CN107122796A - A kind of remote sensing image sorting technique based on multiple-limb network integration model - Google Patents
A kind of remote sensing image sorting technique based on multiple-limb network integration model Download PDFInfo
- Publication number
- CN107122796A CN107122796A CN201710213577.4A CN201710213577A CN107122796A CN 107122796 A CN107122796 A CN 107122796A CN 201710213577 A CN201710213577 A CN 201710213577A CN 107122796 A CN107122796 A CN 107122796A
- Authority
- CN
- China
- Prior art keywords
- layer
- image
- branch
- input
- main split
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of remote sensing image sorting technique based on multiple-limb network integration model, this method comprises the following steps:S1, builds multiple-limb network integration model, includes a main split S, at least one other branch { Tk, k=1 ..., K, K >=1;S2, collection remote sensing image carries out classification mark as training image, and to training image, and multiple-limb network integration model parameter is gone out from the training image learning marked using back-propagation algorithm;S3, image is input in the multiple-limb network integration model for having learnt parameter by the input layer I in main split S, successively processing and all other branch { T of the image by main split SkSuccessively processing, finally by main split S cutting layer W output category shot chart pictures.The present invention utilizes the multiple-limb network integration shallow-layer and deep layer characteristics of image, drastically increases remote sensing image nicety of grading, and can once generate the intensive semantic segmentation result of size identical with input picture, improves image classification efficiency.
Description
Technical field
This hair belongs to the technical field of remote sensing image classification, more particularly to a kind of based on multiple-limb network integration model
Remote sensing image sorting technique.
Background technology
With developing rapidly and extensive use for image sensor technologies, view data explosive growth, image data amount
Magnanimity level is turned into.Conventional artificial image's sorting technique is no longer feasible, therefore is currently based on the figure of computer vision technique
As automatic classification has become a very powerful and exceedingly arrogant research topic.Recent study persons have proposed many image classification sides
Method, but remote sensing image classification at this stage still faces huge challenge:Content complexity, object classification are numerous in remote sensing image
Many, difference is very big in image scene feature class.Previous shallow-layer method does not extract the layering of image using multilayered structure
Feature, realizes merely with shallow-layer feature and shallow-layer grader and classifies, thus the expression ability of shallow-layer method, classification capacity have very much
Limit, image classification accuracy is not high.In recent years, with the development of deep learning, many image classification methods based on deep learning
Also it is suggested, such as SAE (stacked autoencoder), DBN (deep belief network), CNN
(convolutional neural network) even depth model, although the method based on depth model compares shallow-layer method pole
It is big to improve image classification accuracy, but for complicated remote sensing image, nicety of grading still has much room for improvement.It is most of in addition
Depth model, which does not merge layered characteristic, to be used to classify, and only receives the input picture of fixed dimension, it is impossible to once calculated
The intensive semantic segmentation result of (needing by multiple network calculations) size identical with input picture.
The content of the invention
The technical problems to be solved by the invention are:Current depth model most of in the prior art does not merge layering
Feature is used to classify, and only receives the input picture of fixed dimension, it is impossible to which once calculating (needs by multiple network meter
Calculate) the intensive semantic segmentation result of size identical with input picture.
To solve technical problem above, the invention provides a kind of optical remote sensing based on multiple-limb network integration model
Image classification method, the algorithm comprises the following steps:
S1, builds multiple-limb network integration model, includes a main split S, at least one other branch { Tk, k=1 ...,
K, K >=1, wherein K represents the bar number of other branch;
S2, collection remote sensing image carries out classification mark as training image, and to training image, utilizes backpropagation
Algorithm goes out multiple-limb network integration model parameter from the training image learning marked;
S3, image is input to the multiple-limb network integration model for having learnt parameter by the input layer I in main split S
In, successively processing and all other branch { T of the image by main split SkSuccessively processing, finally by the sanction in main split S
Cut a layer W output category shot chart pictures.
Beneficial effects of the present invention:Can by multiple-branching construction in the multiple-limb network integration model that is used in the present invention
Shallow-layer and further feature are merged, and by fusion feature classification remote sensing image, drastically increases nicety of grading, many points
Layer is up-sampled in branch network integration model uses the image for enabling input layer in the inventive method to receive arbitrary dimension defeated
Enter, and can once generate the intensive semantic segmentation result of (without multiple network calculations) size identical with input picture, improve
Image classification efficiency.
Further, the step S2 is specifically, set the number of species of remote sensing image, and classification number is C, gathers light
Remote sensing images are learned as training image, and classification mark is carried out to training image according to the remote sensing image species, are utilized
Back-propagation algorithm goes out multiple-limb network integration model parameter from the training image learning marked.
Further, in the step S1, described main split S is successively comprising input layer I, at least one convolutional layer
{ni, i=1 ..., N, N >=1 }, at least one excitation layer { lj, j=1 ..., L, L >=1 }, at least one dimensionality reduction layer { mp, p=
1 ..., M, M >=1 }, at least one up-sampling layer { uq, q=1 ..., U, U >=1 }, a cutting layer W, as N > 1 or M > 1
Convolutional layer and dimensionality reduction layer cross arrangement, wherein N represent the number of plies of convolutional layer, and M represents the number of plies of dimensionality reduction layer, and L represents excitation layer
The number of plies, U represents to up-sample the number of plies of layer.
Further, in main split S, the excitation layer ljIn, each excitation layer ljAbove one layer is all a convolution
Layer;Last convolutional layer n in main split SNConvolution kernel number be the classification number C.
Further, described every Tiao Pang branches TkSuccessively comprising a convolutional layer ak, one cut layer bk, a superposition
Layer ck, a up-sampling layer dk;Other branch { Tk, k=1 ..., K, K >=1 } in all convolutional layer { ak, k=1 ..., K, K >=1
Convolution kernel number all be the classification number C.
Further, the step S3 includes:
Input layer I in S31, the main split S receives image input;
Convolutional layer n in S32, main split SiOutput image to adjacent last layer performs convolution operation, extracts characteristics of image,
And export characteristic image;
Excitation layer l in S33, main split SjUsing correcting output characteristic image of the linear unit to adjacent last layer convolutional layer
Perform nonlinear function to calculate, and the image for calculating generation is exported;
Dimensionality reduction layer m in S34, main split SpTo adjacent last layer excitation layer ljOutput image perform dimensionality reduction, and by dimensionality reduction figure
As output;
Layer u is up-sampled in S35, main split SqOutput image to adjacent last layer performs deconvolution operation, realizes to figure
The liter dimension of picture, and export a liter dimension image;
S36, K dimensionality reduction layer is selected from main split S, the output image of K dimensionality reduction layer is separately input into other point
Branch { Tk, k=1 ..., K, K >=1 in, other branch's classification is arranged to make up multilevel hierarchy, by the successively meter of Ge Pang branches
Calculate, by branch T by afterbodyKOutput image be input in the cutting layer W in main split S, cutting in main split S layer W
The input picture is cut into the image identical size with the input layer I inputs in main split S, while the image after cutting is made
For the output of whole multiple-limb network integration model, shot chart picture of classifying is obtained.
Above-mentioned further beneficial effect:Above-mentioned main split S is by the processing of each level, and Ge Pang branches are successively
Processing, eventually through output category shot chart picture in cutting layer W last in main split.
Further, the step S32, it is specially:
The convolutional layer niOutput image to adjacent last layer performs convolution operation, if convolutional layer niAdjacent last layer
For excitation layer lj, then to excitation layer ljThe image of output performs convolution operation;If convolutional layer niAdjacent last layer is convolutional layer
ni-1, then to convolutional layer ni-1The image of output performs convolution operation;If convolutional layer niAdjacent last layer is dimensionality reduction layer mp, then it is right
Dimensionality reduction layer mpThe image of output performs convolution operation.
Further, specifically included in the step S36:
S361, K dimensionality reduction layer is selected from main split S, the output image of K dimensionality reduction layer is separately input into other point
Branch { Tk, k=1 ..., K, K >=1 in, i.e. a dimensionality reduction layer m in main split SpOutput image be input to a Ge Pang branches TkIn
Convolutional layer akIn, convolutional layer akConvolution operation is performed to the input picture of this layer;
S362, according to the S361 Zhong Pang branches { Tk, k=1 ..., K, K >=1 } convolutional layer { ak, k=1 ..., K }
The size of input picture, by ascending order by other branch { Tk, k=1 ..., K } and multilevel hierarchy is arranged as, simultaneously
Per Tiao Pang branches TkOutput image by side branch TkIn up-sampling layer dkOutput, and side branch TkOutput image
It is used as the input picture of branch by adjacent next stage;
S363, per Tiao Pang branches TkMiddle cutting layer bkTwo inputs are received, if side branch TkIt is branch by the first order, then
Side branch TkMiddle cutting layer bkFirst input be current other branch TkMiddle convolutional layer akOutput image, second input
It is up-sampling layer u in main split SUOutput;If side branch TkBranch by the first order, then side branch TkMiddle cutting layer
bkFirst input be current other branch TkMiddle convolutional layer akOutput image, second input is branch T by upper levelk-1's
Up-sample layer dk-1Output image;Side branch TkMiddle cutting layer bkFirst input is cut into and second input size phase
Same image, and the image after cutting is exported;
S364, per Tiao Pang branches TkMiddle superimposed layer ckTwo inputs are received, if side branch TkIt is branch by the first order, then
Side branch TkMiddle superimposed layer ckFirst input be current other branch TkMiddle cutting layer bkOutput image, second input
It is up-sampling layer u in main split SUOutput;If side branch TkBranch by the first order, then side branch TkMiddle superimposed layer
ckFirst input be current other branch TkMiddle cutting layer bkOutput image, second input is branch T by upper levelk-1's
Up-sample layer dk-1Output image;Side branch TkMiddle superimposed layer ckTwo input pictures are added up, and will it is cumulative after image
Output;
S365, per Tiao Pang branches TkMiddle up-sampling layer dkTo the superimposed layer c in current other branchkOutput image perform it is anti-
Convolution operation, and the image output after deconvolution is operated;
S366, by branch T by afterbodykIn up-sampling layer dkThe image of output is input to the cutting layer in main split S
In W, the input picture is cut into the image identical with the input layer I inputs in main split S by the cutting layer W in main split S
Size, while the image after cutting obtains shot chart picture of classifying as the output of whole multiple-limb network integration model.
Brief description of the drawings
Fig. 1 is a kind of remote sensing image sorting technique flow chart based on multiple-limb network integration model of the invention.
Fig. 2 is the schematic diagram of multiple-limb network integration model structure of the embodiment of the present invention.
Embodiment
The principle and feature of the present invention are described below in conjunction with accompanying drawing, the given examples are served only to explain the present invention, and
It is non-to be used to limit the scope of the present invention.
As shown in Figure 1, a kind of remote sensing image sorting technique based on multiple-limb network integration model of the present invention should
Algorithm comprises the following steps:
The first step, first builds multiple-limb network integration model, and wherein multiple-limb network integration model is specifically included:One master
Branch S, at least one other branch { Tk, k=1 ..., K, K >=1;Described main split S successively comprising input layer I, extremely
A few convolutional layer { ni, i=1 ..., N, N >=1 }, at least one excitation layer { lj, j=1 ..., L, L >=1 }, at least one drop
Tie up layer { mp, p=1 ..., M, M >=1 }, at least one up-sampling layer { uq, q=1 ..., U, U >=1 }, a cutting layer W, work as N
Convolutional layer and dimensionality reduction layer cross arrangement during > 1 or M > 1, wherein N represent the number of plies of convolutional layer, and M represents the number of plies of dimensionality reduction layer, L tables
Show the number of plies of excitation layer, U represents to up-sample the number of plies of layer;In main split S, the excitation layer ljIn, each excitation layer ljAbove
One layer is all a convolutional layer;Last convolutional layer n in main split SNConvolution kernel number be the classification number C;It is described
Every Tiao Pang branches TkSuccessively comprising a convolutional layer ak, one cut layer bk, a superimposed layer ck, a up-sampling layer dk;Often
The image for receiving two different layers outputs in main split S is inputted and merges the two inputs by Tiao Pang branches as two, is realized
The fusion of the characteristics of image of different levels;Other branch { Tk, k=1 ..., K, K >=1 } in all convolutional layer { ak, k=1 ...,
K, K >=1 } convolution kernel number be all the classification number C;Convolutional layer in the present invention is carried using convolution operation from input picture
Characteristics of image is taken, convolution operation mainly uses convolution collecting image to carry out convolution algorithm.Convolution kernel is generally h × h square,
H represents that the value of each position in the size of convolution kernel, square needs to come out from training image learning;In the present invention
Excitation layer using correct linear unit to image perform nonlinear function calculate, enhance the non-linear property of network;The present invention
In dimensionality reduction layer to image perform dimensionality reduction operation, realize the dimensionality reduction of image;Up-sampling layer in the present invention performs warp to image
Product operation, realizes the liter dimension of image;Two image croppings into identical size, are realized picture size by the cutting layer in the present invention
Normalization;Two imaging importings are realized the fusion of characteristics of image by the superimposed layer in the present invention
Second step, sets the number of species of remote sensing image, and classification number is C, and such as remote sensing image species is to build
Build thing, road, meadow, trees etc.;Remote sensing image is gathered as training image, and according to the remote sensing image species
Classification mark is carried out to training image, goes out multiple-limb network from the training image learning marked using back-propagation algorithm and melts
Matched moulds shape parameter;
3rd step, image is input to the multiple-limb network integration mould for having learnt parameter by the input layer I in main split S
In type, successively processing and all other branch { T of the image by main split SkSuccessively processing, finally by main split S
Cut layer W output category shot chart pictures;Wherein start, the input layer I in the main split S receives image input;Then, main point
Convolutional layer n in branch SiOutput image to adjacent last layer performs convolution operation, extracts characteristics of image, and characteristic image is defeated
Go out;Followed by excitation layer lj is held using linear unit is corrected to the output characteristic image of adjacent last layer convolutional layer in main split S
Row nonlinear function is calculated, and the image for calculating generation is exported;Then, dimensionality reduction layer m in main split SpTo adjacent upper one
Layer excitation layer ljOutput image perform dimensionality reduction, and dimensionality reduction image is exported;Subsequently, layer u is up-sampled in main split SqTo adjacent
The output image of last layer performs deconvolution operation, realizes and the liter of image is tieed up, and exports a liter dimension image;Finally, from main split S
Middle K dimensionality reduction layer of selection, other branch { T is separately input to by the output image of K dimensionality reduction layerk, k=1 ..., K, K >=1
In, other branch's classification is arranged to make up multilevel hierarchy, by the successively calculating of Ge Pang branches, by branch T by afterbodyK's
Output image is input in the cutting layer W in main split S, and the input picture is cut into and main point by the cutting layer W in main split S
The image identical size of input layer I inputs in branch S, while the image after cutting is used as whole multiple-limb network integration model
Output, obtain classify shot chart picture.
Above-mentioned described excitation layer l in the present inventionjAbove all immediately convolutional layer ni, such as convolutional layer n in Fig. 21With swash
Encourage a layer l1, convolutional layer n2With excitation layer l2, convolutional layer n3With excitation layer l3Deng.In addition, dimensionality reduction layer is to intersect with convolutional layer, excitation layer
Arrangement, such as in Fig. 2, all dimensionality reduction layer m1、m2、m3、m4、m5It is interspersed in convolutional layer n1-n15And excitation layer l1-l15In.
Preferably, the convolutional layer niOutput image to adjacent last layer performs convolution operation, if convolutional layer niIt is adjacent
Last layer is excitation layer lj, then to excitation layer ljThe image of output performs convolution operation;If convolutional layer niAdjacent last layer is volume
Lamination ni-1, then to convolutional layer ni-1The image of output performs convolution operation;If convolutional layer niAdjacent last layer is dimensionality reduction layer mp,
Then to dimensionality reduction layer mpThe image of output performs convolution operation.
In the present invention, K dimensionality reduction layer is selected from main split S, the output image of K dimensionality reduction layer is separately input to
Other branch { Tk, k=1 ..., K, K >=1 in, i.e. a dimensionality reduction layer m in main split SpOutput image be input to a Ge Pang branches
TkMiddle convolutional layer akIn, convolutional layer akConvolution operation is performed to the input picture of this layer;
Then, according to the S361 Zhong Pang branches { Tk, k=1 ..., K, K >=1 } convolutional layer { ak, k=1 ..., K }
The size of input picture, by ascending order by other branch { Tk, k=1 ..., K } and multilevel hierarchy is arranged as, simultaneously
Per Tiao Pang branches TkOutput image by side branch TkIn up-sampling layer dkOutput, and side branch TkOutput image
It is used as the input picture of branch by adjacent next stage;
Followed by per Tiao Pang branches TkMiddle cutting layer bkTwo inputs are received, if side branch TkIt is branch by the first order,
Then side branch TkMiddle superimposed layer bkFirst input be current other branch TkMiddle convolutional layer akOutput image, second is defeated
It is up-sampling layer u in main split S to enterUOutput;If side branch TkBranch by the first order, then side branch TkIt is middle to cut
Layer bkFirst input be current other branch TkMiddle convolutional layer akOutput image, second input is branch T by upper levelk-1
Up-sampling layer dk-1Output image;Side branch TkMiddle cutting layer bkFirst input is cut into and second input size
Identical image, and the image after cutting is exported;
Followed by per Tiao Pang branches TkMiddle superimposed layer ckTwo inputs are received, if side branch TkIt is branch by the first order,
Then side branch TkMiddle superimposed layer ckFirst input be current other branch TkMiddle cutting layer bkOutput image, second is defeated
It is up-sampling layer u in main split S to enterUOutput;If side branch TkBranch by the first order, then side branch TkMiddle superposition
Layer ckFirst input be current other branch TkMiddle cutting layer bkOutput image, second input is branch T by upper levelk-1
Up-sampling layer dk-1Output image;Side branch TkMiddle superimposed layer ckTwo input pictures are added up, and will it is cumulative after figure
As output;
Followed by per Tiao Pang branches TkMiddle up-sampling layer dkTo the superimposed layer c in current other branchkOutput image perform
Deconvolution is operated, and the image output after deconvolution is operated;
Finally, by branch T by afterbodykIn up-sampling layer dkThe image of output is input to the cutting layer in main split S
In W, the input picture is cut into the image identical with the input layer I inputs in main split S by the cutting layer W in main split S
Size, while the image after cutting obtains shot chart picture of classifying as the output of whole multiple-limb network integration model.
As shown in Fig. 2 the hierarchy for being main split S in the embodiment in the present invention, embodiment is:Input layer-convolution
The convolutional layer 4- excitation layers 4- of layer 1- excitation layer 1- convolutional layer 2- excitation layer 2- dimensionality reduction layer 1- convolutional layer 3- excitation layers 3
Dimensionality reduction layer 2- convolutional layer 5- excitation layer 5- convolutional layer 6- excitation layer 6- convolutional layer 7- excitation layer 7- dimensionality reduction layer 3- convolution
Layer 8- excitation layer 8- convolutional layer 9- excitation layer 9- convolutional layer 10- excitation layer 10- dimensionality reduction layer 4- convolutional layer 11- excitation layers
11- convolutional layer 12- excitation layer 12- convolutional layer 13- excitation layer 13- dimensionality reductions layer 5- convolutional layer 14- excitation layer 14- convolution
Layer 15- excitation layer 15- convolutional layers 16, are finally to cut layer 3 followed by up-sampling layer 1- up-sampling layers 2.Wherein, in Fig. 2
First layer is main split S input layer, and image is inputted by the input layer, is then passed through convolutional layer, and convolutional layer starts to input
The image of layer carries out convolution, and then performing nonlinear function by excitation layer is calculated, and a volume is connected to before each excitation layer
Lamination, dimensionality reduction layer 1-5 is interspersed in the middle of convolutional layer 1-15 and excitation layer 1-15, and convolutional layer 16 is using C convolution kernel to excitation layer
15 input picture performs convolution operation, and the input picture of up-sampling 1 pair of convolutional layer 16 of layer performs deconvolution operation, up-samples layer
The output image of 2 pairs of up-sampling layers 1 continues to rise dimension, finally cuts the output category shot chart picture of layer 3.
In the embodiment of the present invention as shown in Figure 2, other branch receives image from the dimensionality reduction layer of main split, if dimensionality reduction layer
Output image size is minimum, then the side branch is that the input of convolutional layer 17 is drop in branch by the first order in the first order, Fig. 2
The output image of layer 4 is tieed up, the output image for cutting the up-sampling layer 2 that layer 1 and superimposed layer 1 receive main split is inputted as one,
Finally by up-sampling 3 output image of layer of the side branch;The input of convolutional layer 18 is dimensionality reduction layer 3 in the other branch of the second level
Output image, cuts layer 2 and superimposed layer 2 and receives the output of up-sampling layer 3 by the first order in branch as input, finally by
Up-sampling 4 output image of layer of the side branch;For the 3rd, fourth estate, other other branch is same reason, until most
Branch process is finished by rear stage, by the up-sampling layer output image of branch by afterbody, while the output image is entered
In the cutting layer 3 of main split, the classification shot chart picture that layer 3 exports the application is finally cut.
In this manual, identical embodiment or example are necessarily directed to the schematic representation of above-mentioned term.
Moreover, specific features, structure, material or the feature of description can be in any one or more embodiments or example with suitable
Mode is combined.In addition, in the case of not conflicting, those skilled in the art can be by the difference described in this specification
The feature of embodiment or example and non-be the same as Example or example is combined and combined.
The foregoing is only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent substitution and improvements made etc. should be included in the scope of the protection.
Claims (7)
1. a kind of remote sensing image sorting technique based on multiple-limb network integration model, it is characterised in that this method includes
Following steps:
S1, builds multiple-limb network integration model, includes a main split S, at least one other branch { Tk, k=1 ..., K, K >=
1 }, wherein K represents the bar number of other branch;
S2, collection remote sensing image carries out classification mark as training image, and to training image, utilizes back-propagation algorithm
Go out multiple-limb network integration model parameter from the training image learning marked;
S3, image is input in the multiple-limb network integration model for having learnt parameter by the input layer I in main split S, figure
As the successively processing by main split S and all other branch { TkSuccessively processing, finally by main split S cutting layer W
Output category shot chart picture.
2. the method for remote sensing image classification according to claim 1, it is characterised in that the step S2 is to set
The number of species of remote sensing image, classification number is C, gathers remote sensing image as training image, and according to the optics
Remote sensing images species carries out classification mark to training image, is gone out using back-propagation algorithm from the training image learning marked
Multiple-limb network integration model parameter.
3. the method for remote sensing image according to claim 1 or 2 classification, it is characterised in that described main split S according to
It is secondary to include an input layer I, at least one convolutional layer { ni, i=1 ..., N, N >=1 }, at least one excitation layer { lj, j=
1 ..., L, L >=1 }, at least one dimensionality reduction layer { mp, p=1 ..., M, M >=1 }, at least one up-sampling layer { uq, q=1 ...,
U, U >=1 }, a cutting layer W, convolutional layer and dimensionality reduction layer cross arrangement as N > 1 or M > 1, wherein N represents the layer of convolutional layer
Number, M represents the number of plies of dimensionality reduction layer, and L represents the number of plies of excitation layer, and U represents to up-sample the number of plies of layer.
4. the method for remote sensing image classification according to claim 3, it is characterised in that described to swash in main split S
Encourage a layer ljIn, each excitation layer ljAbove one layer is all a convolutional layer;Last convolutional layer n in main split SNConvolution
Core number is the classification number C.
5. the method for remote sensing image classification according to claim 1, it is characterised in that described every Tiao Pang branches Tk
Successively comprising a convolutional layer ak, one cut layer bk, a superimposed layer ck, a up-sampling layer dk;Other branch { Tk, k=
1 ..., K, K >=1 } all convolutional layer { a ink, k=1 ..., K, K >=1 } convolution kernel number be all the classification number C.
6. the method for remote sensing image classification according to claim 1, it is characterised in that the step S3 includes:
Input layer I in S31, the main split S receives image input;
Convolutional layer n in S32, main split SiOutput image to adjacent last layer performs convolution operation, extracts characteristics of image, and will
Characteristic image is exported;
Excitation layer l in S33, main split SjThe output characteristic image of adjacent last layer convolutional layer is performed using linear unit is corrected
Nonlinear function is calculated, and the image for calculating generation is exported;
Dimensionality reduction layer m in S34, main split SpTo adjacent last layer excitation layer ljOutput image perform dimensionality reduction, it is and dimensionality reduction image is defeated
Go out;
Layer u is up-sampled in S35, main split SqOutput image to adjacent last layer performs deconvolution operation, realizes the liter to image
Dimension, and export a liter dimension image;
S36, K dimensionality reduction layer is selected from main split S, the output image of K dimensionality reduction layer is separately input into other branch { Tk,
K=1 ..., K, K >=1 in, other branch's classification is arranged to make up multilevel hierarchy, will most by the successively calculating of Ge Pang branches
Branch T by rear stageKOutput image be input in the cutting layer W in main split S, cutting in main split S layer W inputs this
Image cropping is into the image identical size with the input layer I inputs in main split S, while the image after cutting is as whole more
The output of branching networks Fusion Model, obtains shot chart picture of classifying.
7. the method for remote sensing image classification according to claim 6, it is characterised in that the step S36 includes:
S361, K dimensionality reduction layer is selected from main split S, the output image of K dimensionality reduction layer is separately input into other branch
{Tk, k=1 ..., K, K >=1 in, i.e. a dimensionality reduction layer m in main split SpOutput image be input to a Ge Pang branches TkMiddle volume
Lamination akIn, convolutional layer akConvolution operation is performed to the input picture of this layer;
S362, according to the S361 Zhong Pang branches { Tk, k=1 ..., K, K >=1 } convolutional layer { ak, k=1 ..., K } input
The size of image, by ascending order by other branch { Tk, k=1 ..., K } and multilevel hierarchy is arranged as, while every
Other branch TkOutput image by side branch TkIn up-sampling layer dkOutput, and side branch TkOutput image conduct
The input picture of branch by adjacent next stage;
S363, per Tiao Pang branches TkMiddle cutting layer bkTwo inputs are received, if side branch TkIt is branch by the first order, then by this
Branch TkMiddle superimposed layer bkFirst input be current other branch TkMiddle convolutional layer akOutput image, second input is main
Layer u is up-sampled in branch SUOutput;If side branch TkBranch by the first order, then side branch TkMiddle cutting layer bk's
First input is current other branch TkMiddle convolutional layer akOutput image, second input is branch T by upper levelk-1On adopt
Sample layer dk-1Output image;Side branch TkMiddle cutting layer bkFirst input is cut into and second input size identical
Image, and the image after cutting is exported;
S364, per Tiao Pang branches TkMiddle superimposed layer ckTwo inputs are received, if side branch TkIt is branch by the first order, then by this
Branch TkMiddle superimposed layer ckFirst input be current other branch TkMiddle cutting layer bkOutput image, second input is main
Layer u is up-sampled in branch SUOutput;If side branch TkBranch by the first order, then side branch TkMiddle superimposed layer ck's
First input is current other branch TkMiddle cutting layer bkOutput image, second input is branch T by upper levelk-1On adopt
Sample layer dk-1Output image;Side branch TkMiddle superimposed layer ckTwo input pictures are added up, and will it is cumulative after image output;
S365, per Tiao Pang branches TkMiddle up-sampling layer dkTo the superimposed layer c in current other branchkOutput image perform deconvolution
Operation, and the image output after deconvolution is operated;
S366, by branch T by afterbodykIn up-sampling layer dkThe image of output is input in the cutting layer W in main split S,
The input picture is cut into the image identical size inputted with the input layer I in main split S by the cutting layer W in main split S,
Image cut simultaneously after obtains shot chart picture of classifying as the output of whole multiple-limb network integration model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710213577.4A CN107122796B (en) | 2017-04-01 | 2017-04-01 | A kind of remote sensing image classification method based on multiple-limb network integration model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710213577.4A CN107122796B (en) | 2017-04-01 | 2017-04-01 | A kind of remote sensing image classification method based on multiple-limb network integration model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107122796A true CN107122796A (en) | 2017-09-01 |
CN107122796B CN107122796B (en) | 2019-07-12 |
Family
ID=59724599
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710213577.4A Active CN107122796B (en) | 2017-04-01 | 2017-04-01 | A kind of remote sensing image classification method based on multiple-limb network integration model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107122796B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107993217A (en) * | 2017-12-22 | 2018-05-04 | 北京奇虎科技有限公司 | Video data real-time processing method and device, computing device |
CN108664994A (en) * | 2018-04-17 | 2018-10-16 | 哈尔滨工业大学深圳研究生院 | A kind of remote sensing image processing model construction system and method |
CN108875751A (en) * | 2017-11-02 | 2018-11-23 | 北京旷视科技有限公司 | Image processing method and device, the training method of neural network, storage medium |
CN109146885A (en) * | 2018-08-17 | 2019-01-04 | 深圳蓝胖子机器人有限公司 | Image partition method, equipment and computer readable storage medium |
CN109523569A (en) * | 2018-10-18 | 2019-03-26 | 中国科学院空间应用工程与技术中心 | A kind of remote sensing image dividing method and device based on more granularity network integrations |
CN109544554A (en) * | 2018-10-18 | 2019-03-29 | 中国科学院空间应用工程与技术中心 | A kind of segmentation of plant image and blade framework extracting method and system |
CN109558806A (en) * | 2018-11-07 | 2019-04-02 | 北京科技大学 | The detection method and system of high score Remote Sensing Imagery Change |
CN109902584A (en) * | 2019-01-28 | 2019-06-18 | 深圳大学 | A kind of recognition methods, device, equipment and the storage medium of mask defect |
CN110298396A (en) * | 2019-06-25 | 2019-10-01 | 北京工业大学 | Hyperspectral image classification method based on deep learning multiple features fusion |
CN110472668A (en) * | 2019-07-22 | 2019-11-19 | 华北电力大学(保定) | A kind of image classification method |
CN110659726A (en) * | 2019-09-24 | 2020-01-07 | 北京达佳互联信息技术有限公司 | Image processing method and device, electronic equipment and storage medium |
CN111833273A (en) * | 2020-07-17 | 2020-10-27 | 华东师范大学 | Semantic boundary enhancement method based on long-distance dependence |
CN112560706A (en) * | 2020-12-18 | 2021-03-26 | 广东电网有限责任公司电力科学研究院 | Method and device for identifying water body target of multi-source satellite image |
CN113554655A (en) * | 2021-07-13 | 2021-10-26 | 中国科学院空间应用工程与技术中心 | Optical remote sensing image segmentation method and device based on multi-feature enhancement |
CN114419059A (en) * | 2021-11-24 | 2022-04-29 | 哈尔滨工程大学 | Visual sense-based germ-remaining rice germ-remaining degree detection method |
WO2023221266A1 (en) * | 2022-05-16 | 2023-11-23 | 清华大学 | Multi-branch network collaborative reasoning method and system for internet of things |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104077613A (en) * | 2014-07-16 | 2014-10-01 | 电子科技大学 | Crowd density estimation method based on cascaded multilevel convolution neural network |
CN105701508A (en) * | 2016-01-12 | 2016-06-22 | 西安交通大学 | Global-local optimization model based on multistage convolution neural network and significant detection algorithm |
CN105912990A (en) * | 2016-04-05 | 2016-08-31 | 深圳先进技术研究院 | Face detection method and face detection device |
-
2017
- 2017-04-01 CN CN201710213577.4A patent/CN107122796B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104077613A (en) * | 2014-07-16 | 2014-10-01 | 电子科技大学 | Crowd density estimation method based on cascaded multilevel convolution neural network |
CN105701508A (en) * | 2016-01-12 | 2016-06-22 | 西安交通大学 | Global-local optimization model based on multistage convolution neural network and significant detection algorithm |
CN105912990A (en) * | 2016-04-05 | 2016-08-31 | 深圳先进技术研究院 | Face detection method and face detection device |
Non-Patent Citations (3)
Title |
---|
ZHICHENG CUI 等: "Multi-Scale Convolutional Neural Networks for Time Series Classification", 《ARXIV:1603.06995V4 [CS.CV]》 * |
林培群 等: "基于多分支BP网络模型的车型分类研究", 《可持续发展的中国交通——2005全国博士生学术论坛(交通运输工程学科)论文集(下册)》 * |
蔡晓东 等: "基于多分支卷积神经网络的车辆图像比对方法", 《电视技术》 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108875751A (en) * | 2017-11-02 | 2018-11-23 | 北京旷视科技有限公司 | Image processing method and device, the training method of neural network, storage medium |
CN108875751B (en) * | 2017-11-02 | 2021-03-02 | 北京旷视科技有限公司 | Image processing method and device, neural network training method and storage medium |
CN107993217A (en) * | 2017-12-22 | 2018-05-04 | 北京奇虎科技有限公司 | Video data real-time processing method and device, computing device |
CN107993217B (en) * | 2017-12-22 | 2021-04-09 | 北京奇虎科技有限公司 | Video data real-time processing method and device and computing equipment |
CN108664994A (en) * | 2018-04-17 | 2018-10-16 | 哈尔滨工业大学深圳研究生院 | A kind of remote sensing image processing model construction system and method |
CN109146885A (en) * | 2018-08-17 | 2019-01-04 | 深圳蓝胖子机器人有限公司 | Image partition method, equipment and computer readable storage medium |
CN109146885B (en) * | 2018-08-17 | 2021-08-17 | 深圳蓝胖子机器智能有限公司 | Image segmentation method, apparatus, and computer-readable storage medium |
CN109523569B (en) * | 2018-10-18 | 2020-01-31 | 中国科学院空间应用工程与技术中心 | optical remote sensing image segmentation method and device based on multi-granularity network fusion |
CN109523569A (en) * | 2018-10-18 | 2019-03-26 | 中国科学院空间应用工程与技术中心 | A kind of remote sensing image dividing method and device based on more granularity network integrations |
CN109544554B (en) * | 2018-10-18 | 2020-01-31 | 中国科学院空间应用工程与技术中心 | plant image segmentation and leaf skeleton extraction method and system |
CN109544554A (en) * | 2018-10-18 | 2019-03-29 | 中国科学院空间应用工程与技术中心 | A kind of segmentation of plant image and blade framework extracting method and system |
CN109558806A (en) * | 2018-11-07 | 2019-04-02 | 北京科技大学 | The detection method and system of high score Remote Sensing Imagery Change |
CN109558806B (en) * | 2018-11-07 | 2021-09-14 | 北京科技大学 | Method for detecting high-resolution remote sensing image change |
CN109902584A (en) * | 2019-01-28 | 2019-06-18 | 深圳大学 | A kind of recognition methods, device, equipment and the storage medium of mask defect |
CN109902584B (en) * | 2019-01-28 | 2022-02-22 | 深圳大学 | Mask defect identification method, device, equipment and storage medium |
CN110298396A (en) * | 2019-06-25 | 2019-10-01 | 北京工业大学 | Hyperspectral image classification method based on deep learning multiple features fusion |
CN110472668A (en) * | 2019-07-22 | 2019-11-19 | 华北电力大学(保定) | A kind of image classification method |
CN110659726A (en) * | 2019-09-24 | 2020-01-07 | 北京达佳互联信息技术有限公司 | Image processing method and device, electronic equipment and storage medium |
CN111833273A (en) * | 2020-07-17 | 2020-10-27 | 华东师范大学 | Semantic boundary enhancement method based on long-distance dependence |
CN111833273B (en) * | 2020-07-17 | 2021-08-13 | 华东师范大学 | Semantic boundary enhancement method based on long-distance dependence |
CN112560706A (en) * | 2020-12-18 | 2021-03-26 | 广东电网有限责任公司电力科学研究院 | Method and device for identifying water body target of multi-source satellite image |
CN113554655A (en) * | 2021-07-13 | 2021-10-26 | 中国科学院空间应用工程与技术中心 | Optical remote sensing image segmentation method and device based on multi-feature enhancement |
CN113554655B (en) * | 2021-07-13 | 2021-12-31 | 中国科学院空间应用工程与技术中心 | Optical remote sensing image segmentation method and device based on multi-feature enhancement |
CN114419059A (en) * | 2021-11-24 | 2022-04-29 | 哈尔滨工程大学 | Visual sense-based germ-remaining rice germ-remaining degree detection method |
WO2023221266A1 (en) * | 2022-05-16 | 2023-11-23 | 清华大学 | Multi-branch network collaborative reasoning method and system for internet of things |
Also Published As
Publication number | Publication date |
---|---|
CN107122796B (en) | 2019-07-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107122796B (en) | A kind of remote sensing image classification method based on multiple-limb network integration model | |
CN111079795B (en) | Image classification method based on CNN (content-centric networking) fragment multi-scale feature fusion | |
CN106845529B (en) | Image feature identification method based on multi-view convolution neural network | |
CN106372648A (en) | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method | |
CN108288035A (en) | The human motion recognition method of multichannel image Fusion Features based on deep learning | |
JP2022548712A (en) | Image Haze Removal Method by Adversarial Generation Network Fusing Feature Pyramids | |
CN107220657A (en) | A kind of method of high-resolution remote sensing image scene classification towards small data set | |
CN107527318A (en) | A kind of hair style replacing options based on generation confrontation type network model | |
CN109543502A (en) | A kind of semantic segmentation method based on the multiple dimensioned neural network of depth | |
CN108549893A (en) | A kind of end-to-end recognition methods of the scene text of arbitrary shape | |
CN107437092A (en) | The sorting algorithm of retina OCT image based on Three dimensional convolution neutral net | |
CN106650690A (en) | Night vision image scene identification method based on deep convolution-deconvolution neural network | |
CN108986058A (en) | The image interfusion method of lightness Consistency Learning | |
CN105631415A (en) | Video pedestrian recognition method based on convolution neural network | |
CN108416314B (en) | Picture important face detection method | |
CN110046550A (en) | Pedestrian's Attribute Recognition system and method based on multilayer feature study | |
CN113158862A (en) | Lightweight real-time face detection method based on multiple tasks | |
CN112381764A (en) | Crop disease and insect pest detection method | |
CN113034506B (en) | Remote sensing image semantic segmentation method and device, computer equipment and storage medium | |
CN107958067A (en) | It is a kind of based on without mark Automatic Feature Extraction extensive electric business picture retrieval system | |
CN111161244B (en) | Industrial product surface defect detection method based on FCN + FC-WXGboost | |
CN113762265B (en) | Classified segmentation method and system for pneumonia | |
CN108416397A (en) | A kind of Image emotional semantic classification method based on ResNet-GCN networks | |
CN109636764A (en) | A kind of image style transfer method based on deep learning and conspicuousness detection | |
CN107958219A (en) | Image scene classification method based on multi-model and Analysis On Multi-scale Features |
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 |