CN110298211A - A kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image - Google Patents
A kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image Download PDFInfo
- Publication number
- CN110298211A CN110298211A CN201810234115.5A CN201810234115A CN110298211A CN 110298211 A CN110298211 A CN 110298211A CN 201810234115 A CN201810234115 A CN 201810234115A CN 110298211 A CN110298211 A CN 110298211A
- Authority
- CN
- China
- Prior art keywords
- river
- point
- sample
- column
- remote sensing
- 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
- G06F18/21—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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/182—Network patterns, e.g. roads or rivers
Abstract
The invention discloses a kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image, comprising: obtain No. two multi-spectrum remote sensing images of more scape high scores, remote sensing image is divided into training sample and detection sample；Remote sensing image pretreatment, image of the subsequent processing only with the 4th wave band and the 2nd wave band；River sample point and non-river sample point are chosen as training sample point；Deep learning model is established, training sample input deep learning model is trained, trained deep learning model is obtained；Divide to obtain detection sample by mesh segmentation method by the remote sensing image in network of waterways region to be extracted；Classification and Identification is carried out by trained model；Generate the network of waterways.It is more true, accurate, fine, continuous that technical solution of the present invention may make the network of waterways of extraction, and not by the interference of the factors such as mountain range shade, building.
Description
Technical field
The invention belongs to remote Sensing Image Analysis, artificial intelligence and hydrology technology crossing domains, are related to hydrographic information extraction
Technology more particularly to a kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image.
Background technique
River network of watershed is a kind of important geographic element, is the important component for constituting topography and geomorphology ＂ skeleton ＂, in number
Play a significant role in the expression of word basin.Traditional Methods Deriving Drainage Network mainly passes through digital elevation model (DEM) extraction.But
It is that existing digital elevation model influences during automatically extracting the virtual network of waterways vulnerable to depression, plains region, generates discontinuous
Or the parallel wrong network of waterways.
In recent years, with the rapid development of remote sensing technology, high-resolution remote sensing image using more more and more universal.Numerous
Person has carried out the research that water body, river are extracted with remote sensing image, and still, the research for extracting the network of waterways using remote sensing image is less.With
Remote sensing image extract water body, river information algorithm mainly include water body index, supervised classification, unsupervised classification, gray level image,
The methods of decision tree classification, mathematical morphology.But the river that these algorithms extract there is also some problems, the river such as extracted
Flow the interference vulnerable to mountain range shade, be difficult to extract tiny river, there is partial interruption in the river of extraction.
Summary of the invention
In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a kind of based on deep learning and high-definition remote sensing shadow
The Methods Deriving Drainage Network of picture, so that the network of waterways extracted is more true, accurate, fine, continuous.
Present invention provide the technical scheme that
A kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image, steps are as follows:
Step 1: obtaining No. two remote sensing images of more scape high scores, determine that each scape is as training sample or as detection sample
This.
The high-resolution remote sensing image that the present invention uses is high score two (GF-2) satellite remote-sensing image.No. two (GF- of high score
2) satellite is that first spatial resolution of China's independent development is better than 1 meter of civilian Optical remote satellite, equipped with two high scores
1 meter of resolution panchromatic, 4 meters of multispectral cameras have sub-meter grade spatial resolution, high position precision and rapid attitude maneuver ability etc.
Feature.The spectral region of 4m resolution multi-spectral be 0.45-0.52 μm, 0.52-0.59 μm, 0.63-0.69 μm, 0.77-0.89 μ
M is successively denoted as the 1st, 2,3,4 wave bands.
Step 2: remote sensing image pretreatment, including ortho-rectification processing and by 4-4-2 band combination at pseudo color coding hologram figure,
Tif format is saved as under ENVI software, so that pixel gray value is up to 255, and further normalizes, method for normalizing
It is the gray value of each pixel divided by 255.Image of the subsequent processing only with the 4th wave band and the 2nd wave band；
Step 3: training sample prepares
(1) input of training sample
On remote sensing image, the point being located on river, including big river and tiny river are chosen, for big river, not only
The sample point at river center is chosen, also to choose river boundaries position and belongs to the sample point in river, it is main for small river
Choose the sample point being located on small river.Record the transverse and longitudinal coordinate of river sample point.
On remote sensing image, the point, including mountain range, shade, cloud, building, farmland, highway, railway in non-river etc. are chosen, it is non-
River sample point tries not to be selected in nearby water bodies, records the transverse and longitudinal coordinate of non-river sample point.
On remote sensing image, centered on the abscissa and ordinate of training sample point, the preceding m/2-1 row of Selection Center point
Right m/2 column are arranged to rear m/2 row, left m/2-1, constitute m × m Square Neighborhood, wherein m belongs to positive integer and is even number；It adopts
With the 4th wave band of image and the 2nd wave band, m × m Square Neighborhood is the tensor of m × m × 2 (being determined by the coordinate of sample point), this is
The input data of one training sample.
(2) output of training sample
The output of each training sample is river or non-river, is two classification problems.The output of river sample is denoted as
The output of [1 0], non-river sample is denoted as [0 1].
Step 4: detection sample prepares
(1) input of sample is detected, detection sample is obtained by the Remote Sensing Image Segmentation in network of waterways region to be extracted；
Remote sensing image of at least scape remote sensing image as network of waterways region to be extracted is obtained, detects sample by the network of waterways to be extracted
The remote sensing image in region is divided to obtain by mesh segmentation method；Mesh segmentation method (is wrapped using a variety of partitioning schemes when specific implementation
Include 16 kinds of partitioning schemes), it is the submatrix that multiple sizes are m row m column by Image Segmentation, obtains the tensor of multiple m × m × 2；It will be every
The tensor of a m × m × 2 is as a detection sample；
(2) output of sample is detected
For the needs of deep learning program operation, the output of detection sample need to be got out, sets each detection here
The output of sample is [0 1].Since network of waterways detection is two classification problems, the result of prediction model output should be two values, export
As a result default to be set to output valve there are two [0 1] i.e. representatives.
Step 5: establishing deep learning model
(1) foundation of model
Using TensorFlow frame, deep learning model is established with python language under Spyde software.Deep learning
The structure of model is 1 input layer, 1 convolutional layer, 1 pond layer, 1 full articulamentum, 1 output layer.
(2) training of model
The input of training sample, output data are put into deep learning model to be trained, and save trained model.
(3) application of model
The model saved is read, the input of each detection sample, output data are put into trained model and are known
Not, recognition result is obtained.
Step 6: generating the network of waterways
(1) for each partitioning scheme, the recognition result of each detection sample is reduced to m × m (such as 16 × 16)
Matrix, each element value in matrix is consistent with recognition result or is 0 or be 1.
(2) for each partitioning scheme, recognition result is combined into complete big matrix by corresponding position when segmentation.
(3) recognition result that different partitioning schemes are obtained is superimposed, that is, big matrix is added, as long as the element in matrix
1 is replaced with greater than 0.
(4) matrix of superposition being corresponded to image up, recognition result is non-river, then image greyscale value is replaced with 0,
Recognition result is that the image greyscale value in river is constant.
(5) since the recognition result of each detection sample can not be completely correct, for the sample of identification mistake, from image
On see it is scattered point, pass through delete small surfaces area image available apparent, the more accurate network of waterways of algorithm.
(6) whether the google map of the network of waterways and corresponding position extracted compares, ideal to detect extraction of drainage effect.
Compared with prior art, the beneficial effects of the present invention are:
(1) compared with common remote sensing image, high-resolution remote sensing image has higher resolution ratio, in conjunction with depth
Learning method can identify more tiny river.
It (two), can be from bigger using Square Neighborhood as recognition unit compared with the identification of the single pixel point of the prior art
Range integrated survey, river image be locally blocked bridge on such as river, river image part not enough it is clear will not image know
Other result.Similarly, it can also preferably identify non-river, reduce interference.
(3) compared with prior art, using the artificial intelligence approach based on deep learning, it is capable of handling bigger number
It is more intelligent according to amount, do not need that threshold value artificially is arranged, be entirely it is adaptive, can preferably identify river, it is ensured that extract
The network of waterways it is more true, accurate.
(4) compared with prior art, post-processing technology, including mesh segmentation method, deletion small area region are increased, is obtained
To be more clear, the network of waterways continuously, finely.
(5) compared with prior art, extraction of drainage is not influenced by Plain, freeze thawing, city, mountain range shade.
Detailed description of the invention
Fig. 1 is the flow diagram of the method for the present invention.
Fig. 2 is mesh segmentation method schematic diagram；
Wherein, F1~F16 respectively indicates 16 kinds of partitioning schemes that mesh segmentation method proposed by the present invention includes；A indicate to
Extract the remote sensing image matrix of the network of waterways；A1x1For the element that the 1st row the 1st of matrix A arranges, A1x16、A16x1、A16x16Respectively matrix A
The 1st row the 16th column, the 16th row the 1st column and the 16th row the 16th column element.
Fig. 3 is No. two remote sensing images of high score for needing to extract the network of waterways in the embodiment of the present invention.
Fig. 4 is the result of extraction of drainage in the embodiment of the present invention.
Fig. 5 is the google map of extraction of drainage corresponding position in the embodiment of the present invention.
Specific embodiment
With reference to the accompanying drawing, the present invention, the model of but do not limit the invention in any way are further described by embodiment
It encloses.
The present invention provides a kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image, realize it is fine,
Network of waterways information is continuously extracted, network of waterways information extraction accuracy is improved and extracts the really degree of result.
Fig. 1 show the stream of the Methods Deriving Drainage Network proposed by the present invention based on deep learning and high-resolution remote sensing image
Journey；Specific implementation step is as follows:
Step 1: obtaining No. two remote sensing images of high score；Determine that each scape is as training sample or as detection sample.
No. two multi-spectrum remote sensing images of more scape high scores are obtained, as the materials of training sample, more scape remote sensing images need to wrap
Contain river, mountain range, city, farmland, highway, railway.In addition an at least scape remote sensing image is also obtained as the network of waterways to be extracted
Materials.Using No. two multi-spectrum remote sensing images of high score, and panchromatic image is not had to.
Step 2: remote sensing image pretreatment, including ortho-rectification processing and by 4-4-2 band combination at pseudo color coding hologram figure,
Tif format is saved as under ENVI software, so that pixel gray value is up to 255, and further normalizes, method for normalizing
It is the gray value of each pixel divided by 255.Image of the subsequent processing only with the 4th wave band and the 2nd wave band；
Step 3: training sample prepares
(1) input of training sample
On remote sensing image, the point for being located at river, including big river and tiny river are chosen.The point in non-river is chosen, is wrapped
Mountain range, shade, cloud, building, farmland, highway, railway etc. are included, the transverse and longitudinal coordinate of sample point is recorded.
On remote sensing image, centered on the transverse and longitudinal coordinate of sample point, preceding 7 row of Selection Center point is arranged to rear 8 row, a left side 7
16 × 16 Square Neighborhoods are constituted to 8 column of the right side, since with two wave bands, each sample is 16 × 16 × 2 tensors, this
For the input data of a training sample.
(2) output of training sample
The output of each training sample is river or non-river, is two classification problems.The output of river sample is denoted as
The output of [1 0], non-river sample is denoted as [0 1],
Step 4: detection sample prepares
(1) input of sample is detected
The remote sensing image for selecting network of waterways region to be extracted uses mesh segmentation method Image Segmentation for 16 × 16 × 2 tensors,
This detects the input data of sample for one.Mesh segmentation method includes 16 kinds of partitioning schemes.
(2) output of sample is detected
For the needs of deep learning program operation, the output of detection sample need to be got out, sets each detection here
The output of sample is [0 1].
Step 5: establishing deep learning model
(1) foundation of model
Using TensorFlow frame, deep learning model is established with python language under Spyde software.Deep learning
The structure of model is 1 input layer, 1 convolutional layer, 1 pond layer, 1 full articulamentum, 1 output layer.
(2) training of model
The input of training sample, output data are put into deep learning model to be trained, and save trained model.
(3) application of model
The model saved is read, the input of each detection sample, output data are put into trained model and are known
Not, recognition result is obtained.
Step 6: generating the network of waterways
(1) for each dividing method, the recognition result of each detection sample is reduced to 16 × 16 matrix, square
Each element value in battle array is consistent with recognition result or is 0 or be 1.
(2) for each dividing method, recognition result is combined into complete big matrix by corresponding position when segmentation.
(3) recognition result that different dividing methods are obtained is superimposed, that is, big matrix is added, as long as the element in matrix
1 is replaced with greater than 0.
(4) matrix of superposition being corresponded to image up, recognition result is that image greyscale value is then replaced with 0 by non-river,
Recognition result is that the image greyscale value in river is constant.
(5) since the recognition result of each detection sample can not be completely correct, for the sample of identification mistake, from image
On see it is scattered point, pass through delete small surfaces area image available apparent, the more accurate network of waterways of algorithm.
(6) whether the google map of the network of waterways and corresponding position extracted compares, ideal to detect extraction of drainage effect.
Embodiment:
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.
Step 1: obtaining No. two remote sensing images of high score
No. two remote sensing images of high score derive from Ministry of Water Resources.
Step 2: remote sensing image pretreatment
(1) under ENVI software, ortho-rectification is done to each scape remote sensing image.
(2) it under ENVI software, the 4-4-2 band combination of the image after ortho-rectification at pseudo color coding hologram figure, protects
Save as tif format, so that pixel gray value is up to 255, and further normalizes, and method for normalizing is each pixel
Gray value is divided by 255.Image of the subsequent processing only with the 4th wave band and the 2nd wave band；
Step 3: training sample prepares
(1) remote sensing image is opened under ENVI software, chooses the point for being located at river, center, big river including big river
Boundary and tiny river center, and record in Excel transverse and longitudinal coordinate a little.20000 river samples are chosen in this example
Point.
(2) remote sensing image is opened under ENVI software, chooses the point in non-river, including mountain range, shade, cloud, building, agriculture
Field, highway, railway etc., and record in Excel transverse and longitudinal coordinate a little.20000 non-river sample points are chosen in this example.
(3) under Matlab software, the 4th and the 2nd two wave band is chosen, 16 are generated centered on the transverse and longitudinal coordinate of sample point
× 16 Square Neighborhood matrixes, since with two wave bands, each sample is 16 × 16 × 2 tensors, this is a trained sample
This input data.
(4) output of training sample is denoted as the output of river sample [1 0], non-river sample due to being two classification
This output is denoted as [0 1], and the output valve of all training samples is generated under Matlab software.
Step 4: detection sample prepares
(1) determine that the remote sensing image in network of waterways region to be extracted as research object, calculates to simplify, this example is from a Jing Jing
It crosses and intercepts a part in the pretreated image of step 2, choose the image of 1600 × 1600 sizes, such as Fig. 3 here.It converges in two rivers
Latitude and longitude coordinates at chalaza, longitude be 114 degree 28 points 52.91 seconds, dimension for 27 degree 7 points 58.37 seconds.
(2) 1600 × 1600 Image Segmentation at 100 × 100=10000 detection sample, each detection sample is 16
× 16 × 2 tensor.
(3) there is also the need to get out the output of detection sample, i.e., raw for the needs of deep learning program operation below
At the matrix that 10000 rows 2 arrange, each behavior [0 1] is intended merely to the needs of program operation, does not represent practical significance.
(4) the more complete network of waterways, the present invention are split using a variety of partitioning schemes, this hair image in order to obtain
It is bright to be referred to as mesh segmentation method, as Fig. 2 uses 16 kinds of partitioning schemes.As shown in Fig. 2, every kind of partitioning scheme is with its starting point
Name, is denoted as partitioning scheme F1~F16 respectively.Point F1~F16 is first sub- square of the remote sensing image in network of waterways region to be extracted
Battle array carries out 4*4 and divides equally 16 obtained elements.
To 1600 × 1600 matrix in this present embodiment, partitioning scheme F1 indicates the impartial segmentation down since point F1
100 rows, the width of every a line are 16, and impartial 100 column of segmentation of turning right since point F1, the width of each column is 16.In this way, segmentation
Mode F1 obtains 100 × 100=10000 submatrix, and the size of each submatrix is the square matrix of 16 rows 16 column, each submatrix
For a detection sample, the tensor that then each detection sample is 16 × 16 × 2.Partitioning scheme F11 is expert at by cut-off point F11
The upper surface of all rows (upper 7 row in submatrix), give up all column (left side 7 column) in submatrix in the left side of F11 point column,
Row and column where F11 point is not given up, and 99 rows of impartial segmentation, the width of every a line are 16 down since point F11, opens from point F11
Begin impartial 99 column of segmentation of turning right, and the width of each column is 16.Finally, giving up bottom remaining row and the remaining column of rightmost.
In this way, partitioning scheme F11 obtains 99 × 99=9801 submatrix, the size of each submatrix is the square matrix of 16 rows 16 column.Its
Its partitioning scheme is similarly analogized.
Step 5: establishing deep learning model
(1) foundation of model
Using TensorFlow frame, deep learning model is established with python language under Spyde software.Deep learning
The structure of model (specific implementation uses convolutional neural networks) is 1 input layer, 1 convolutional layer, 1 pond layer, 1 full connection
Layer, 1 output layer.
(2) training of model
The input of training sample, output data are put into deep learning model to be trained, it is 30000 times trained in total, often
Primary training randomly selects 50 samples, preservation model after training.
(3) application of model
The model saved is read, the input data for the detection sample that 16 kinds of partitioning schemes obtain is put into trained model
Classification and Identification is carried out, output result is two values, the probability that the detection sample belongs to river, non-river is respectively represented, which
It is non-river that probability, which then illustrates that greatly the detection sample belongs to river also, to obtain the classification results of each detection sample.For
The following convenience for generating network of waterways description is that river is denoted as 1 the result that detection specimen discerning obtains, for being denoted as non-river
0。
Step 6: generating the network of waterways
(1) recognition result that each detection sample obtains is 1 or 0.If 1,16 × 16 matrixes are reduced to, matrix
Each element is 1；If 0, it is reduced to 16 × 16 null matrix.Recognition result is combined into completely by corresponding position when segmentation
Big matrix.
1st kind of dividing method F1 is combined into 1600 × 1,600 0,1 matrix, remaining 15 kinds of dividing method be combined into (16 ×
99) 0,1 matrix of × (16 × 99), according to segmentation when corresponding position, the boundary that use 0 fills up surrounding obtain 1600 × 1600 0,
1 matrix.
To which each dividing method all obtains one corresponding 1600 × 1,600 0,1 matrix.16 kinds of dividing methods
Obtained 0,1 matrix superposition summation, the element in matrix replace with 1 as long as being greater than 0, obtain in this way one comprehensive 1600 ×
1600 0,1 matrix.Rule of thumb, it is not necessarily intended to 16 results to be all superimposed, there may come a time when only with several in 16
A effect is more preferable.This step can pass through Matlab software realization.
(2) it is intercepted 1600 × 1,600 0,1 matrix of the synthesis that detection specimen discerning obtains and after pretreatment
1600 × 1600 remote sensing images are corresponded to, if corresponding position is 0, the gray value of image is replaced with 0；If 1, then image
Gray value remain unchanged.
Specifically, due to 1600 × 1,600 0,1 matrix of obtained synthesis with by pretreatment after intercept 1600 ×
1600 remote sensing image matrix sizes are the same, can be corresponding by the transverse and longitudinal coordinate of matrix.To the member in two matrixes of same location
Element compares, and the matrix of remote sensing image is denoted as A, and the 0-1 matrix of recognition result is denoted as B, if B (i, j)=0, A (i, j)=
0；If B (i, j) is remained unchanged not equal to 0, A (i, j).
It is scattered point from image for the sample of identification mistake since recognition result may be not exclusively accurate, leads to
The available better network of waterways figure of algorithm for deleting small surfaces area image is crossed, tif format is finally saved as.This step exists
It is realized under Matlab software.
(3) network of waterways remote sensing image that extraction is opened under ENVI software checks extraction of drainage as a result, such as Fig. 4.
(4) map in corresponding geographical location is found on google map according to the longitude and latitude of detection sample, such as Fig. 5,
It can be seen from the figure that the more continuous network of waterways can be extracted using Methods Deriving Drainage Network proposed by the present invention, and can mention
Take the more tiny river than showing on current country's google map.
It should be noted that the purpose for publicizing and implementing example is to help to further understand the present invention, but the skill of this field
Art personnel, which are understood that, not to be departed from the present invention and spirit and scope of the appended claims, and various substitutions and modifications are all
It is possible.Therefore, the present invention should not be limited to embodiment disclosure of that, and the scope of protection of present invention is with claim
Subject to the range that book defines.
Claims (10)
1. a kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image, includes the following steps:
Step 1: obtaining No. two multi-spectrum remote sensing images of more scape high scores, remote sensing image is divided into training sample and detection sample；
No. two multi-spectrum remote sensing images of high score the 1st, 2,3, the spectral regions of 4 wave bands be respectively 0.45-0.52 μm,
0.52-0.59 μm, 0.63-0.69 μm, 0.77-0.89 μm；
Step 2: remote sensing image pretreatment, including ortho-rectification processing and by 4-4-2 band combination at pseudo color coding hologram figure, go forward side by side one
Step normalization, image of the subsequent processing only with the 4th wave band and the 2nd wave band；
Step 3: training sample prepares；
31) training sample point is obtained, as mode input；It performs the following operations:
On training sample remote sensing image, river sample point and non-river sample point are chosen as training sample point, record training
The abscissa and ordinate of sample point；
On remote sensing image, centered on the abscissa and ordinate of training sample point, after the preceding m/2-1 row of Selection Center point arrives
M/2 row, left m/2-1 arrange right m/2 column, constitute m × m Square Neighborhood, and wherein m belongs to positive integer and is even number；Using shadow
As the 4th wave band and the 2nd wave band, m × m Square Neighborhood is the tensor of m × m × 2, as a training sample；
32) setting model exports；
The recognition result of each training sample is river or non-river, and output result is two values, respectively represents the sample category
In river and the probability for belonging to non-river；The output of river sample is denoted as [1 0], the output of non-river sample is denoted as [0 1]；
33) deep learning model is established, training sample input deep learning model is subjected to classification based training, obtains trained depth
Spend learning model；
Step 4: detection sample prepares；
41) remote sensing image of at least scape remote sensing image as network of waterways region to be extracted is obtained, detects sample by the network of waterways to be extracted
The remote sensing image in region is divided to obtain by mesh segmentation method；The mesh segmentation method uses a variety of partitioning schemes, by image point
It is segmented into the submatrix that multiple sizes are m row m column, obtains the tensor of multiple m × m × 2；It is examined the tensor of each m × m × 2 as one
Test sample sheet；
42) Classification and Identification is carried out by trained model；
Each detection sample data is put into trained model and carries out Classification and Identification, recognition result is obtained, thus identifies
To river or non-river, it is expressed as 1 or 0；
Step 5: the network of waterways is generated, is performed the following operations:
51) recognition result of each detection sample is reduced to the matrix of m × m, each element value and identification in matrix are tied
Fruit is consistent, and the matrix element value for belonging to river classification is 1, and the matrix element value for belonging to non-river classification is 0；
52) it is directed to each partitioning scheme, corresponding position when by recognition result by segmentation is combined, complete after being combined
Whole big matrix；
53) the big matrix after the combination for obtaining different partitioning schemes is added, and obtains superposition matrix；If being superimposed the element in matrix
Value is greater than 0, then the element value replaces with 1；
54) superposition matrix is corresponded to image；If recognition result is non-river, image greyscale value is replaced with 0；If identification knot
Fruit is river, and image greyscale value is constant；Thus it obtains the network of waterways and shows image；
Through the above steps, it realizes and the network of waterways is extracted based on deep learning and high-resolution remote sensing image and is shown.
2. Methods Deriving Drainage Network as described in claim 1, characterized in that the network of waterways obtained for step 54) shows image, leads to
Point scattered in the algorithm deletion image for delete small surfaces area image is crossed, for removing the sample of identification mistake.
3. Methods Deriving Drainage Network as described in claim 1, characterized in that step 2) is specifically by pseudo color coding hologram figure under ENVI software
Save as tif format so that pixel gray value is up to 255, and further by by the gray value of each pixel divided by 255 into
Row normalization.
4. Methods Deriving Drainage Network as described in claim 1, characterized in that step 31) obtains on training sample remote sensing image
Training sample point, the river sample point are specifically chosen the point being located on river and are obtained；River includes big river and tiny river；
Big river sample point includes being located at the sample point at river center and being located at river boundaries and belong to the sample point in river；The non-river
Flow the sample point on mountain range, shade, cloud, building, farmland, highway, railway that sample point includes separate water body.
5. Methods Deriving Drainage Network as described in claim 1, characterized in that preferably, step 31) is with the horizontal seat of training sample point
Centered on mark and ordinate, preceding 7 row of Selection Center point constitutes 16 × 16 Square Neighborhoods to rear 8 row, 7 column of a left side to 8 column of the right side,
Each training sample is 16 × 16 × 2 tensors；The Remote Sensing Image Segmentation in network of waterways region to be extracted is multiple big by step 41)
Small is the submatrix of 16 rows 16 column, obtains multiple 16 × 16 × 2 tensors；Using each 16 × 16 × 2 tensor as a detection sample
This；The recognition result of each detection sample is reduced to 16 × 16 matrix by step 51).
6. Methods Deriving Drainage Network as described in claim 1, characterized in that step 31) establishes deep learning model, concrete application
TensorFlow frame and Spyde software are realized using python language；The structure of the deep learning model includes input
Layer, convolutional layer, pond layer, full articulamentum and output layer.
7. Methods Deriving Drainage Network as described in claim 1, characterized in that step 41) obtains detection sample by mesh segmentation method
This, the mesh segmentation method includes 16 kinds of partitioning schemes.
8. Methods Deriving Drainage Network as claimed in claim 7, characterized in that 16 kinds of partitioning schemes are denoted as F1~F16；Segmentation
Mode F1 specifically: it is assumed that remote sensing image data is the square matrix of n row n column, n=km, n, k, m belong to positive integer, open from point F1
Begin equalization segmentation k row down, and the width of every a line is m, and the impartial segmentation k that turns right since point F1 is arranged, and the width of each column is m；
K*k submatrix is obtained, the size of each submatrix is the square matrix of m row m column, and each submatrix is a detection sample；
Point F1~F16 is that first submatrix of the remote sensing image in network of waterways region to be extracted carries out 16 that 4*4 is respectively obtained again
Element.
9. Methods Deriving Drainage Network as claimed in claim 8, characterized in that partitioning scheme F11 is on cut-off point F11 is of the row
All rows in face give up all column in the left side of F11 point column；It is impartial down the row and column where point F11 to divide k-1 row,
The width of every a line is m, and impartial segmentation k-1 column of turning right since point F11, the width of each column is m；Finally, giving up bottom
Remaining row and the remaining column of rightmost；Thus a submatrix of (k-1) × (k-1) is obtained, the size of each submatrix is m row m
The square matrix of column.
10. Methods Deriving Drainage Network as claimed in claim 7, characterized in that partitioning scheme F1 specifically: since point F1 down
Equalization 100 rows of segmentation, the width of every a line are 16, and impartial 100 column of segmentation of turning right since point F1, the width of each column is 16；
10000 submatrixs are obtained, the size of each submatrix is the square matrix of 16 rows 16 column, and each submatrix is a detection sample；
Partitioning scheme F11 is cut-off point F11 all above row of the row, gives up all column in the left side of F11 point column；From point F11
Place row and column starts 99 rows of impartial segmentation down, and the width of every a line is 16, impartial 99 column of segmentation of turning right since point F11,
The width of each column is 16；Finally, giving up bottom remaining row and the remaining column of rightmost；Thus 99 × 99=9801 is obtained
A submatrix, the size of each submatrix are the square matrix of 16 rows 16 column.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810234115.5A CN110298211B (en) | 2018-03-21 | 2018-03-21 | River network extraction method based on deep learning and high-resolution remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810234115.5A CN110298211B (en) | 2018-03-21 | 2018-03-21 | River network extraction method based on deep learning and high-resolution remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110298211A true CN110298211A (en) | 2019-10-01 |
| CN110298211B CN110298211B (en) | 2021-03-23 |
Family
ID=68025336
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810234115.5A Active CN110298211B (en) | 2018-03-21 | 2018-03-21 | River network extraction method based on deep learning and high-resolution remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110298211B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110992330A (en) * | 2019-11-28 | 2020-04-10 | 桂林理工大学 | Multi-level integral relaxation matching high-resolution ortho-image shadow detection under artificial shadow drive |
| CN111178230A (en) * | 2019-12-26 | 2020-05-19 | 武汉大学 | Intelligent extraction method for river beach in remote sensing image |
| CN111738168A (en) * | 2020-06-24 | 2020-10-02 | 中水北方勘测设计研究有限责任公司 | Satellite image river two-side sand production extraction method and system based on deep learning |
| CN112084843A (en) * | 2020-07-28 | 2020-12-15 | 北京工业大学 | Multispectral river channel remote sensing monitoring method based on semi-supervised learning |
| CN112785521A (en) * | 2021-01-19 | 2021-05-11 | 澜途集思生态科技集团有限公司 | Remote sensing image processing method under haze condition |
| CN113033386A (en) * | 2021-03-23 | 2021-06-25 | 广东电网有限责任公司广州供电局 | High-resolution remote sensing image-based transmission line channel hidden danger identification method and system |
| CN113888670A (en) * | 2021-10-18 | 2022-01-04 | 珠江水利委员会珠江水利科学研究院 | Method for generating high-precision two-dimensional terrain of one-way river channel based on deep learning |
| CN114155253A (en) * | 2021-12-09 | 2022-03-08 | 电子科技大学 | Connection method for strip-shaped element fracture based on remote sensing image segmentation |
| CN116630811A (en) * | 2023-06-07 | 2023-08-22 | 自然资源部国土卫星遥感应用中心 | River extraction method, river extraction device, terminal equipment and readable storage medium |
| CN117392539A (en) * | 2023-10-13 | 2024-01-12 | 哈尔滨师范大学 | River water body identification method based on deep learning, electronic equipment and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855490A (en) * | 2012-07-23 | 2013-01-02 | 黑龙江工程学院 | Object-neural-network-oriented high-resolution remote-sensing image classifying method |
| CN104462494A (en) * | 2014-12-22 | 2015-03-25 | 武汉大学 | Remote sensing image retrieval method and system based on non-supervision characteristic learning |
| CN105303198A (en) * | 2015-11-17 | 2016-02-03 | 福州大学 | Remote-sensing image semi-supervision classification method based on customized step-size learning |
| CN106127784A (en) * | 2016-07-01 | 2016-11-16 | 辽宁工程技术大学 | A kind of high-resolution remote sensing image dividing method |
| CN106991411A (en) * | 2017-04-17 | 2017-07-28 | 中国科学院电子学研究所 | Remote Sensing Target based on depth shape priori becomes more meticulous extracting method |
| US20170293825A1 (en) * | 2016-04-08 | 2017-10-12 | Wuhan University | Method and system for reconstructing super-resolution image |
| CN107247938A (en) * | 2017-06-08 | 2017-10-13 | 中国科学院遥感与数字地球研究所 | A kind of method of high-resolution remote sensing image City Building function classification |
| CN107610141A (en) * | 2017-09-05 | 2018-01-19 | 华南理工大学 | A kind of remote sensing images semantic segmentation method based on deep learning |
| CN107688818A (en) * | 2016-08-05 | 2018-02-13 | 中国电力科学研究院 | A kind of path intelligent selecting method and system based on satellite remote-sensing image signature analysis |
| CN107729922A (en) * | 2017-09-20 | 2018-02-23 | 千寻位置网络有限公司 | Remote sensing images method for extracting roads based on deep learning super-resolution technique |
-
2018
- 2018-03-21 CN CN201810234115.5A patent/CN110298211B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855490A (en) * | 2012-07-23 | 2013-01-02 | 黑龙江工程学院 | Object-neural-network-oriented high-resolution remote-sensing image classifying method |
| CN104462494A (en) * | 2014-12-22 | 2015-03-25 | 武汉大学 | Remote sensing image retrieval method and system based on non-supervision characteristic learning |
| CN105303198A (en) * | 2015-11-17 | 2016-02-03 | 福州大学 | Remote-sensing image semi-supervision classification method based on customized step-size learning |
| US20170293825A1 (en) * | 2016-04-08 | 2017-10-12 | Wuhan University | Method and system for reconstructing super-resolution image |
| CN106127784A (en) * | 2016-07-01 | 2016-11-16 | 辽宁工程技术大学 | A kind of high-resolution remote sensing image dividing method |
| CN107688818A (en) * | 2016-08-05 | 2018-02-13 | 中国电力科学研究院 | A kind of path intelligent selecting method and system based on satellite remote-sensing image signature analysis |
| CN106991411A (en) * | 2017-04-17 | 2017-07-28 | 中国科学院电子学研究所 | Remote Sensing Target based on depth shape priori becomes more meticulous extracting method |
| CN107247938A (en) * | 2017-06-08 | 2017-10-13 | 中国科学院遥感与数字地球研究所 | A kind of method of high-resolution remote sensing image City Building function classification |
| CN107610141A (en) * | 2017-09-05 | 2018-01-19 | 华南理工大学 | A kind of remote sensing images semantic segmentation method based on deep learning |
| CN107729922A (en) * | 2017-09-20 | 2018-02-23 | 千寻位置网络有限公司 | Remote sensing images method for extracting roads based on deep learning super-resolution technique |
Non-Patent Citations (2)
| Title |
|---|
| 刘扬 等: "高分辨率遥感影像目标分类与识别研究进展", 《地球信息科学学报》 * |
| 杜培军 等: "高光谱遥感影像分类研究进展", 《遥感学报》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110992330A (en) * | 2019-11-28 | 2020-04-10 | 桂林理工大学 | Multi-level integral relaxation matching high-resolution ortho-image shadow detection under artificial shadow drive |
| CN111178230A (en) * | 2019-12-26 | 2020-05-19 | 武汉大学 | Intelligent extraction method for river beach in remote sensing image |
| CN111738168A (en) * | 2020-06-24 | 2020-10-02 | 中水北方勘测设计研究有限责任公司 | Satellite image river two-side sand production extraction method and system based on deep learning |
| CN112084843B (en) * | 2020-07-28 | 2024-03-12 | 北京工业大学 | Multispectral river channel remote sensing monitoring method based on semi-supervised learning |
| CN112084843A (en) * | 2020-07-28 | 2020-12-15 | 北京工业大学 | Multispectral river channel remote sensing monitoring method based on semi-supervised learning |
| CN112785521A (en) * | 2021-01-19 | 2021-05-11 | 澜途集思生态科技集团有限公司 | Remote sensing image processing method under haze condition |
| CN113033386A (en) * | 2021-03-23 | 2021-06-25 | 广东电网有限责任公司广州供电局 | High-resolution remote sensing image-based transmission line channel hidden danger identification method and system |
| CN113888670A (en) * | 2021-10-18 | 2022-01-04 | 珠江水利委员会珠江水利科学研究院 | Method for generating high-precision two-dimensional terrain of one-way river channel based on deep learning |
| CN113888670B (en) * | 2021-10-18 | 2022-05-20 | 珠江水利委员会珠江水利科学研究院 | Method for generating high-precision two-dimensional terrain of one-way river channel based on deep learning |
| CN114155253B (en) * | 2021-12-09 | 2023-04-07 | 电子科技大学 | Connection method for strip-shaped element fracture based on remote sensing image segmentation |
| CN114155253A (en) * | 2021-12-09 | 2022-03-08 | 电子科技大学 | Connection method for strip-shaped element fracture based on remote sensing image segmentation |
| CN116630811A (en) * | 2023-06-07 | 2023-08-22 | 自然资源部国土卫星遥感应用中心 | River extraction method, river extraction device, terminal equipment and readable storage medium |
| CN116630811B (en) * | 2023-06-07 | 2024-01-02 | 自然资源部国土卫星遥感应用中心 | River extraction method, river extraction device, terminal equipment and readable storage medium |
| CN117392539A (en) * | 2023-10-13 | 2024-01-12 | 哈尔滨师范大学 | River water body identification method based on deep learning, electronic equipment and storage medium |
| CN117392539B (en) * | 2023-10-13 | 2024-04-09 | 哈尔滨师范大学 | River water body identification method based on deep learning, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110298211B (en) | 2021-03-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110298211A (en) | A kind of Methods Deriving Drainage Network based on deep learning and high-resolution remote sensing image | |
| Jia et al. | Spectral matching based on discrete particle swarm optimization: A new method for terrestrial water body extraction using multi-temporal Landsat 8 images | |
| CN109934200B (en) | RGB color remote sensing image cloud detection method and system based on improved M-Net | |
| Fang et al. | Recognizing global reservoirs from Landsat 8 images: A deep learning approach | |
| CN111881816A (en) | Long-time-sequence river and lake ridge culture area monitoring method | |
| Verhulp et al. | Transferability of decision trees for land cover classification in a heterogeneous area | |
| Xiong et al. | Comparison of artificial neural network and support vector machine methods for urban land use/cover classifications from remote sensing images a case study of Guangzhou, South China | |
| CN116012723A (en) | Wetland type extraction method and device based on time sequence remote sensing image and electronic equipment | |
| Sahu et al. | Conceptual framework of combined pixel and object-based method for delineation of debris-covered glaciers | |
| Liu et al. | Vietnam wetland cover map: Using hydro-periods Sentinel-2 images and Google Earth Engine to explore the mapping method of tropical wetland | |
| Liu et al. | Environment monitoring of Shanghai Nanhui intertidal zone with dual-polarimetric SAR data based on deep learning | |
| Dibs et al. | Detection wetland dehydration extent with multi-temporal remotely sensed data using remote sensing analysis and GIS techniques | |
| Parajuli et al. | Attentional dense convolutional neural network for water body extraction from sentinel-2 images | |
| Zhang | A time-series approach to detect urbanized areas using biophysical indicators and landsat satellite imagery | |
| Huang et al. | An improved deep learning approach for retrieving outfalls into rivers from UAS imagery | |
| Li et al. | Characterizing ancient channel of the yellow river from spaceborne sar: Case study of chinese gaofen-3 satellite | |
| Qi et al. | An adaptive threshold selected method from remote sensing image based on water index | |
| Khan et al. | Analyzing the Land Cover Change and Degradation in Sundarbans Mangrove Forest Using Machine Learning and Remote Sensing Technique | |
| Hao et al. | A subpixel mapping method for urban land use by reducing shadow effects | |
| Dehghani | Study of changes in land use and land cover in and around the Chah Nimeh water reservoir in Sistan and Baluchistan Province, Iran | |
| Sekertekin | Mapping of surface water resources: a comparative analysis of eight image classification methods | |
| Liu et al. | A new convolutional neural networks for land use classification | |
| Yashin et al. | Comparative analysis of classification algorithms for landuse/landcover change over a part of the east coast region of tamil nadu and its environs | |
| Lehrbass et al. | Techniques for object-based classification of urban tree cover from high-resolution multispectral imagery | |
| Matci | AUTOMATIC DELINEATION OF WATER BODIES USING CORINE DATA FROM REMOTELY SENSED IMAGES |
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 |