CN108898603A - Plot segmenting system and method on satellite image - Google Patents
Plot segmenting system and method on satellite image Download PDFInfo
- Publication number
- CN108898603A CN108898603A CN201810534898.9A CN201810534898A CN108898603A CN 108898603 A CN108898603 A CN 108898603A CN 201810534898 A CN201810534898 A CN 201810534898A CN 108898603 A CN108898603 A CN 108898603A
- Authority
- CN
- China
- Prior art keywords
- plot
- satellite image
- image
- unit
- prediction
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30188—Vegetation; Agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the plot segmenting system and method on a kind of satellite image, wherein the plot segmenting system on satellite image includes：First unit, obtains training pictures, and the trained picture has plot and edge labelling；Second unit obtains satellite image to be split；Third unit, builds neural network model, and the third unit is trained neural network model to obtain prediction model with the trained pictures；Satellite image to be split is inputted the prediction model and carries out plot prediction by the third unit；Unit the 4th carries out connective region search to the forecast image of prediction model output；Unit the 5th compares the connected domain area searched and preset threshold interval, wherein the connected domain area in the threshold interval, then the connected domain is plot.The mark speed of the plot on satellite image can be improved in the present invention.
Description
Technical field
The present invention relates to satellite image processing technology field, in particular to plot segmenting system on a kind of satellite image and
Method.
Background technique
The present invention belongs to the relevant technologies related to the present invention for the description of background technique, be only used for explanation and just
In understanding summary of the invention of the invention, it should not be construed as applicant and be specifically identified to or estimate applicant being considered of the invention for the first time
The prior art for the applying date filed an application.
Satellite image technology has had the application of many decades, but in agriculture fine-grained management, applied satellite image pair
Farmland, which is managed, is not applied effectively also.With the development of science and technology, agricultural modernization needs more, more advanced technology to do
It supports.And satellite image is utilized, the division in plot is carried out, is the first step of land management and planning, satellite image is drawn in plot
There is natural advantage in point, however, basic or place also immature at present using the method that satellite image carries out Parcel division
In the stage of complete artificial mark boundary of land block, however the crop-planting type on soil is becoming, and boundary of land block is also to change, this
With regard to needing constantly to repeat this work, moreover, satellite image size is generally large, complete artificial mark, the time needed
Cost is also very big.
Summary of the invention
In view of this, the embodiment of the present invention provides plot segmenting system and method on a kind of satellite image, main purpose
It is the plot mark speed improved on satellite image.
In order to achieve the above objectives, present invention generally provides following technical solutions：
In a first aspect, the embodiment of the invention provides the plot segmenting systems on a kind of satellite image, including：
First unit, obtains training pictures, and the trained picture has plot and edge labelling；
Second unit obtains satellite image to be split；
Third unit builds neural network model, and the third unit is with the trained pictures to neural network model
It is trained to obtain prediction model；The third unit is pre- by satellite image input prediction model progress plot to be split
It surveys；
Unit the 4th carries out connective region search to the forecast image of prediction model output；
Unit the 5th compares the connected domain area searched and preset threshold interval, wherein
The connected domain area is in the threshold interval, then the connected domain is plot.
Preferably, the first unit includes cutting module and flip module, the module that cuts is marked to plot
The satellite image sample of label is cut, and the flip module overturns the satellite image cut, and the satellite image sample is by having
The satellite image sample of plot label is cut and overturning obtains.
Preferably, the first unit further includes color correction module, the extension rectification module is to cutting and overturn
The satellite image sample afterwards carries out color correction, and the color correction includes in histogram stretching and Mean-Variance standard
At least one.
Preferably, the first unit further includes expansion process module, the expansion process module is to the satellite shadow
Decent progress expansion process, the satellite image sample cutting after expansion process obtain training pictures.
Preferably, the first layer in the neural network model is using expansion convolution.
Preferably, the loss function that different phase generates in the neural network model is weighting decreasing fashion.
Preferably, further including post-processing unit, the post-processing unit includes Morphological scale-space module, the morphology
Processing module first carries out closing in morphology to the forecast image that the prediction model exports before carrying out the connective region search
At least one of operation and skeletal extraction processing.
Preferably, the post-processing unit further includes simple interest module, the simple interest module is to the prediction
The forecast image of model output first carries out simple interest processing before carrying out the connective region search, so that ground block edge is straight
The connection of line gap.
Preferably, the line processing includes：
It is slided in the image handled to simple interest using the window of a 3*3 pixel, when the window center of sliding is
When 1, the sum of the pixel value in window being judged, assert that the point is endpoint when the sum of pixel is 2, area is then executed since the point
Domain growth algorithm stops after increasing enough points n, then with this n point, carries out straight line fitting using least square method, so
Afterwards along the slope direction of fitting a straight line, increase the value of x-axis to the right, then stopping when finding corresponding y value and being also 1 connects
The two points are connect, image or so overturning are then executed into the above process again, the image after simple interest can be obtained.
Preferably, the prediction model carries out plot prediction to satellite image to be split using redundant prediction method.
Second aspect, the embodiment of the invention provides the plot dividing methods on a kind of satellite image, include the following steps：
Training pictures are obtained, the trained picture has plot and edge labelling；
Neural network model is trained to obtain prediction model with the trained pictures；
Satellite image to be split is inputted into the prediction model and carries out plot prediction；
Connective region search is carried out to the forecast image of prediction model output；
The connected domain area searched and preset threshold interval are compared, wherein
The connected domain area is in the threshold interval, then the connected domain is plot.
Preferably, the trained pictures are obtained by the satellite image sample cutting with plot label, the cutting
Including cutting to the satellite image sample and being overturn to the satellite image cut.
Preferably, carrying out color correction, the color correction includes histogram Tula after the satellite image sample cutting
At least one of stretch with Mean-Variance standard.
Preferably, the satellite image sample carries out expansion process, the satellite image sample after expansion process is cut
Get training pictures.
Preferably, the size of the trained picture is 200*200 pixel between 1000*1000 pixel.
Preferably, the first layer in the neural network model is using expansion convolution.
Preferably, the loss function that different phase generates in the neural network model is weighting decreasing fashion.
Preferably, the forecast image of the prediction model output first carries out form before carrying out the connective region search
At least one of closed operation and skeletal extraction are handled in.
Preferably, the forecast image of the prediction model output first carries out straight line before carrying out the connective region search
Growth process, so that the straight line gap of ground block edge connects.
Preferably, the line processing includes：
It is slided in the image handled to simple interest using the window of a 3*3 pixel, when the window center of sliding is
When 1, the sum of the pixel value in window being judged, assert that the point is endpoint when the sum of pixel is 2, area is then executed since the point
Domain growth algorithm stops after increasing enough points n, then with this n point, carries out straight line fitting using least square method, so
Afterwards along the slope direction of fitting a straight line, increase the value of x-axis to the right, then stopping when finding corresponding y value and being also 1 connects
The two points are connect, image or so overturning are then executed into the above process again, the image after simple interest can be obtained.
Preferably, the prediction model carries out plot prediction to satellite image to be split using redundant prediction method.
The third aspect, the embodiment of the invention provides a kind of computer readable storage mediums, are stored thereon with computer journey
The step of sequence, which realizes above-mentioned method when being executed by processor.
Fourth aspect the embodiment of the invention provides a kind of computer equipment, including memory, processor and is stored in
On reservoir and the computer program that can run on a processor, the processor realize above-mentioned method when executing described program
Step.
Compared with prior art, the beneficial effect of the embodiment of the present invention is：
Plot segmenting system and method satellite image on satellite image provided in an embodiment of the present invention, with deep learning
Based on model, automatically obtain ground after then having carried out a series of post-processing using the model output result after deep learning
Result after block segmentation.Cost of labor and the duplication of labour are greatly reduced, while greatly improving speed.The present invention couple
Future, basic impetus and technical support were played in the fine-grained management in plot.
Scheme attached explanation
Fig. 1 shows the schematic diagram of an embodiment of the plot segmenting system on satellite image of the present invention.
Fig. 2 shows the schematic diagrames of another embodiment of the plot segmenting system on satellite image of the present invention.
Fig. 3 shows the flow chart of an embodiment of the plot dividing method on satellite image of the present invention.
Fig. 4 shows the flow chart of another embodiment of the plot dividing method on satellite image of the present invention.
Specific embodiment
Present invention is further described in detail combined with specific embodiments below, but not as a limitation of the invention.?
In following the description, what different " embodiment " or " embodiment " referred to is not necessarily the same embodiment.In addition, one or more are implemented
Special characteristic, structure or feature in example can be combined by any suitable form.
Fig. 1 shows the schematic diagram of an embodiment of the plot segmenting system on satellite image of the present invention.Fig. 2 shows this
The schematic diagram of another embodiment of plot segmenting system on invention satellite image.Referring to Fig. 1 and Fig. 2, on a kind of satellite image
Plot segmenting system, including：
First unit 10, obtains training pictures, and training picture has plot and edge labelling；
Second unit 20 obtains satellite image to be split；
Third unit 30 builds neural network model, and third unit 30 is to train pictures to carry out neural network model
Training obtains prediction model；Satellite image input prediction model to be split is carried out plot prediction by third unit 30；
4th unit 40 carries out connective region search to the forecast image of prediction model output；
5th unit 50 compares the connected domain area searched and preset threshold interval, wherein
If connected domain area, in the threshold interval, which is plot.
Plot segmenting system satellite image on satellite image provided in an embodiment of the present invention is with deep learning model
Basis automatically obtains plot segmentation after then having carried out a series of post-processing using the model output result after deep learning
Result afterwards.Cost of labor and the duplication of labour are greatly reduced, while greatly improving speed.The present invention is in the future
The impetus and technical support on basis are played in the fine-grained management of block.
In one embodiment of the invention, training pictures are from training sample cutting, and training sample is that a width includes
The satellite image of hundreds of thousands of pixels, cannot be directly used to model training.Shapefile vector file can not be directly used in mould
Type training needs for label file (shapefile file) to be converted into the label image with sizes such as training images.Therefore,
One unit 10 carries out cutting to the satellite image with plot label, and above-mentioned cutting includes cutting and overturning.Referring to fig. 2, one
In a embodiment, first unit 10 includes cutting module 11 and flip module 12, cuts module 11 and defends to plot label
Star image sample is cut, and flip module 12 overturns the satellite image cut, and satellite image sample is by defending with plot label
Star image sample is cut and overturning obtains.Therefore, with plot label satellite image sample in block edge and non-edge portion
The value for dividing imparting different is marked, for example, plot marginal value is 1 in image, remaining place value is 0.
In one embodiment, referring to fig. 2, first unit 10 further includes expansion process module 13, expansion process module 13
Expansion process is carried out to satellite image sample, the satellite image sample cutting after expansion process obtains training pictures.This implementation
The robustness of model can be enhanced in example, while eliminating the deviation that label label occurs, and carrying out expansive working to label image makes to mark
In label block edge it is more obvious.After expansion process, using cutting out at random, the operations such as overturning obtain as more as possible while superfluous
The few training pictures of remaining property, such as thousands of Zhang Xunlian pictures can be obtained.It is all a certain size instruction in training pictures
Practice picture.The size of training picture can be 200*200 pixel between 1000*1000 pixel.For example, training picture size is
512*512 pixel.
Satellite image is usually to have made of more scape image mosaics.Since the time of satellite shooting is different, weather, light,
Angle etc. all has image to image quality, therefore will appear serious color error ratio.If directly this picture is trained
It is unfavorable for model convergence.Therefore, in one embodiment, referring to fig. 2, first unit 10 further includes color correction module 14, is prolonged
Stretch rectification module 14 to after cutting and overturning satellite image sample carry out color correction, color correction include histogram stretch and
At least one of Mean-Variance standard.In the present embodiment, to the training picture after segmentation, (such as 512*512 pixel is big
It is small) operations such as histogram stretching, Mean-Variance standardization have been carried out, eliminate the influence of illumination, weather etc..
In one embodiment of the invention, model construction is based on edge parted pattern, and considers satellite image and traditional natural
Image difference is larger, and the method for having abandoned pre-training weight is directly trained from the beginning.
In one embodiment, the first layer in neural network model makes model compared to common convolution using expansion convolution
Obtain better receptive field.
In one embodiment of the invention, model loss function uses Focal Loss, the positive and negative sample proportion in balance sample,
It is more preferable compared to cross entropy loss function effect.
In another embodiment, the loss function that different phase generates in neural network model is weighting decreasing fashion.
In one embodiment of the invention, the deep learning frame Tensorflow to be increased income using Google builds above-mentioned nerve
The learning rate that successively decreases is arranged in network model, tall and handsome up to the left back model convergence of 10 hours of training on P100 video card.In training process
Weight parameter need to be adjusted for the distribution of the output of model each section (such as five parts), expand expansion ratio, the loss letter of volume machine
Hyper parameter for including in number etc. and then re -training.Until model each section can preferably predict image border, while mould
The value of type loss function can deconditioning when tending towards stability.The prediction model obtained at this time can be used for dividedly block edge.
It further include post-processing unit in one embodiment of the invention to improve the effect of plot segmentation, it is referring to fig. 2, rear to locate
Managing unit 60 includes Morphological scale-space module 61, and the forecast image that Morphological scale-space module 61 exports prediction model is connecting
At least one of closed operation and skeletal extraction processing in morphology are first carried out before logical domain search.The output of model is and inputs
Image single channel image of a size exports the every bit on image and is regarded as whether corresponding input picture point is plot side
The probability of edge.Such as choosing 0.5 is cut-point, is ground block edge more than or equal to 0.5, is background less than 0.5, is obtained preliminary
Segmentation result, the result is thicker in the presence of the edge being partitioned into, and having in cavity and background in edge has the case where noise spot, therefore,
The present embodiment using in morphological image closed operation and framework extraction method it is available eliminating noise, and by unidirectional
The segmentation result of the image border of vegetarian refreshments composition.
In one embodiment of the invention, post-processing unit 60 further includes simple interest module 62, simple interest module
The forecast image of 62 pairs of prediction models output first carries out simple interest processing before carrying out connective region search, so that ground block edge
Straight line gap connection.Due to model accuracy, plot changes situations such as complicated, the image of prediction model output, even across
The phenomenon that Morphological scale-space, also there is also broken string (the straight line disconnections of description ground block edge).To solve this problem, the present embodiment
A kind of scheme of simple interest is proposed, first with the window of a simple 3*3 pixel in the image handled to simple interest
Middle sliding judges the sum of the pixel value in window when the window center of sliding is 1, assert that the point is when the sum of pixel is 2
Then endpoint executes region growing algorithm since the point, increasing enough points n (can according to need adjustment, such as n=
15) stop after, then with this n point, straight line fitting is carried out using least square method, then along the slope side of fitting a straight line
To the right then the value of increase x-axis, stopping when finding corresponding y value and being also 1 connect the two points, then by an image left side
Right overturning executes the above process again, the image after simple interest can be obtained.
In the embodiment of the present invention, it can receive the input picture of arbitrary size in prediction, then use the method for redundant prediction
Entire image is predicted, the serious disadvantage of the splicing edge noise occurred when image cutting prediction is spliced then is eliminated.
Second aspect, the embodiment of the invention provides the plot dividing method on a kind of satellite image, Fig. 3 shows this hair
The flow chart of one embodiment of the plot dividing method on bright satellite image.Fig. 4 shows the plot on satellite image of the present invention
The flow chart of another embodiment of dividing method.Referring to Fig. 3 and Fig. 4, the plot on the satellite image of the embodiment of the present invention is divided
Method includes the following steps：
Training pictures are obtained, training picture has plot and edge labelling；
To train pictures to be trained to obtain prediction model to neural network model；
Satellite image input prediction model to be split is subjected to plot prediction；
Connective region search is carried out to the forecast image of prediction model output；
The connected domain area searched and preset threshold interval are compared, wherein
Connected domain area is in threshold interval, then the connected domain is plot.
Plot dividing method satellite image on satellite image provided in an embodiment of the present invention is with deep learning model
Basis automatically obtains plot segmentation after then having carried out a series of post-processing using the model output result after deep learning
Result afterwards.Cost of labor and the duplication of labour are greatly reduced, while greatly improving speed.The present invention is in the future
The impetus and technical support on basis are played in the fine-grained management of block.
In one embodiment of the invention, training pictures are from training sample cutting, and training sample is that a width includes
The satellite image of hundreds of thousands of pixels, cannot be directly used to model training.Shapefile vector file can not be directly used in mould
Type training needs for label file (shapefile file) to be converted into the label image with sizes such as training images.Therefore,
One unit 10 carries out cutting to the satellite image with plot label, and above-mentioned cutting includes cutting and overturning.Referring to fig. 4, one
In a embodiment, training pictures are obtained by the satellite image sample cutting with plot label, and cutting includes to the satellite
Image sample cuts and overturns to the satellite image cut.Therefore, with plot label satellite image sample in block edge
And non-edge part assigns different values and is marked, for example, plot marginal value is 1 in image, remaining place value is 0.
Referring to fig. 4, in one embodiment of the invention, after satellite image sample cutting, color correction is carried out, color correction includes
At least one of histogram stretching and Mean-Variance standard.The robustness of model can be enhanced in the present embodiment, while eliminating mark
The deviation that label label occurs, to label image carry out expansive working make in label block edge it is more obvious.After expansion process,
Using cutting out at random, the operations acquisition such as overturning is as more as possible, while the training pictures that redundancy is few, such as can obtain thousands of
Zhang Xunlian pictures.It is all a certain size training picture in training pictures.The size of training picture can be 200*200 picture
Element is between 1000*1000 pixel.For example, training picture size is 512*512 pixel.In the embodiment of the present invention, expansion process
Both expansion process first can be carried out to satellite image before cutting, the slicing operations such as is then cut out, overturns, it can also be with
It is that expansion process is carried out to the picture that cutting obtains, obtains training pictures after the slicing operations such as cutting out, overturning.
Satellite image is usually to have made of more scape image mosaics.Since the time of satellite shooting is different, weather, light,
Angle etc. all has image to image quality, therefore will appear serious color error ratio.If directly this picture is trained
It is unfavorable for model convergence.Therefore, in one embodiment, referring to fig. 4, after the cutting of satellite image sample, color correction, institute are carried out
Stating color correction includes at least one of histogram stretching and Mean-Variance standard.In the present embodiment, to the instruction after segmentation
Practice picture (such as 512*512 pixel size) and carried out the operations such as histogram stretching, Mean-Variance standardization, eliminates illumination, day
The influence of gas etc..
In one embodiment of the invention, model construction is based on edge parted pattern, and considers satellite image and traditional natural
Image difference is larger, and the method for having abandoned pre-training weight is directly trained from the beginning.
In one embodiment, the first layer in neural network model makes model compared to common convolution using expansion convolution
Obtain better receptive field.
In one embodiment of the invention, model loss function uses Focal Loss, the positive and negative sample proportion in balance sample,
It is more preferable compared to cross entropy loss function effect.
In another embodiment, the loss function that different phase generates in neural network model is weighting decreasing fashion.
In one embodiment of the invention, the deep learning frame Tensorflow to be increased income using Google builds above-mentioned nerve
The learning rate that successively decreases is arranged in network model, tall and handsome up to the left back model convergence of 10 hours of training on P100 video card.In training process
Weight parameter need to be adjusted for the distribution of the output of model each section (such as five parts), expand expansion ratio, the loss letter of volume machine
Hyper parameter for including in number etc. and then re -training.Until model each section can preferably predict image border, while mould
The value of type loss function can deconditioning when tending towards stability.The prediction model obtained at this time can be used for dividedly block edge.
In order to improve the effect of plot segmentation, in one embodiment of the invention, the forecast image of prediction model output is being carried out
At least one of closed operation and skeletal extraction processing in morphology are first carried out before the connective region search.The output of model is
With input picture single channel image of a size, export the every bit on image be regarded as corresponding input picture point whether be
The probability of ground block edge.Such as choosing 0.5 is cut-point, is ground block edge more than or equal to 0.5, is background less than 0.5, obtains
To primary segmentation, as a result, the result exists, the edge being partitioned into is thicker, and having in cavity and background in edge has the case where noise spot,
Therefore, the present embodiment using in morphological image closed operation and framework extraction method it is available eliminating noise, Yi Jiyou
The segmentation result of the image border of unidirectional vegetarian refreshments composition.
In one embodiment of the invention, the forecast image of prediction model output is before carrying out the connective region search
Simple interest processing is first carried out, so that the straight line gap of ground block edge connects.Due to model accuracy, plot changes the feelings such as complexity
Condition, the image of prediction model output, even across Morphological scale-space, also there is also broken string (the straight line disconnections of description ground block edge)
The phenomenon that.To solve this problem, the present embodiment proposes a kind of scheme of simple interest, first with a simple 3*3 picture
The window of element slides in the image handled to simple interest, when the window center of sliding is 1, judges the pixel value in window
The sum of, assert that the point is endpoint when the sum of pixel is 2, region growing algorithm is then executed since the point, increases enough
Points n (can according to need adjustment, such as n=15) stop afterwards, then with this n point, carry out straight line using least square method
Fitting increases to the right the value of x-axis then along the slope direction of fitting a straight line, stopping when finding corresponding y value and being also 1,
Then the two points are connected, image or so overturning are then executed into the above process again, the image after simple interest can be obtained.
In the embodiment of the present invention, it can receive the input picture of arbitrary size in prediction, then use the method for redundant prediction
Entire image is predicted, the serious disadvantage of the splicing edge noise occurred when image cutting prediction is spliced then is eliminated.
The third aspect, the embodiment of the invention provides a kind of computer readable storage mediums, are stored thereon with computer journey
The step of sequence, which realizes above-mentioned method when being executed by processor.
Fourth aspect the embodiment of the invention provides a kind of computer equipment, including memory, processor and is stored in
On reservoir and the computer program that can run on a processor, the processor realize above-mentioned method when executing described program
Step.
Those skilled in the art can be understood that the embodiment of the present invention technical solution can by software and/or
Hardware is realized." unit " or " unit " in this specification is to refer to complete independently or complete with other component cooperation specific
The software and/or hardware of function, wherein hardware for example can be FPGA (Field-Programmable Gate Array, scene
Programmable gate array), IC (Integrated Circuit, integrated circuit) etc..
The embodiment of the invention also provides a kind of computer readable storage mediums, are stored thereon with computer program, the journey
The step of any of the above-described embodiment method is realized when sequence is executed by processor.Wherein, computer readable storage medium may include
But be not limited to any kind of disk, including floppy disk, CD, DVD, CD-ROM, mini drive and magneto-optic disk, ROM, RAM,
EPROM, EEPROM, DRAM, VRAM, flash memory device, magnetic or optical card, nanosystems (including molecular memory lC),
Or it is suitable for any kind of medium or equipment of store instruction and/or data.
The embodiment of the invention also provides a kind of computer equipments, including memory, processor and storage are on a memory
And the computer program that can be run on a processor, the step of realizing any of the above-described embodiment method when processor executes program.
In embodiments of the present invention, processor is the control centre of computer system, can be the processor of physical machine, is also possible to void
The processor of quasi- machine.
In the present invention, term " first ", " second " etc. are only used for the purpose of description, are not understood to indicate or imply
Relative importance or sequence；Term " multiple " then refers to two or more, unless otherwise restricted clearly.Term " installation ",
The terms such as " connected ", " connection ", " fixation " shall be understood in a broad sense, for example, " connection " may be a fixed connection, being also possible to can
Dismantling connection, or be integrally connected；" connected " can be directly connected, can also be indirectly connected through an intermediary.For this
For the those of ordinary skill in field, the specific meanings of the above terms in the present invention can be understood according to specific conditions.
In description of the invention, it is to be understood that the orientation or positional relationship of the instructions such as term " on ", "lower" be based on
Orientation or positional relationship shown in the drawings, is merely for convenience of description of the present invention and simplification of the description, rather than indication or suggestion institute
The device or unit of finger must have specific direction, be constructed and operated in a specific orientation, it is thus impossible to be interpreted as to this hair
Bright limitation.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (23)
1. the plot segmenting system on satellite image, which is characterized in that including：
First unit, obtains training pictures, and the trained picture has plot and edge labelling；
Second unit obtains satellite image to be split；
Third unit, builds neural network model, and the third unit carries out neural network model with the trained pictures
Training obtains prediction model；Satellite image to be split is inputted the prediction model and carries out plot prediction by the third unit；
Unit the 4th carries out connective region search to the forecast image of prediction model output；
Unit the 5th compares the connected domain area searched and preset threshold interval, wherein
The connected domain area is in the threshold interval, then the connected domain is plot.
2. system according to claim 1, which is characterized in that the first unit includes cutting module and flip module,
The module that cuts cuts the satellite image sample with plot label, and the flip module turns over the satellite image cut
Turn, the satellite image sample is cut by the satellite image sample with plot label and overturning obtains.
3. system according to claim 2, which is characterized in that the first unit further includes color correction module, described
Extend rectification module and color correction is carried out to the satellite image sample after cutting and overturning, the color correction includes histogram
Tula at least one of is stretched with Mean-Variance standard.
4. system according to claim 1, which is characterized in that the first unit further includes expansion process module, described
Expansion process module carries out expansion process to the satellite image sample, and the satellite image sample cutting after expansion process obtains
To training pictures.
5. system according to claim 1, which is characterized in that the first layer in the neural network model is using expansion volume
Product.
6. system according to claim 1, which is characterized in that the loss that different phase generates in the neural network model
Function is weighting decreasing fashion.
7. system according to claim 1, which is characterized in that further include post-processing unit, the post-processing unit includes
Morphological scale-space module, the Morphological scale-space module are carrying out the connected domain to the forecast image that the prediction model exports
At least one of closed operation and skeletal extraction processing in morphology are first carried out before search.
8. system according to claim 7, which is characterized in that the post-processing unit further includes simple interest module, institute
It states simple interest module and straight line is first carried out before carrying out the connective region search to the forecast image that the prediction model exports
Growth process, so that the straight line gap of ground block edge connects.
9. system according to claim 8, which is characterized in that the line processing includes：
It is slided in the image handled to simple interest using the window of a 3*3 pixel, when the window center of sliding is 1,
Judge the sum of the pixel value in window, assert that the point is endpoint when the sum of pixel is 2, then executes region since the point and increase
Long algorithm stops after increasing enough points n, then with this n point, carries out straight line fitting using least square method, then edge
The slope direction of fitting a straight line, increase the value of x-axis to the right, then stopping when finding corresponding y value and being also 1 connects this
Then image or so overturning is executed the above process, the image after simple interest can be obtained by two points again.
10. system according to claim 1, which is characterized in that the prediction model is using redundant prediction method to be split
Satellite image carry out plot prediction.
11. the plot dividing method on satellite image, includes the following steps：
Training pictures are obtained, the trained picture has plot and edge labelling；
Neural network model is trained to obtain prediction model with the trained pictures；
Satellite image to be split is inputted into the prediction model and carries out plot prediction；
Connective region search is carried out to the forecast image of prediction model output；
The connected domain area searched and preset threshold interval are compared, wherein
The connected domain area is in the threshold interval, then the connected domain is plot.
12. according to the method for claim 11, which is characterized in that the trained pictures are by the satellite with plot label
Image sample cutting obtains, and the cutting includes cutting to the satellite image sample and overturning to the satellite image cut.
13. according to the method for claim 12, which is characterized in that after the satellite image sample cutting, carry out color and rectify
Just, the color correction includes at least one of histogram stretching and Mean-Variance standard.
14. according to the method for claim 11, which is characterized in that the satellite image sample carries out expansion process, expansion
The satellite image sample cutting that treated obtains training pictures.
15. according to the method for claim 11, which is characterized in that the size of the trained picture be 200*200 pixel extremely
Between 1000*1000 pixel.
16. according to the method for claim 11, which is characterized in that the first layer in the neural network model is using expansion
Convolution.
17. according to the method for claim 11, which is characterized in that the damage that different phase generates in the neural network model
Losing function is weighting decreasing fashion.
18. according to the method for claim 11, which is characterized in that the forecast image of the prediction model output is carrying out institute
At least one of closed operation and skeletal extraction processing in morphology are first carried out before stating connective region search.
19. according to the method for claim 11, which is characterized in that the forecast image of the prediction model output is carrying out institute
Simple interest processing is first carried out before stating connective region search, so that the straight line gap of ground block edge connects.
20. according to the method for claim 19, which is characterized in that the line processing includes：
It is slided in the image handled to simple interest using the window of a 3*3 pixel, when the window center of sliding is 1,
Judge the sum of the pixel value in window, assert that the point is endpoint when the sum of pixel is 2, then executes region since the point and increase
Long algorithm stops after increasing enough points n, then with this n point, carries out straight line fitting using least square method, then edge
The slope direction of fitting a straight line, increase the value of x-axis to the right, then stopping when finding corresponding y value and being also 1 connects this
Then image or so overturning is executed the above process, the image after simple interest can be obtained by two points again.
21. according to the method for claim 11, which is characterized in that the prediction model is using redundant prediction method to be split
Satellite image carry out plot prediction.
22. a kind of computer readable storage medium, is stored thereon with computer program, power is realized when which is executed by processor
Benefit requires the step of any one of 11-21 the method.
23. a kind of computer equipment including memory, processor and stores the meter that can be run on a memory and on a processor
The step of calculation machine program, the processor realizes any one of claim 11-21 the method when executing described program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810534898.9A CN108898603A (en) | 2018-05-29 | 2018-05-29 | Plot segmenting system and method on satellite image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810534898.9A CN108898603A (en) | 2018-05-29 | 2018-05-29 | Plot segmenting system and method on satellite image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108898603A true CN108898603A (en) | 2018-11-27 |
Family
ID=64343452
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810534898.9A Pending CN108898603A (en) | 2018-05-29 | 2018-05-29 | Plot segmenting system and method on satellite image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108898603A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109766400A (en) * | 2019-01-14 | 2019-05-17 | 北京天下图数据技术有限公司 | Plot labeling method and device, block identifying method and device |
| CN110189341A (en) * | 2019-06-05 | 2019-08-30 | 北京青燕祥云科技有限公司 | A kind of method, the method and device of image segmentation of Image Segmentation Model training |
| CN111210095A (en) * | 2020-03-12 | 2020-05-29 | 深圳前海微众银行股份有限公司 | Power generation amount prediction method, device, equipment and computer readable storage medium |
| CN111382645A (en) * | 2018-12-29 | 2020-07-07 | 顺丰科技有限公司 | Method and system for identifying expired buildings in electronic map |
| CN112949711A (en) * | 2021-02-26 | 2021-06-11 | 中国科学院软件研究所 | Neural network model reusable training method and device for software-defined satellite |
| US20220156493A1 (en) * | 2020-11-17 | 2022-05-19 | Insurance Services Office, Inc. | Computer Vision Systems and Methods for Detecting and Aligning Land Property Boundaries on Aerial Imagery |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103218618A (en) * | 2013-01-09 | 2013-07-24 | 重庆交通大学 | Highway route automatic extraction method based on remote-sensing digital image |
| CN103761526A (en) * | 2014-01-26 | 2014-04-30 | 北京理工大学 | Urban area detecting method based on feature position optimization and integration |
| CN106384081A (en) * | 2016-08-30 | 2017-02-08 | 水利部水土保持监测中心 | Slope farmland extracting method and system based on high-resolution remote sensing image |
| CN107154048A (en) * | 2017-05-09 | 2017-09-12 | 中国科学院遥感与数字地球研究所 | The remote sensing image segmentation method and device of a kind of Pulse-coupled Neural Network Model |
| CN107274418A (en) * | 2017-07-07 | 2017-10-20 | 江苏省无线电科学研究所有限公司 | A kind of crop image partition method based on AP clustering algorithms |
| CN107341795A (en) * | 2017-06-30 | 2017-11-10 | 武汉大学 | A kind of high spatial resolution remote sense image method for detecting automatic variation of Knowledge driving |
-
2018
- 2018-05-29 CN CN201810534898.9A patent/CN108898603A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103218618A (en) * | 2013-01-09 | 2013-07-24 | 重庆交通大学 | Highway route automatic extraction method based on remote-sensing digital image |
| CN103761526A (en) * | 2014-01-26 | 2014-04-30 | 北京理工大学 | Urban area detecting method based on feature position optimization and integration |
| CN106384081A (en) * | 2016-08-30 | 2017-02-08 | 水利部水土保持监测中心 | Slope farmland extracting method and system based on high-resolution remote sensing image |
| CN107154048A (en) * | 2017-05-09 | 2017-09-12 | 中国科学院遥感与数字地球研究所 | The remote sensing image segmentation method and device of a kind of Pulse-coupled Neural Network Model |
| CN107341795A (en) * | 2017-06-30 | 2017-11-10 | 武汉大学 | A kind of high spatial resolution remote sense image method for detecting automatic variation of Knowledge driving |
| CN107274418A (en) * | 2017-07-07 | 2017-10-20 | 江苏省无线电科学研究所有限公司 | A kind of crop image partition method based on AP clustering algorithms |
Non-Patent Citations (2)
| Title |
|---|
| 冯子亮 等: "一种基于主动生长的边缘连接算法", 《计算机应用研究》 * |
| 量子学园: "Kanggle优胜者详解：如何用深度学习实现卫星图像分割与识别", 《知乎：HTTPS://ZHUANLAN.ZHIHU.COM/P/26377387》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111382645A (en) * | 2018-12-29 | 2020-07-07 | 顺丰科技有限公司 | Method and system for identifying expired buildings in electronic map |
| CN111382645B (en) * | 2018-12-29 | 2023-11-28 | 顺丰科技有限公司 | Method and system for identifying overdue building in electronic map |
| CN109766400A (en) * | 2019-01-14 | 2019-05-17 | 北京天下图数据技术有限公司 | Plot labeling method and device, block identifying method and device |
| CN110189341A (en) * | 2019-06-05 | 2019-08-30 | 北京青燕祥云科技有限公司 | A kind of method, the method and device of image segmentation of Image Segmentation Model training |
| CN110189341B (en) * | 2019-06-05 | 2021-08-10 | 北京青燕祥云科技有限公司 | Image segmentation model training method, image segmentation method and device |
| CN111210095A (en) * | 2020-03-12 | 2020-05-29 | 深圳前海微众银行股份有限公司 | Power generation amount prediction method, device, equipment and computer readable storage medium |
| CN111210095B (en) * | 2020-03-12 | 2022-02-18 | 深圳前海微众银行股份有限公司 | Power generation amount prediction method, device, equipment and computer readable storage medium |
| US20220156493A1 (en) * | 2020-11-17 | 2022-05-19 | Insurance Services Office, Inc. | Computer Vision Systems and Methods for Detecting and Aligning Land Property Boundaries on Aerial Imagery |
| US11847824B2 (en) * | 2020-11-17 | 2023-12-19 | Insurance Services Office, Inc. | Computer vision systems and methods for detecting and aligning land property boundaries on aerial imagery |
| CN112949711A (en) * | 2021-02-26 | 2021-06-11 | 中国科学院软件研究所 | Neural network model reusable training method and device for software-defined satellite |
| CN112949711B (en) * | 2021-02-26 | 2023-10-27 | 中国科学院软件研究所 | Neural network model multiplexing training method and device for software defined satellites |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108898603A (en) | Plot segmenting system and method on satellite image | |
| CN106157319B (en) | The conspicuousness detection method in region and Pixel-level fusion based on convolutional neural networks | |
| CN109859171B (en) | Automatic floor defect detection method based on computer vision and deep learning | |
| CN106157267A (en) | A kind of image mist elimination absorbance optimization method based on dark channel prior | |
| CN101571950B (en) | Image restoring method based on isotropic diffusion and sparse representation | |
| WO2020098158A1 (en) | Pedestrian re-recognition method and apparatus, and computer readable storage medium | |
| CN109410129A (en) | A kind of method of low light image scene understanding | |
| CN108776772B (en) | Cross-time building change detection modeling method, detection device, method and storage medium | |
| CN101777180B (en) | Complex background real-time alternating method based on background modeling and energy minimization | |
| CN108319972A (en) | A kind of end-to-end difference online learning methods for image, semantic segmentation | |
| CN103164855B (en) | A kind of Bayesian decision foreground extracting method in conjunction with reflected light photograph | |
| CN108198201A (en) | A kind of multi-object tracking method, terminal device and storage medium | |
| CN106204494B (en) | A kind of image defogging method and system comprising large area sky areas | |
| CN105678813A (en) | Skin color detection method and device | |
| CN103942758A (en) | Dark channel prior image dehazing method based on multiscale fusion | |
| CN109584170A (en) | Underwater image restoration method based on convolutional neural networks | |
| CN106339984A (en) | Distributed image super-resolution method based on K-means driven convolutional neural network | |
| CN103971338A (en) | Variable-block image repair method based on saliency map | |
| CN110310241A (en) | A kind of more air light value traffic image defogging methods of fusion depth areas segmentation | |
| CN106686278A (en) | Heterogeneous processor parallel coordination processing method facing real-time object identification | |
| CN109858487A (en) | Weakly supervised semantic segmentation method based on watershed algorithm and image category label | |
| CN110148093A (en) | A kind of image defogging improved method based on dark channel prior | |
| CN103578083A (en) | Single image defogging method based on joint mean shift | |
| CN110223251A (en) | Suitable for manually with the convolutional neural networks underwater image restoration method of lamp | |
| CN108986101A (en) | Human body image dividing method based on circulation " scratching figure-segmentation " optimization |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181127 |