SAR image;
(4) super-pixel segmentation is carried out to SAR image, obtains the super-pixel segmentation result of SAR image, and by itself and SAR image
First category merges, and obtains the second category of SAR image;
(5) the 3rd category of SAR image is obtained according to Markov random field models, and by its super-pixel segmentation with SAR image
As a result merge, obtain the 4th category of SAR image;
(6) according to SAR image gradient map, the second category of SAR image and the 4th category is merged, obtain final segmentation
As a result:
(6.1) ratio operator r, right-angled intersection operator c and first difference operator ξ are merged, the operation after being merged is calculated
Sub- f:
Wherein,
(6.2) the operation operator f after fusion and SAR image are subjected to convolution, obtain SAR image gradient map, and by pixel (x, y)
Gradient be designated as G (x, y);
(6.3) gradient G (x, y) given threshold T=185 of pixel (x, y) is directed to, and by G (x, y) compared with T, if G
(x, y) >=T, then using the second category of the pixel as final category, otherwise, using the 4th category of the pixel as final class
Mark.
2. the SAR image segmentation method of convolutional neural networks according to claim 1, wherein to spy in the step (1)
Sign figure carries out wavelet transformation and takes its approximation subband to form small echo pond layer, carries out as follows:
(1.1) according to the following formula to characteristic pattern CFkCarry out wavelet transformation,
Wherein k=1 ..., N, N are characterized the number of figure,It is characteristic pattern CFkIn neighborhood centered on point (x, y)
Block, * represent convolution operation, and ψ is wavelet basis function, and down () represents lower 2 sampling operations,It isPoint in each subband obtained by wavelet transformation, a, h, v, d represent approximation subband, horizontal subband, vertical son respectively
Band and diagonal subband;
(1.2) approximation subband characteristic pattern is usedSmall echo pond layer is formed, wherein k=1 ..., N, N are characterized the number of figure.
3. SAR image segmentation method according to claim 1, wherein by the super-pixel of SAR image point in the step (4)
Cut result to merge with the first category of SAR image, be first to be calculated every according to the first category of the SAR image obtained in step (3)
Number of pixels of all categories in individual super-pixel;Again by each super-pixel labeled as that most class of number of pixels in the super-pixel
Not, the second category of SAR image is obtained.
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 CN105139395A (en) | 2015-12-09 |
CN105139395B true 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) |
Families Citing this family (17)
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 |
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 |
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 |
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 |
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 |
CN107341488B (en) * | 2017-06-16 | 2020-02-18 | 电子科技大学 | SAR image target detection and identification integrated method |
CN107480707B (en) * | 2017-07-26 | 2020-08-07 | 天津大学 | Deep neural network method based on information lossless pooling |
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 |
CN108648161B (en) * | 2018-05-16 | 2020-09-01 | 江苏科技大学 | Binocular vision obstacle detection system and method of asymmetric kernel convolution neural network |
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 |
CN109583333B (en) * | 2018-11-16 | 2020-12-11 | 中证信用增进股份有限公司 | Image identification method based on flooding method and convolutional neural network |
CN109993050B (en) * | 2018-11-28 | 2019-12-27 | 中国人民解放军国防科技大学 | Synthetic aperture radar image identification method |
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 |
Citations (3)
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 |
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 |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8744185B2 (en) * | 2011-06-20 | 2014-06-03 | Massachusetts Institute Of Technology | Efficient MCMC sampling with implicit shape representations |
-
2015
- 2015-08-19 CN CN201510512535.1A patent/CN105139395B/en active Active
Patent Citations (3)
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 |
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)
Title |
---|
Local maximal homogeneous region search for SAR speckle reduction with skech-based geometrical kernel function;Wu J et al;《IEEE transactions on Geoscience and Remote Sensing》;20141231;第5751-5764页 * |
基于神经网络的SAR图像多尺度分割的研究;权金娟;《万方在线学位论文》;20081229;第31-40页 * |
Also Published As
Publication number | Publication date |
---|---|
CN105139395A (en) | 2015-12-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105139395B (en) | SAR image segmentation method based on small echo pond convolutional neural networks | |
CN111986099B (en) | Tillage monitoring method and system based on convolutional neural network with residual error correction fused | |
CN108647585B (en) | Traffic identifier detection method based on multi-scale circulation attention network | |
CN106157319B (en) | The conspicuousness detection method in region and Pixel-level fusion based on convolutional neural networks | |
CN107092870B (en) | A kind of high resolution image Semantic features extraction method | |
CN110852316B (en) | Image tampering detection and positioning method adopting convolution network with dense structure | |
CN107918776B (en) | Land planning method and system based on machine vision and electronic equipment | |
CN108596055B (en) | Airport target detection method of high-resolution remote sensing image under complex background | |
CN108710863A (en) | Unmanned plane Scene Semantics dividing method based on deep learning and system | |
CN111161311A (en) | Visual multi-target tracking method and device based on deep learning | |
CN108537824B (en) | Feature map enhanced network structure optimization method based on alternating deconvolution and convolution | |
CN107480726A (en) | A kind of Scene Semantics dividing method based on full convolution and shot and long term mnemon | |
CN113192040A (en) | Fabric flaw detection method based on YOLO v4 improved algorithm | |
CN107862668A (en) | A kind of cultural relic images restored method based on GNN | |
CN106981080A (en) | Night unmanned vehicle scene depth method of estimation based on infrared image and radar data | |
CN105809194B (en) | A kind of method that SAR image is translated as optical image | |
CN107832797B (en) | Multispectral image classification method based on depth fusion residual error network | |
CN106651887A (en) | Image pixel classifying method based convolutional neural network | |
CN109829507B (en) | Aerial high-voltage transmission line environment detection method | |
CN111986193B (en) | Remote sensing image change detection method, electronic equipment and storage medium | |
CN109858487A (en) | Weakly supervised semantic segmentation method based on watershed algorithm and image category label | |
CN112613350A (en) | High-resolution optical remote sensing image airplane target detection method based on deep neural network | |
CN113971764B (en) | Remote sensing image small target detection method based on improvement YOLOv3 | |
CN115049619A (en) | Efficient flaw detection method for complex scene | |
Maggiolo et al. | Improving maps from CNNs trained with sparse, scribbled ground truths using fully connected CRFs |
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 |