CN106447687A - Boundary extraction method of remote sensing image pixel through neighbor Filter - Google Patents
Boundary extraction method of remote sensing image pixel through neighbor Filter Download PDFInfo
- Publication number
- CN106447687A CN106447687A CN201610898413.5A CN201610898413A CN106447687A CN 106447687 A CN106447687 A CN 106447687A CN 201610898413 A CN201610898413 A CN 201610898413A CN 106447687 A CN106447687 A CN 106447687A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- neighborhood
- sensing image
- sample
- pixel
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a boundary extraction method of a remote sensing image pixel through a neighbor Filter, relates to a remote sensing image boundary extraction method, and brings forward a boundary extraction method of a remote sensing image pixel through a neighbor Filter, for solving the problems of inaccuracy of threshold selection and heavy workload caused by extraction of superfluous boundary. The method is realized through the following steps: step one, constructing a sample position Set; step two, constructing the neighbor Filter; step three, obtaining a training sample Set by filtering position samples of the sample position Set through the neighbor Filter; step four, obtaining a neural network prediction model; step five, outputting a boundary extraction result, and the like. The method is applied to the field of remote sensing image boundary extraction.
Description
Technical field
The present invention relates to boundary extraction method, the more particularly to side of a kind of remote sensing image picture element and its one-dimensional filter of neighborhood
Boundary's extracting method.
Background technology
Spectrum remote sensing image have recorded the spectral information of atural object in earth's surface certain area coverage, permissible by boundary extraction algorithm
In remote sensing image border atural object between is automatically obtained;These borders for analysis remote sensing image in characters of ground object, obtain
The distribution situation of building and vegetation, construction vector quantization numerical map are particularly significant, therefore the Boundary Extraction of remote sensing image is wide
General is applied to remote Sensing Interpretation, land use pattern change, agricultural and environmental monitoring, with higher practical value.
Sense image boundary extracts the main method for adopting at present is:By an algorithm filter to certain convolution window model
The pixel of the remote sensing image in enclosing is calculated and is obtained a result, if this result is side higher than certain threshold marker
Boundary, on the contrary it is labeled as non-border.The problem that current method is primarily present is:First, the more difficult selection of threshold value, threshold value value is larger
Excessive border pixel can be so caused to be ignored, less can the causing a large amount of on remote sensing image of threshold value value itself is not border
Pixel be marked as border.Second, the atural object for including in a width remote sensing image is numerous, user be not usually want to extract all
Border, but according to application need extract the boundary information between specific one group of atural object, and existing algorithm is of overall importance
All possible border in whole image is all extracted, a lot of boundary information users not need, user need
The result of extraction and former remote sensing image are carried out manually position one by one compare into line deletion, workload is larger.
It is thus desirable to a kind of method, by the one of user input group of border and non-boundary sample, need not be input into threshold value
In the case of identifying user need the border of which feature on earth and set up corresponding model, extract the ground that user really needs
Thing boundary information.
Content of the invention
Border inaccurate and that extraction is unnecessary is selected to cause workload big the invention aims to solving threshold value
Problem, and the boundary extraction method of a kind of remote sensing image picture element of proposition and its one-dimensional filter of neighborhood.
Above-mentioned goal of the invention is achieved through the following technical solutions:
Step one, input remote sensing image InputMap, choose n on the remote sensing image and are in borderline pixel and n
Individual be not located at borderline pixel, borderline pixel is according to n and n be not located at borderline pixel and construct sample bit
Put collection positionSet;
Step 2, the one-dimensional filter neighborFilter of construction neighborhood;
Step 3, filter the position of sample position collection positionSet by the one-dimensional filter neighborFilter of neighborhood
Put sample and obtain training sample set simpleSet;
Step 4, each training sample sample for utilizing in nerve net Algorithm Learning training sample set simpleSet,
Obtain nerve net forecast model model;
Step 5, by the one-dimensional filter neighborFilter of neighborhood and nerve net forecast model model to be input into distant
All pixels of sense image InputMap are processed, and output boundary extracts result.
Invention effect
For the problem that prior art is present, the present invention provides a kind of remote sensing shadow based on the one-dimensional filter of sample and neighborhood
As boundary extraction method.One group of border of user input and non-boundary sample can be passed through by the method, need not be input into
In the case of threshold value, identifying user needs the border of which feature on earth and sets up corresponding model, and extracting user really needs
Atural object boundary information (as Fig. 7~9).
The present invention provides a kind of remote sensing image boundary extraction method based on the one-dimensional filter of sample and neighborhood.By the party
Method can be by one group of border of user input and non-boundary sample, and in the case of it need not be input into threshold value, identifying user is on earth
Need the border of which feature and corresponding model is set up, extract the atural object boundary information that user really needs.By we
Method can obtain the atural object boundary information of better quality, be widely used in remote Sensing Interpretation, land use pattern change, agricultural with
Environmental monitoring, with higher using value.
Description of the drawings
Fig. 1 is the Boundary Extraction of a kind of remote sensing image picture element of the proposition of specific embodiment one and its one-dimensional filter of neighborhood
Method flow diagram;
Fig. 2 is the construction sample position collection positionSet procedure chart of the proposition of specific embodiment one;
Fig. 3 is the one-dimensional filter neighborFilter procedure chart of construction neighborhood of the proposition of specific embodiment four;
Fig. 4 is the acquisition training sample set simpleSet detailed process figure of the proposition of specific embodiment eight;
Fig. 5 is the acquisition nerve net forecast model model flow chart of the proposition of specific embodiment nine;
Fig. 6 is that the output boundary of the proposition of specific embodiment ten extracts outcome procedure figure;
Fig. 7 is the loading remote sensing image schematic diagram of the proposition of specific embodiment one;
Fig. 8 is that choosing on image for the proposition of specific embodiment one is in the pixel on border and is not located at the pixel on border
Schematic diagram;
Fig. 9 is the result schematic diagram for obtaining after the program of the proposition of specific embodiment one is run.
Specific embodiment
Specific embodiment one:A kind of remote sensing image picture element in conjunction with Fig. 1 present embodiment and its one-dimensional filter of neighborhood
Boundary extraction method, specifically according to following steps prepare:
Step one, input remote sensing image InputMap, choose n on the remote sensing image and are in borderline pixel and n
Individual be not located at borderline pixel, borderline pixel is according to n and n be not located at borderline pixel and construct sample bit
Put collection positionSet;
Step 2, the one-dimensional filter neighborFilter of construction neighborhood;
Step 3, filter the position of sample position collection positionSet by the one-dimensional filter neighborFilter of neighborhood
Put sample and obtain training sample set simpleSet;
Step 4, each training sample sample for utilizing in nerve net Algorithm Learning training sample set simpleSet,
Obtain nerve net forecast model model;(there are a lot of sample inside simpleSet, numerous sample are through nerve net algorithm
Practise and could obtain a forecast model)
Step 5, by the one-dimensional filter neighborFilter of neighborhood and nerve net forecast model model to be input into distant
All pixels of sense image InputMap are processed, and output boundary extracts result.
Present embodiment effect:
For the problem that prior art is present, present embodiment provides a kind of based on the distant of the one-dimensional filter of sample and neighborhood
Sense image boundary extracting method.One group of border of user input and non-boundary sample can be passed through by the method, do not needed
In the case of input threshold value, identifying user needs the border of which feature on earth and sets up corresponding model, extracts user real
The atural object boundary information (as Fig. 7~9) of needs.
Present embodiment provides a kind of remote sensing image boundary extraction method based on the one-dimensional filter of sample and neighborhood.Pass through
The method can pass through one group of border of user input and non-boundary sample, identifying user in the case of it need not be input into threshold value
Need the border of which feature on earth and corresponding model is set up, extract the atural object boundary information that user really needs.Pass through
This method can obtain the atural object boundary information of better quality, be widely used in remote Sensing Interpretation, land use pattern change, agriculture
Industry and environmental monitoring, with higher using value.
Specific embodiment two:Present embodiment from unlike specific embodiment one:Remote sensing shadow is input in step one
As InputMap, choose on the remote sensing image n be in borderline pixel and n be not located at borderline pixel, according to n
Individual be in borderline pixel and n and be not located at borderline pixel construction sample position collection positionSet detailed process be
As shown in Figure 2:
Step one by one, input remote sensing image InputMap;
Step one two, on remote sensing image InputMap artificial choose the n pixel being on boundary position and n do not locate
Pixel on boundary position;Each pixel is configured to position sample positionSimple={ x, y, L };Wherein, position
Sample is a tlv triple;X is the residing columns on remote sensing image of pixel, y is the residing row on remote sensing image of the pixel
Number;L is boundary label, the boundary label L=1 when pixel is in border, the boundary label L=when pixel is not located at boundary position
0;
Step one three, construction sample position collection positionSet, corresponding for all pixels position sample is added to sample
In the collection positionSet of position.Other steps and parameter are identical with specific embodiment one.
Specific embodiment three:Present embodiment from unlike specific embodiment one or two:In step one, n is arrived for 20
Between 500,100 are defaulted as.Other steps and parameter are identical with specific embodiment one or two.
Specific embodiment four:Unlike one of present embodiment and specific embodiment one to three:Structure in step 2
It is as shown in Figure 3 to make the one-dimensional filter neighborFilter detailed process of neighborhood:
Step 2 one, the one-dimensional filter neighborFilter of neighborhood is set up, remote sensing is input in the one-dimensional filter of neighborhood
The upper pixel for being set to (x, y) of image;
Step 2 two, on remote sensing image centered on position (x, y) point, on the remote sensing image in residing line number in
The both sides of heart point respectively take M pixel and are stored among temporary array L1 of left and right neighborhood, on remote sensing image on residing columns, center
Respectively take M pixel in Y direction above and below point to be stored among temporary array L2 of upper and lower neighborhood;
Step 2 three, 3 neighboring mean value arrays meanL1 of L1 are set up, the 3 neighborhood elemental standards of L1 are set up according to meanL1
Difference group stdL1;
Wherein, the element number W of meanL1 is identical with the element number of L1, first element of meanL1 and last
The value of element is set as that first element of 0, stdL1 and the value of last element are set as 0;
Step 2 four, 3 neighboring mean value arrays meanL2 of L2 are set up, the 3 neighborhood elemental standards of L2 are set up according to meanL2
Difference group stdL2, wherein, the element number N of meanL2 is identical with the element number of L2, and first element of meanL2 is with finally
The value of one element is set as 0;First element of stdL2 and the value of last element are set as 0;
Step 2 five, stdL1, meanL1, stdL2 and meanL2 are combined as one-dimensional characteristic array oneArray;
Step 2 six, determine that the one-dimensional filter neighborFilter of neighborhood is output as one-dimensional characteristic array
oneArray.One of other steps and parameter and specific embodiment one to three are identical.
Specific embodiment five:Unlike one of present embodiment and specific embodiment one to four:Two institute of step 2
The odd number that M is between 11 to 51 is stated, default value is 11.One of other steps and parameter and specific embodiment one to four are identical.
Specific embodiment six:Unlike one of present embodiment and specific embodiment one to five:In step 2 three
The computing formula of second of meanL1 to penultimate element is as follows:
Wherein, meanL1 (i) is i-th element of meanL1, and L1 (i-1) is the i-th -1 element of L1, and L1 (i) is L1
I-th element, L1 (i+1) for L1 i+1 element;
The computing formula of second of the stdL1 to penultimate element is as follows:
Wherein, stdL1 (i) is i-th element of stdL1, and L1 (i-1) is the i-1 element of L1, and L1 (i) is i of L1
Element, L1 (i+1) is the i+1 element of L1, and meanL1 (i) is i-th element of meanL1.Other steps and parameter with concrete
One of embodiment one to five is identical.
Specific embodiment seven:Unlike one of present embodiment and specific embodiment one to six:In step 2 four
The computing formula of second of meanL2 to penultimate element is as follows:
Wherein, meanL2 (i) is i-th element of meanL2, and L2 (i-1) is the i-1 element of L2, and L2 (i) is the i of L2
Individual element, L2 (i+1) is the i+1 element of L2;
The computing formula of second of the stdL2 to penultimate element is as follows:
Wherein, stdL2 (i) is i-th element of stdL1, and L2 (i-1) is the i-1 element of L2, and L2 (i) is i of L2
Element, L2 (i+1) is the i+1 element of L2, and meanL2 (i) is i-th element of meanL1.Other steps and parameter with concrete
One of embodiment one to six is identical.
Specific embodiment eight:Unlike one of present embodiment and specific embodiment one to seven:Lead in step 3
Cross the position sample acquisition training sample that the one-dimensional filter neighborFilter of neighborhood filters sample position collection positionSet
Collection simpleSet is concretely comprised the following steps as shown in Figure 4:
Step 3 one, set up an empty training sample set simpleSet;
Step 3 two, setting Sample Counter sampleCounter are 1;
Step 3 three, the sampleCounter element for taking out in positionSet are put into temporary variable EP
(temporaryParameter) in the middle of;
X and y in the middle of step 3 four, taking-up EP, x and y are inputed in the one-dimensional filter neighborFilter of neighborhood,
One-dimensional characteristic array oneArray is exported using the one-dimensional filter neighborFilter of neighborhood;
Step 3 five, set up a training sample sample;Wherein, sample be two tuple input,
Output }, the value of sample input data input is oneArray, and the value of sample output data output is the boundary label of EP
L;(variable label of sample, it is an array to have arrived sample stage input, output be whether label mark)
Step 3 six, sample is added among simpleSet, enumerator sampleCounter increases by 1;
If step Radix Notoginseng increases the sampleCounter after 1 so goes to step 3 eight more than 2 × n, will otherwise increase
SampleCounter after plus 1 goes to step 3 three;
Step 3 eight, the process of construction training sample set simpleSet terminate.Other steps and parameter and specific embodiment party
One of formula one to seven is identical.
Specific embodiment nine:Unlike one of present embodiment and specific embodiment one to eight:Profit in step 4
With the training sample sample in nerve net Algorithm Learning training sample set simpleSet, nerve net forecast model model is obtained
Specially as Fig. 5:
Step 4 one, first element for taking out in simpleSet, obtain the length of input data input of the element
DL;
Step 4 two, a nerve net is set up, wherein, nerve net input layer includes DL neuron, nerve net intermediate layer
Comprising 5 neurons, nerve net output layer corresponds to 1 neuron;
Step 4 three, all elements for passing through in nerve net Algorithm Learning simpleSet, by each unit in simpleSet
Element sample input data input as input, using output data output of each element in simpleSet as output,
Nerve net forecast model model is set up according to input and output.One of other steps and parameter and specific embodiment one to eight
Identical.
Specific embodiment ten:Unlike one of present embodiment and specific embodiment one to nine:Lead in step 5
The one-dimensional filter neighborFilter of neighborhood and nerve net forecast model model are crossed to being input into the institute of remote sensing image InputMap
Pixel is had to be processed, it is as shown in Figure 6 that output boundary extracts result detailed process:
Step May Day, the width width and height height of acquisition input remote sensing image
Step 5 two, set up a black width width and height height output result figure resultMap
Step 5 three, pixel counter pixelCounter are set to 1;
Step the May 4th, take out columns x, distant of the pth ixelCounter pixel of InputMap residing on remote sensing image
Residing line number y on image of sense image;
Step 5 five, x and y are inputed to the one-dimensional Convolution Filter neighborFilter of neighborhood, obtain output
oneArray
Step 5 six, oneArray is inputed to nerve net forecast model model, obtain prediction output result r;
If seven r of step 5 so represents that more than 0.5 currently processed pixel is border, then in result figure
White on the x of resultMap and y location labelling;
Step 5 eight, pixel counter pixelCounter increase by 1;
If step 5 nine increases the pixelCounter after 1, the pixel number more than InputMap so goes to step
On May Day ten, otherwise go to step the May 4th;
Step May Day ten, using resultMap as Boundary Extraction result output, by resultMap in figure white portion
It is allocated as the output boundary for being extracted and extracts result.One of other steps and parameter and specific embodiment one to nine are identical.
Claims (10)
1. the boundary extraction method of a kind of remote sensing image picture element and its one-dimensional filter of neighborhood, it is characterised in that the method is concrete
Follow the steps below:
Step one, input remote sensing image InputMap, on the remote sensing image, selection n is in borderline pixel and n is individual not
Be in borderline pixel, borderline pixel is according to n and n be not located at borderline pixel and construct sample position collection
positionSet;
Step 2, the one-dimensional filter neighborFilter of construction neighborhood;
Step 3, filter the position sample of sample position collection positionSet by the one-dimensional filter neighborFilter of neighborhood
This acquisition training sample set simpleSet;
Step 4, each training sample sample for utilizing in nerve net Algorithm Learning training sample set simpleSet, obtain
Nerve net forecast model model;
Step 5, by the one-dimensional filter neighborFilter of neighborhood and nerve net forecast model model to be input into remote sensing shadow
As all pixels of InputMap are processed, output boundary extracts result.
2. the boundary extraction method of a kind of remote sensing image picture element and its one-dimensional filter of neighborhood according to claim 1, which is special
Levy and be:Remote sensing image InputMap being input in step one, n is chosen on the remote sensing image and is in borderline pixel and n
Individual be not located at borderline pixel, borderline pixel is according to n and n be not located at borderline pixel and construct sample bit
Putting collection positionSet detailed process is:
Step one by one, input remote sensing image InputMap;
Step one two, on remote sensing image InputMap artificial choose the n pixel being on boundary position and n be not located at side
Pixel on boundary position;Each pixel is configured to position sample positionSimple={ x, y, L };Wherein, x exists for pixel
On remote sensing image, residing columns, y are the residing line number on remote sensing image of the pixel;L is boundary label, when pixel is in side
Boundary label L=1 during boundary, the boundary label L=0 when pixel is not located at boundary position;
Step one three, construction sample position collection positionSet, corresponding for all pixels position sample is added to sample position
In collection positionSet.
3. the boundary extraction method of a kind of remote sensing image picture element and its one-dimensional filter of neighborhood according to claim 2, which is special
Levy and be:In step one n be 20 to 500 between.
4. according to claim 1,2 or 3 a kind of remote sensing image picture element and its one-dimensional filter of neighborhood boundary extraction method,
It is characterized in that:In step 2, the one-dimensional filter neighborFilter detailed process of construction neighborhood is:
Step 2 one, the one-dimensional filter neighborFilter of neighborhood is set up, remote sensing image is input in the one-dimensional filter of neighborhood
The upper pixel for being set to (x, y);
Step 2 two, on remote sensing image centered on position (x, y) point, in central point in residing line number on the remote sensing image
Both sides respectively take M pixel and be stored in left and right neighborhood and keep among array L1, on the remote sensing image on residing columns, on central point
Respectively take M pixel in the Y direction of side and lower section to be stored among temporary array L2 of upper and lower neighborhood;
Step 2 three, 3 neighboring mean value arrays meanL1 of L1 are set up, the 3 neighborhood elemental standards differences of L1 are set up according to meanL1
Group stdL1;
Wherein, the element number W of meanL1 is identical with the element number of L1, first element of meanL1 and last element
Value be set as first element of 0, stdL1 and the value of last element is set as 0;
Step 2 four, 3 neighboring mean value arrays meanL2 of L2 are set up, the 3 neighborhood elemental standards differences of L2 are set up according to meanL2
Group stdL2, wherein, the element number N of meanL2 is identical with the element number of L2, first element of meanL2 and last
The value of element is set as 0;First element of stdL2 and the value of last element are set as 0;
Step 2 five, stdL1, meanL1, stdL2 and meanL2 are combined as one-dimensional characteristic array oneArray;
Step 2 six, determine that the one-dimensional filter neighborFilter of neighborhood is output as one-dimensional characteristic array oneArray.
5. the boundary extraction method of a kind of remote sensing image picture element and its one-dimensional filter of neighborhood according to claim 4, which is special
Levy and be:M described in step 2 two is the odd number between 11 to 51.
6. the boundary extraction method of a kind of remote sensing image picture element and its one-dimensional filter of neighborhood according to claim 4, which is special
Levy and be:In step 2 three, the computing formula of second of meanL1 to penultimate element is as follows:
Wherein, meanL1 (i) is i-th element of meanL1, and L1 (i-1) is the i-th -1 element of L1, and L1 (i) is the i-th of L1
Individual element, L1 (i+1) is the i+1 element of L1;
The computing formula of second of the stdL1 to penultimate element is as follows:
Wherein, stdL1 (i) is i-th element of stdL1, and L1 (i-1) is the i-1 element of L1, and L1 (i) is the i unit of L1
Element, L1 (i+1) is the i+1 element of L1, and meanL1 (i) is i-th element of meanL1.
7. the boundary extraction method of a kind of remote sensing image picture element and its one-dimensional filter of neighborhood according to claim 6, which is special
Levy and be:In step 2 four, the computing formula of second of meanL2 to penultimate element is as follows:
Wherein, meanL2 (i) is i-th element of meanL2, and L2 (i-1) is the i-1 element of L2, and L2 (i) is the i unit of L2
Element, L2 (i+1) is the i+1 element of L2;
The computing formula of second of the stdL2 to penultimate element is as follows:
Wherein, stdL2 (i) is i-th element of stdL1, and L2 (i-1) is the i-1 element of L2, and L2 (i) is the i unit of L2
Element, L2 (i+1) is the i+1 element of L2, and meanL2 (i) is i-th element of meanL1.
8. according to claim 1 or 7 a kind of remote sensing image picture element and its one-dimensional filter of neighborhood boundary extraction method, its
It is characterised by:Filter sample position collection positionSet's by the one-dimensional filter neighborFilter of neighborhood in step 3
Position sample obtains training sample set simpleSet and concretely comprises the following steps:
Step 3 one, set up an empty training sample set simpleSet;
Step 3 two, setting Sample Counter sampleCounter are 1;
Step 3 three, the sampleCounter element for taking out in positionSet are put in the middle of temporary variable EP;
X and y in the middle of step 3 four, taking-up EP, x and y is inputed in the one-dimensional filter neighborFilter of neighborhood, is utilized
The one-dimensional filter neighborFilter of neighborhood exports one-dimensional characteristic array oneArray;
Step 3 five, set up a training sample sample;Wherein, sample is two tuple { input, output }, sample
The value of this input data input is oneArray, and the value of sample output data output is the boundary label L of EP;
Step 3 six, sample is added among simpleSet, enumerator sampleCounter increases by 1;
If step Radix Notoginseng increases the sampleCounter after 1 so goes to step 3 eight more than 2 × n, after otherwise increasing by 1
SampleCounter go to step 3 three;
Step 3 eight, the process of construction training sample set simpleSet terminate.
9. the boundary extraction method of a kind of remote sensing image picture element and its one-dimensional filter of neighborhood according to claim 8, which is special
Levy and be:Using the training sample sample in nerve net Algorithm Learning training sample set simpleSet in step 4, god is obtained
It is specially through net forecast model model:
Step 4 one, first element for taking out in simpleSet, obtain length DL of input data input of the element;
Step 4 two, a nerve net is set up, wherein, nerve net input layer includes DL neuron, nerve net intermediate layer is comprising 5
Individual neuron, nerve net output layer corresponds to 1 neuron;
Step 4 three, all elements for passing through in nerve net Algorithm Learning simpleSet, by each element in simpleSet
Sample input data input as input, using output data output of each element in simpleSet as output, according to
Nerve net forecast model model is set up in input and output.
10. according to claim 1 or 9 a kind of remote sensing image picture element and its one-dimensional filter of neighborhood boundary extraction method,
It is characterized in that:By the one-dimensional filter neighborFilter of neighborhood and nerve net forecast model model to defeated in step 5
The all pixels for entering remote sensing image InputMap are processed, and output boundary extracts result detailed process and is:
Step May Day, the width width and height height of acquisition input remote sensing image
Step 5 two, set up a black width width and height height output result figure resultMap
Step 5 three, pixel counter pixelCounter are set to 1;
Step the May 4th, take out InputMap pth ixelCounter pixel on remote sensing image residing for columns x, remote sensing shadow
As residing line number y on image;
Step 5 five, x and y are inputed to the one-dimensional Convolution Filter neighborFilter of neighborhood, obtain output oneArray
Step 5 six, oneArray is inputed to nerve net forecast model model, obtain prediction output result r;
If seven r of step 5 so represents that more than 0.5 currently processed pixel is border, then in result figure resultMap
White on x and y location labelling;
Step 5 eight, pixel counter pixelCounter increase by 1;
If step 5 nine increases the pixelCounter after 1, the pixel number more than InputMap so goes to step May Day
Ten, otherwise go to step the May 4th;
Step May Day ten, using resultMap as Boundary Extraction result output, by resultMap in figure white part make
Output boundary for being extracted extracts result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610898413.5A CN106447687B (en) | 2016-10-14 | 2016-10-14 | A kind of boundary extraction method based on remote sensing image picture element and its one-dimensional filter of neighborhood |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610898413.5A CN106447687B (en) | 2016-10-14 | 2016-10-14 | A kind of boundary extraction method based on remote sensing image picture element and its one-dimensional filter of neighborhood |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106447687A true CN106447687A (en) | 2017-02-22 |
CN106447687B CN106447687B (en) | 2019-02-01 |
Family
ID=58174123
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610898413.5A Active CN106447687B (en) | 2016-10-14 | 2016-10-14 | A kind of boundary extraction method based on remote sensing image picture element and its one-dimensional filter of neighborhood |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106447687B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111062403A (en) * | 2019-12-26 | 2020-04-24 | 哈尔滨工业大学 | Hyperspectral remote sensing data depth spectral feature extraction method based on one-dimensional group convolution neural network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080095461A1 (en) * | 2006-10-20 | 2008-04-24 | Samsung Electronics Co.; Ltd | Apparatus and method for determining filter condition area in deblocking filter |
KR101007384B1 (en) * | 2009-11-26 | 2011-01-13 | 삼성탈레스 주식회사 | Method for adaptive filtering reflecting features of neighbor blocks |
CN104299006A (en) * | 2014-07-23 | 2015-01-21 | 中国传媒大学 | Vehicle license plate recognition method based on deep neural network |
CN105894013A (en) * | 2016-03-29 | 2016-08-24 | 西安电子科技大学 | Method for classifying polarized SAR image based on CNN and SMM |
-
2016
- 2016-10-14 CN CN201610898413.5A patent/CN106447687B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080095461A1 (en) * | 2006-10-20 | 2008-04-24 | Samsung Electronics Co.; Ltd | Apparatus and method for determining filter condition area in deblocking filter |
KR101007384B1 (en) * | 2009-11-26 | 2011-01-13 | 삼성탈레스 주식회사 | Method for adaptive filtering reflecting features of neighbor blocks |
CN104299006A (en) * | 2014-07-23 | 2015-01-21 | 中国传媒大学 | Vehicle license plate recognition method based on deep neural network |
CN105894013A (en) * | 2016-03-29 | 2016-08-24 | 西安电子科技大学 | Method for classifying polarized SAR image based on CNN and SMM |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111062403A (en) * | 2019-12-26 | 2020-04-24 | 哈尔滨工业大学 | Hyperspectral remote sensing data depth spectral feature extraction method based on one-dimensional group convolution neural network |
Also Published As
Publication number | Publication date |
---|---|
CN106447687B (en) | 2019-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109800736B (en) | Road extraction method based on remote sensing image and deep learning | |
CN107220980B (en) | A kind of MRI image brain tumor automatic division method based on full convolutional network | |
CN107609526A (en) | Rule-based fine dimension city impervious surface rapid extracting method | |
CN106845401A (en) | A kind of insect image-recognizing method based on many spatial convoluted neutral nets | |
CN110458201B (en) | Object-oriented classification method and classification device for remote sensing image | |
CN111080652B (en) | Optical remote sensing image segmentation method based on multi-scale lightweight cavity convolution | |
CN106203331A (en) | A kind of crowd density evaluation method based on convolutional neural networks | |
CN107067043A (en) | A kind of diseases and pests of agronomic crop detection method | |
CN107480679A (en) | The road network extracting method of classification and connected component analysis based on convolutional neural networks | |
CN113989681B (en) | Remote sensing image change detection method and device, electronic equipment and storage medium | |
CN110298322B (en) | Cultivated land extraction method and system based on remote sensing data | |
Liang et al. | Segmentation and weight prediction of grape ear based on SFNet-ResNet18 | |
CN112464766A (en) | Farmland automatic identification method and system | |
Zhang et al. | Evaluation of efficacy of fungicides for control of wheat fusarium head blight based on digital imaging | |
CN109635811A (en) | The image analysis method of spatial plant | |
CN114494821A (en) | Remote sensing image cloud detection method based on feature multi-scale perception and self-adaptive aggregation | |
CN114663771A (en) | Mountain farmland intelligent extraction method based on partition layering theory | |
CN106447687A (en) | Boundary extraction method of remote sensing image pixel through neighbor Filter | |
CN113569772A (en) | Remote sensing image farmland instance mask extraction method, system, equipment and storage medium | |
CN107067384B (en) | A kind of loess plateau terraced fields extracting method based on ray method | |
CN117372881A (en) | Intelligent identification method, medium and system for tobacco plant diseases and insect pests | |
CN113469122A (en) | Deep learning based crop space-time generalization classification method and system | |
CN114564893B (en) | Wetland plant diversity monitoring and situation optimization method | |
CN109376742A (en) | A kind of accumulated snow extracting method and system based on remote sensing image | |
CN112861712B (en) | Agricultural pest and disease monitoring method based on artificial intelligence and multi-temporal remote sensing |
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 |