CN105374033B - SAR image segmentation method based on ridge ripple deconvolution network and sparse classification - Google Patents
SAR image segmentation method based on ridge ripple deconvolution network and sparse classification Download PDFInfo
- Publication number
- CN105374033B CN105374033B CN201510675676.5A CN201510675676A CN105374033B CN 105374033 B CN105374033 B CN 105374033B CN 201510675676 A CN201510675676 A CN 201510675676A CN 105374033 B CN105374033 B CN 105374033B
- Authority
- CN
- China
- Prior art keywords
- mrow
- sketch
- region
- sar image
- line
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
Abstract
The invention discloses a kind of SAR image segmentation method based on ridge ripple deconvolution network and sparse classification.Mainly solve the problems, such as that prior art needs artificial experience to extract characteristics of image.Implementation step is as follows：(1) sketch is combined to aperture radar SAR image；(2) it is different semantic regions to divide SAR image；(3) ridge ripple deconvolution network RDN aggregation zone and homogenous region is respectively trained；(4) similar aggregation zone is merged；(5) similar homogenous region is merged；(6) dividing ridge method is based on, the structural region obtained to step (2) is split；(7) the synthetic aperture radar SAR image after being split.The segmentation result of the present invention has preferable region consistency, and improves the segmentation effect of synthetic aperture radar SAR image, available for Target detection and identification.
Description
Technical field
The invention belongs to technical field of image processing, the one kind further related in target identification technology field is based on ridge
Ripple deconvolution network and the synthetic aperture radar of sparse classification (Synthetic Aperture Radar, SAR) image segmentation side
Method.The present invention can be split exactly to the different zones of synthetic aperture radar SAR image, and can be used for follow-up conjunction
Into the object detection and recognition of aperture radar SAR image.
Background technology
The segmentation of synthetic aperture radar SAR image refers to synthetic aperture according to features such as gray scale, texture, structure, aggregations
Radar SAR image is divided into several mutually disjoint regions, and these features is showed similitude in the same area, and
The process of obvious otherness is showed between different zones.The purpose of synthetic aperture radar SAR image segmentation is to simplify or change
Become the representation of image so that image is easier to understand and analyzed.The segmentation of synthetic aperture radar SAR image be image understanding with
The basis of interpretation, the quality for splitting quality directly affect follow-up analysis, identification etc..Generally, split it is more accurate, identification more into
Work(.
Existing synthetic aperture radar SAR image segmentation method is broadly divided into the method for feature based and based on statistical model
Method.The method of feature based is mainly that the feature for extracting some synthetic aperture radar SAR images is split, such as texture
Feature, side feature and composite character etc..Based on the method for statistical model by synthetic aperture radar SAR image segmentation problem with generally
The mode of rate is expressed, and the feature of image is described as to the distribution of some experiences, such as Nakagami distributions, Gamma are distributed, K points
Cloth, G distributions etc..
Linda, Xu Xin, Pan Xuefeng, paper that Haitao Zhang is delivered at it " a kind of new MSTAR SAR image segmentation methods,
A kind of new MSTAR SAR image segmentation methods are proposed in 2014,39 (11) ".This method is first according to the scattering machine of atural object
System carries out attribute scattering center feature extraction, structure attribute scattering center characteristic vector, then using markov random file knot
Close attribute scattering center feature and space neighbour city relationship description is carried out to MSTAR SAR images, it is finally excellent with label cost energy
Change algorithm and obtain final segmentation result.Weak point is existing for this method, and this method is entered to synthetic aperture radar SAR image
It is characterized in what is manually extracted used in row segmentation, artificial selected characteristic is a method that is very laborious, needing professional knowledge,
The feature that can have been chosen largely leans on experience and fortune, therefore the quality for the feature manually chosen often turns into whole
The bottleneck of individual systematic function.
Patent " the SAR image segmentation side based on depth own coding and administrative division map that Xian Electronics Science and Technology University applies at it
Disclosed in method " (number of patent application 201410751944.2, publication number CN104392456 A) it is a kind of based on depth own coding and
The SAR image segmentation method of administrative division map.The region that this method is divided according to the sketch map of synthetic aperture radar SAR image
Figure, by administrative division map be mapped to artwork assembled, homogeneous and structural region；Respectively to aggregation, the different depth in homogenous region
Self-encoding encoder is trained, and obtains the feature that aggregation and homogenous region are each put；Dictionary, each point are built to aggregation and homogenous region respectively
Projection is to corresponding dictionary and converges out the provincial characteristics of all subregion, and the sub-district characteristic of field in two class regions is clustered respectively；
Structural region is merged under the guidance of sketch line segment using super-pixel and split；Merge each region segmentation result and complete SAR figures
As segmentation.Weak point is existing for this method, and the input of the depth self-encoding encoder for automatically extracting characteristics of image used is one
Dimensional vector, destroys the spatial structure characteristic of image, it is thus impossible to extract the substantive characteristics of image, reduces SAR image segmentation
Precision.
The content of the invention
It is an object of the invention to overcome the shortcomings of above-mentioned prior art, propose that one kind is based on ridge ripple deconvolution neutral net
With (Synthetic Aperture Radar, SAR) image partition method of sparse classification.This method is based on ridge ripple deconvolution god
Through network, not only can learning sample automatically feature, broken through the bottleneck of artificial extraction feature, and can learn to synthesize hole
Spatial relationship in the radar SAR image of footpath between pixel, the substantive characteristics of synthetic aperture radar SAR image is extracted, improves synthesis
The segmentation effect of aperture radar SAR image.
To achieve the above object, present invention specific implementation step includes as follows：
(1) sketch is combined to aperture radar SAR image：
To the synthetic aperture radar SAR image sketch of input, the sketch map of synthetic aperture radar SAR image is obtained；
(2) it is different semantic regions to divide synthetic aperture radar SAR image：
Using ray completion administrative division map, synthetic aperture radar SAR image is divided into aggregation zone, homogenous region and structure
Region；
(3) ridge ripple deconvolution network RDN aggregation zone and homogenous region is respectively trained：
(3a) constructs 4 layers of ridge ripple deconvolution network RDN；
(3b) utilizes ridge ripple function, respectively to 3 warp laminations in 4 layers of ridge ripple deconvolution network RDN having constructed
Wave filter group is initialized；
(3c) disconnected each homogenous region to spatially disconnected each aggregation zone and spatially, is respectively trained
One 4 layers of ridge ripple deconvolution network RDN, obtains the optimal value of ridge ripple deconvolution network RDN median filter groups；
(4) similar aggregation zone is merged：
(4a) trains the filter of last layers of gained ridge ripple deconvolution network RDN with spatially disconnected each aggregation zone
Disconnected each aggregation zone on the optimal value representation space of ripple device group；
The method that (4b) uses sparse classification, calculating represent architectural feature phase between spatially disconnected each aggregation zone
Like estimating for property；
Aggregation zone architectural feature similarity measure is more than the corresponding region of its threshold tau as similar accumulation regions by (4c)
Domain, merge all similar aggregation zones, wherein, τ represents the threshold value of aggregation zone architectural feature similarity measure, τ value
Scope is τ ∈ [0,1]；
(5) similar homogenous region is merged：
(5a) trains the filter of last layers of gained ridge ripple deconvolution network RDN with spatially disconnected each homogenous region
Disconnected each homogenous region on the optimal value representation space of ripple device group；
The method that (5b) uses sparse classification, calculating represent architectural feature phase between spatially disconnected each homogenous region
Like estimating for property；
(5c) is using corresponding region of the homogenous region architectural feature similarity measure more than its threshold value σ as similar homogeneous area
Domain, merge all similar homogenous regions, wherein, σ represents the threshold value of homogenous region architectural feature similarity measure, σ value
Scope is σ ∈ [0,1]；
(6) structural region is split：
The structural region obtained to step (2) is split, and obtains the segmentation result of structural region；
(7) the synthetic aperture radar SAR image after being split：
The homogenous region that the aggregation zone and step (5) obtained using step (4) is obtained, and the knot that step (6) obtains
Structure region, the synthetic aperture radar SAR image after being split.
The present invention has advantages below compared with prior art：
First, because the present invention constructs 4 layers of ridge ripple deconvolution network RDN, prior art is overcome to synthetic aperture
Radar SAR image split used in the shortcomings that being characterized in manually extracting so that figure can be automatically extracted using the present invention
The feature of picture, it is more time saving and energy saving than artificial extraction feature, and the characteristics of image automatically extracted is than the feature manually extracted more
It is accurate to add.
Second, because the present invention utilizes ridge ripple function, respectively to 3 in 4 layers of ridge ripple deconvolution network RDN having constructed
The wave filter group of warp lamination is initialized, and overcoming the depth self-encoding encoder that prior art automatically extracts characteristics of image does not have
The shortcomings that paying close attention to the architectural characteristic of image so that the structural specificity feature of image can be automatically extracted using the present invention, is improved
The precision of SAR image segmentation.
3rd, due to the present invention to spatially disconnected each aggregation zone and spatially disconnected each homogeneous area
4 layers of ridge ripple deconvolution network RDN are respectively trained in domain, overcome the depth own coding that prior art automatically extracts characteristics of image
Device is not concerned with image the shortcomings that spatial relationship between pixel so that the essence that image can be automatically extracted using the present invention is special
Sign, therefore, obtains more preferable region segmentation uniformity.
Brief description of the drawings
Fig. 1 is the flow chart of the present invention；
Fig. 2 is the analogous diagram of the present invention.
Embodiment
The present invention will be further described below in conjunction with the accompanying drawings.
Referring to the drawings 1, of the invention comprises the following steps that.
Step 1, sketch is combined to aperture radar SAR image.
Synthetic aperture radar SAR image is inputted, by its sketch, obtains the sketch map of synthetic aperture radar SAR image.
The synthetic aperture radar SAR image sketch model that the present invention uses is that Jie-Wu et al. was published in IEEE in 2014
Article on Transactions on Geoscience and Remote Sensing magazines《Local maximal
homogenous region search for SAR speckle reduction with sketch-based
geometrical kernel function》Proposed in model.
Construct the side with different directions and yardstick, line template, and using the direction of template and dimensional information construction it is each to
Anisotropic Gaussian function calculates the weight coefficient of every bit in the template, and its mesoscale number value is 3~5, and direction number takes
It is worth for 18.
According to the following formula, the average and variance of the corresponding pixel in synthetic aperture radar SAR image of calculation template different zones：
Wherein, μ represents the average of the corresponding pixels in synthetic aperture radar SAR image of region Ω, and Ω is represented in template
One region, g represent the position of a pixel in the Ω of region, and ∈ represents to belong to symbol, and ∑ represents sum operation, wgRepresent area
Weight coefficient in the Ω of domain at the g of position, wgSpan be wg∈ [0,1], AgPosition g is corresponding in expression region Ω is synthesizing
Pixel value in aperture radar SAR image, ν represent the variance of the corresponding pixels in synthetic aperture radar SAR image of region Ω.
According to the following formula, the response of each pixel comparison value operator in synthetic aperture radar SAR image is calculated：
Wherein, R represents the response of each pixel comparison value operator in synthetic aperture radar SAR image, and min { } is represented
Minimize operation, a and b represent the numbering of any two different zones in template, μ respectivelyaAnd μbRegion a and area are represented respectively
Domain b and the average of respective pixel in synthetic aperture radar SAR image.
According to the following formula, response of each pixel to correlation operator in calculating synthetic aperture radar SAR image：
Wherein, C represents that each pixel is to the response of correlation operator, ν in synthetic aperture radar SAR imageaAnd νbRespectively
The variance of the corresponding pixels in synthetic aperture radar SAR image of region a and region b is represented,Represent square root functions.
According to the following formula, the response and synthetic aperture thunder of pixel comparison value operator in synthetic aperture radar SAR image are merged
Pixel is to the response of correlation operator up in SAR image, calculates in synthetic aperture radar SAR image each pixel to each mould
The response of plate：
Wherein, F represents that each pixel is to the response of each template in synthetic aperture radar SAR image.
Template template as in synthetic aperture radar SAR image pixel of the selection with maximum response, and by maximum
Intensity of the response as the pixel, the direction using the direction of the template with maximum response as the pixel, is synthesized
The sideline response diagram and directional diagram of aperture radar SAR image.
Using the template selected by each pixel in synthetic aperture radar SAR image, synthetic aperture radar SAR image is obtained
Gradient map.
According to the following formula, the sideline response diagram for normalizing to [0,1] is merged with normalizing to the gradient map of [0,1],
Obtain intensity map：
Wherein, I represents the intensity level in intensity map, and x represents the value in the response diagram of sideline, and y represents the value in gradient map.
Using non-maxima suppression method, intensity map is detected, obtains suggestion sketch.
The pixel for suggesting that there is maximum intensity in sketch is chosen, will suggest what is connected in sketch with the pixel of the maximum intensity
Pixel connects to form suggestion line segment, obtains suggestion sketch map.
According to the following formula, the code length gain CLG for suggesting sketch line in sketch map is calculated：
Wherein, CLG represents to suggest the code length gain of sketch line in sketch map, and m represents picture in current sketch line neighborhood
The number of element, t represent the numbering of pixel in current sketch line neighborhood, AtRepresent the sight of t-th of pixel in current sketch line neighborhood
Measured value, At,0Represent in the case where current sketch line can not represent the hypothesis of structural information, t-th pixel estimates in the sketch line neighborhood
Evaluation, ln () represent the log operations using e the bottom of as, At,1Represent in the case where current sketch line can represent the hypothesis of structural information,
The estimate of t-th of pixel in the sketch line neighborhood.
Given threshold T, T span are 5~50, select CLG>T suggestion sketch line is as in final sketch map
Sketch line, obtain sketch map corresponding to input synthetic aperture radar SAR image.
Step 2, it is different semantic regions to divide synthetic aperture radar SAR image.
According to the concentration class of sketch line segment in the sketch map of synthetic aperture radar SAR image, sketch line is divided into expression
Assemble the aggregation sketch line of atural object and represent the sketch line of border, line target and isolated target.
According to the statistics with histogram of sketch line segment concentration class, the sketch line segment conduct that concentration class is equal to optimal concentration class is chosen
Seed line-segment sets { Ek, k=1,2 ..., m }, wherein EkAny bar sketch line segment in seed line-segment sets is represented, k is seed line segment
The label of any bar sketch line segment is concentrated, m is the total number of line segment, and { } represents set operation.
If seed line-segment sets { Ek, k=1,2 ..., m in sketch line segment EkSome line segment aggregate is not added to,
Then with sketch line segment EkNew line segment aggregate is solved for basic point is recursive.
Actionradius is the circular primitive in the optimal concentration class section upper bound, and first the line segment in line segment aggregate is expanded,
Line segment aggregate ecto-entad after expansion is corroded, the aggregation zone in units of sketch point is obtained in sketch map.
The length of each sketch line in the sketch line for representing border, line target and isolated target is calculated, according to length
These sketch lines are ranked up from long to short, the sketch line set after being sorted.
The initial value of counter is arranged to 1, counter α threshold value E value is arranged to 21.
Judge whether counter is less than threshold value, if it is, performing the 8th step, otherwise, perform the 13rd step.
3 seed points are selected on α root sketch lines after sequence, sketch line is divided into 4 bisectors with this 3 seed points
Section, if some seed point is precisely the end points of sketch line, the midpoint of sketch line segment where the seed point is moved to.
In α root sketch lines both sides, using each seed point on α root sketch lines as starting point, along with sketch line it
Between angle be that 180 directions of 1~180 degree stretch out, to running into expression border, line target and isolated mesh during extension
Target sketch line, the border of aggregation zone are any in three kinds of the border situation for the closed area that other sketch line completions obtain
One kind, stop extension, obtain the ray using seed point as starting point, each seed point respectively produces 180 articles in α root sketch lines both sides
Ray, these caused rays are sorted according to the size at ray and sketch wire clamp angle.
The length of all rays is calculated, first ray cluster is generated with first ray, for remaining 179 rays,
The length ratio of the length of each of which ray previous bar ray adjacent thereto is judged whether between 1.25~1.5, if
It is the ray cluster added the ray where its previous bar ray, otherwise, a new ray cluster is generated with the ray.
Ray comprising ray cluster of the number of rays less than 5 and length mutation is modified.
The end points of current sketch line and the ray terminal with its arest neighbors on locus are connected, obtains penetrating for seed point
Line sealing ring.
Merge the ray sealing ring of 3 seed points, obtain the ray sealing ring of sketch line, and obtain using the ray sealing ring
To the sketch line of completion, the closed area of current sketch line is obtained, counter α value is increased by 1, performs the 7th step.
To representing border, line target and the sketch line of isolated target and the sketch line of its completion, with each sketch line
Each sketch point centered on construct size be 5 × 5 geometry window obtain structural region.
Using the part removed in sketch map beyond aggregation zone and structural region as can not sketch region.
By the aggregation zone in sketch map, structural region and can not sketch region correspond to synthetic aperture radar SAR image
On, obtain the aggregation zone, structural region and homogenous region of synthetic aperture radar SAR image.
Step 3, ridge ripple deconvolution network RDN aggregation zone and homogenous region is respectively trained.
1st step, construct 4 layers of ridge ripple deconvolution network RDN.
Input layer is arranged to by the 1st layer of ridge ripple deconvolution network RDN；By ridge ripple deconvolution network RDN the 2nd layer of setting
For warp lamination, warp lamination includes the wave filter of 97 × 7 sizes and the characteristic pattern of 9 37 × 37 sizes, 97 × 7 big
Small wave filter group is into a wave filter group；Warp lamination, warp lamination are arranged to by the 3rd layer of ridge ripple deconvolution network RDN
The characteristic pattern of wave filter and 45 43 × 43 sizes comprising 45 7 × 7 sizes, the wave filter groups of 45 7 × 7 sizes is into one
Individual wave filter group；Warp lamination is arranged to by the 4th layer of ridge ripple deconvolution network RDN, warp lamination includes 100 7 × 7 sizes
Wave filter and 100 49 × 49 sizes characteristic pattern, the wave filter groups of 100 7 × 7 sizes is into a wave filter group.
2nd step, using ridge ripple function, respectively to 3 warp laminations in 4 layers of ridge ripple deconvolution network RDN having constructed
Wave filter group initialized.
Discretization is carried out to the parameter of continuous ridge ripple function, obtains the discretization parameter of ridge ripple function, the continuous ridge ripple
Functional expression is as follows：
Wherein, A represents continuous ridge ripple function, and a represents the scale parameter of continuous ridge ripple function, and a span is a ∈
(0,3], discretization represents to belong to symbol at intervals of 1, ∈, and ψ () represents wavelet function, x1And x2The filter of warp lamination is represented respectively
The abscissa and ordinate of pixel in wave filter in ripple device group, θ represent the directioin parameter of continuous ridge ripple function, θ value
Scope be θ ∈ [0, π), discretization is at intervals of π/18, and b represents the displacement parameter of continuous ridge ripple function, when directioin parameter θ is in θ ∈
[0, pi/2) in the range of value when, b span is b ∈ [0, n × (sin θ+cos θ)], when directioin parameter θ θ ∈ [pi/2,
In the range of π) during value, b span is b ∈ [n × cos θ, n × sin θ], and n represents displacement parameter b threshold parameter, n's
Span be n ∈ (0,1], sin represent SIN function, cos represent cosine function, b discretization is at intervals of 1.
Appoint the ridge ripple function parameter value for taking 9 groups of discretizations, using the continuous ridge ripple functional expression in the 1st step, to ridge ripple warp
The 2nd layer of product network RDN wave filter group is initialized.
Appoint the ridge ripple function parameter value for taking 45 groups of discretizations, using the continuous ridge ripple functional expression in the 1st step, to ridge ripple warp
The 3rd layer of product network RDN wave filter group is initialized.
Appoint the ridge ripple function parameter value for taking 100 groups of discretizations, it is anti-to ridge ripple using the continuous ridge ripple functional expression in the 1st step
Convolutional network RDN the 4th layer of wave filter group is initialized.
3rd step, to spatially disconnected each aggregation zone and spatially disconnected each homogenous region is instructed respectively
Practice 4 layers of ridge ripple deconvolution network RDN, obtain the optimal value of ridge ripple deconvolution network RDN median filter groups.
The method of described training deconvolution network, meeting was published in referring to Matthew D.Zeiler et al. in 2010
Article on Computer Vision and Pattern Recognition《Deconvolutional Networks》, this
It is a kind of method of unsupervised level extraction characteristics of image.
Intensive sliding window sampling is carried out to aggregation zone and homogenous region respectively, sampling window size is respectively 31 × 31 pixels
With 17 × 17 pixels, aggregation zone and the sample of homogenous region sampling are obtained.
Respectively by aggregation zone and the sample of homogenous region, it is input in 4 layers of ridge ripple deconvolution network RDN.
The value of characteristic pattern and wave filter group in fixed ridge ripple deconvolution network RDN, is asked by solving an one-dimensional optimization
Topic, obtains the optimal value of auxiliary variable in ridge ripple deconvolution network.
The value of auxiliary variable and wave filter group in fixed ridge ripple deconvolution network RDN, by solving a linear system most
Optimization problem, obtain the optimal value of characteristic pattern in ridge ripple deconvolution network RDN.
The value of characteristic pattern and auxiliary variable in fixed ridge ripple deconvolution network RDN, by using gradient descent method, obtains ridge
The optimal value of ripple deconvolution network RDN median filter groups.
Step 4, similar aggregation zone is merged.
1st step, the wave filter group of last layers of ridge ripple deconvolution network RDN is optimal obtained by the training of each aggregation zone
Value represents each aggregation zone.
2nd step, with the method for sparse classification, obtain representing architectural feature similarity measure between each aggregation zone.
Appoint and take an aggregation zone to be set to project aggregation zone, appoint and take an aggregation zone different from projection aggregation zone
It is set to be projected aggregation zone.
The wave filter group for representing last layers of the ridge ripple deconvolution network RDN of projection aggregation zone is set to include 100
The projection aggregation zone wave filter group of wave filter, the filter of last layers of the ridge ripple deconvolution network RDN of aggregation zone will be projected
Ripple device group is set to be projected aggregation zone wave filter group comprising 100 wave filters.
By any one wave filter projected in aggregation zone wave filter group to being projected in aggregation zone wave filter group
All wave filters are projected, and obtain 100 groups of projection values, and projection formula is：
Wherein, d represents that projection aggregation zone wave filter projects to the projection value for being projected aggregation zone wave filter, and d's takes
Value scope is d ∈ [0,1], and ∈ represents to belong to symbol, F1Represent projection aggregation zone wave filter, F2Expression is projected aggregation zone
Wave filter, * represent dot product operations, | | | | represent modulus operation.
3rd step, aggregation zone architectural feature similarity measure is more than the corresponding region of its threshold tau as similar aggregation
Region, merge all similar aggregation zones, wherein, τ represents the threshold value of aggregation zone architectural feature similarity measure, and τ's takes
Value scope is τ ∈ [0,1].
Step 5, similar homogenous region is merged.
1st step, the optimal value table of the wave filter group of last layers of gained ridge ripple deconvolution network RDN is trained with regional
Show each homogenous region.
2nd step, with the method for sparse classification, obtain representing architectural feature similarity measure between each homogenous region.
Appoint and take a homogenous region to be set to project homogenous region, appoint and take a homogenous region different from projection homogenous region
It is set to be projected homogenous region.
The wave filter group for representing last layers of the ridge ripple deconvolution network RDN of projection homogenous region is set to include 100
The projection homogenous region wave filter group of wave filter, the filter of last layers of the ridge ripple deconvolution network RDN of homogenous region will be projected
Ripple device group is set to be projected homogenous region wave filter group comprising 100 wave filters.
By any one wave filter projected in the wave filter group of homogenous region to being projected in the wave filter group of homogenous region
All wave filters are projected, and obtain 100 groups of projection values, and projection formula is：
Wherein, d represents that projection homogenous region wave filter projects to the projection value for being projected homogenous region wave filter, and d's takes
Value scope is d ∈ [0,1], and ∈ represents to belong to symbol, F1Represent projection homogenous region wave filter, F2Expression is projected homogenous region
Wave filter, * represent dot product operations, | | | | represent modulus operation.
3rd step, using corresponding region of the homogenous region architectural feature similarity measure more than its threshold value σ as similar homogeneous
Region, merge all similar homogenous regions, wherein, σ represents the threshold value of homogenous region architectural feature similarity measure, and σ's takes
Value scope is σ ∈ [0,1].
Step 6, structural region is split.
Using watershed algorithm, structural region is divided into super-pixel.
In the sketch map of synthetic aperture radar SAR image, two sketch lines that parallel and distance is less than to 7 pixels are true
It is set to first kind line target sketch line, the super-pixel between first kind line target sketch line is merged, as first kind line
Target.
In the initial sketch map of synthetic aperture radar SAR image, sketch line both sides are belonged to the sketch line of the same area
It is defined as the second class line target sketch line, the second class line target sketch line both sides is respectively expanded to a pixel as the second class line mesh
Mark, using other sketch lines as the sketch line for portraying border.
To each super-pixel in addition to the super-pixel that line target and border are covered, by adjacent thereto and gray average
Difference merged less than 25 super-pixel, until two super-pixel of the difference in the absence of adjacent and gray average less than 25 are
Only.
By each super-pixel after merging in the 4th step, be incorporated into and the difference of the super-pixel gray value average it is minimum and
In structural region less than 25, the result after splitting to structural region is obtained.
Step 7, the synthetic aperture radar SAR image after being split.
The homogenous region that the aggregation zone and step 5 obtained using step 4 is obtained, and the structural region that step 6 obtains,
Synthetic aperture radar SAR image after being split.
With reference to analogous diagram, the present invention will be further described.
1. simulated conditions：
The hardware condition that the present invention emulates is：Window7, CPU Pentium (R) 4, fundamental frequency 3.0GHZ；Software
Platform is：MatlabR2012a；The present invention emulation used in synthetic aperture radar SAR image be：Ku wave bands resolution ratio is 1 meter
Piperiver figure.
2. emulation content：
The emulation experiment of the present invention is that Piperiver figures are split, and the Piperiver figures as shown in Fig. 2 (a) come
The synthetic aperture radar SAR image that Ku wave bands resolution ratio is 1 meter is come from, extracted region is carried out to Fig. 2 (a), obtained such as Fig. 2 (b)
Shown administrative division map.
Method using the present invention is split to the aggregation zone of Piperiver figures, obtains the aggregation shown in Fig. 2 (c)
The same atural object of region representation of region segmentation result figure, wherein same color, the different atural object of the region representations of different colours.
Method using the present invention is split to the homogenous region of Piperiver figures, obtains point of the homogenous region shown in Fig. 2 (d)
Cut result figure, the wherein same atural object of color identical region representation, the different atural object of the different region representation of color.Using this
The method of invention is split to the structural region of Piperiver figures, and structural region segmentation result is merged into homogenous region
Segmentation result in, obtain the final synthetic aperture radar SAR image segmentation result figure as shown in Fig. 2 (e), wherein color phase
The same same atural object of region representation, the different atural object of the different region representation of color.
3. simulated effect is analyzed：
It can see by the segmentation result of the Piperiver figures shown in above-mentioned Fig. 2 (e), use the inventive method pairing
The precision of segmentation can be improved by carrying out segmentation into aperture radar SAR image, and the region consistency in segmentation result is more preferable.
Claims (8)
1. a kind of SAR image segmentation method based on ridge ripple deconvolution network and sparse classification, comprises the following steps：
(1) sketch is combined to aperture radar SAR image：
To the synthetic aperture radar SAR image sketch of input, the sketch map of synthetic aperture radar SAR image is obtained；
(2) it is different semantic regions to divide synthetic aperture radar SAR image：
Using ray completion administrative division map, synthetic aperture radar SAR image is divided into aggregation zone, homogenous region and structural area
Domain；
(3) ridge ripple deconvolution network RDN aggregation zone and homogenous region is respectively trained：
(3a) constructs 4 layers of ridge ripple deconvolution network RDN；
(3b) utilizes ridge ripple function, respectively to the filtering of 3 warp laminations in 4 layers of ridge ripple deconvolution network RDN having constructed
Device group is initialized；
(3c) disconnected each homogenous region to spatially disconnected each aggregation zone and spatially, is respectively trained one
4 layers of ridge ripple deconvolution network RDN, obtain the optimal value of ridge ripple deconvolution network RDN median filter groups；
(4) similar aggregation zone is merged：
(4a) trains the wave filter of last layers of gained ridge ripple deconvolution network RDN with spatially disconnected each aggregation zone
Disconnected each aggregation zone on the optimal value representation space of group；
The method that (4b) uses sparse classification, calculating represent architectural feature similitude between spatially disconnected each aggregation zone
Estimate；
Aggregation zone architectural feature similarity measure is more than the corresponding region of its threshold tau as similar aggregation zone by (4c),
Merge all similar aggregation zones, wherein, τ represents the threshold value of aggregation zone architectural feature similarity measure, τ span
For τ ∈ [0,1]；
(5) similar homogenous region is merged：
(5a) trains the wave filter of last layers of gained ridge ripple deconvolution network RDN with spatially disconnected each homogenous region
Disconnected each homogenous region on the optimal value representation space of group；
The method that (5b) uses sparse classification, calculating represent architectural feature similitude between spatially disconnected each homogenous region
Estimate；
(5c) using corresponding region of the homogenous region architectural feature similarity measure more than its threshold value σ as similar homogenous region,
Merge all similar homogenous regions, wherein, σ represents the threshold value of homogenous region architectural feature similarity measure, σ span
For σ ∈ [0,1]；
(6) structural region is split：
The structural region obtained to step (2) is split, and obtains the segmentation result of structural region；
(7) the synthetic aperture radar SAR image after being split：
The homogenous region that the aggregation zone and step (5) obtained using step (4) is obtained, and the structural area that step (6) obtains
Domain, the synthetic aperture radar SAR image after being split.
2. the SAR image segmentation method according to claim 1 based on ridge ripple deconvolution network and sparse classification, its feature
It is：Step (1) sketch comprises the following steps that：
1st step, the side with different directions and yardstick, line template are constructed, and it is each using the direction of template and dimensional information construction
Anisotropy Gaussian function calculates the weight coefficient of every bit in the template, and its mesoscale number value is 3~5, direction number
Value is 18；
2nd step, according to the following formula, the corresponding average of pixel and the side in synthetic aperture radar SAR image of calculation template different zones
Difference：
<mrow>
<mi>&mu;</mi>
<mo>=</mo>
<mfrac>
<mrow>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>g</mi>
<mo>&Element;</mo>
<mi>&Omega;</mi>
</mrow>
</munder>
<msub>
<mi>w</mi>
<mi>g</mi>
</msub>
<msub>
<mi>A</mi>
<mi>g</mi>
</msub>
</mrow>
<mrow>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>g</mi>
<mo>&Element;</mo>
<mi>&Omega;</mi>
</mrow>
</munder>
<msub>
<mi>w</mi>
<mi>g</mi>
</msub>
</mrow>
</mfrac>
</mrow>
<mrow>
<mi>v</mi>
<mo>=</mo>
<mfrac>
<mrow>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>g</mi>
<mo>&Element;</mo>
<mi>&Omega;</mi>
</mrow>
</munder>
<msub>
<mi>w</mi>
<mi>g</mi>
</msub>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>A</mi>
<mi>g</mi>
</msub>
<mo>-</mo>
<mi>&mu;</mi>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>g</mi>
<mo>&Element;</mo>
<mi>&Omega;</mi>
</mrow>
</munder>
<msub>
<mi>w</mi>
<mi>g</mi>
</msub>
</mrow>
</mfrac>
</mrow>
Wherein, μ represents the average of the corresponding pixels in synthetic aperture radar SAR image of region Ω, and Ω represents one in template
Region, g represent the position of a pixel in the Ω of region, and ∈ represents to belong to symbol, and ∑ represents sum operation, wgRepresent region Ω
Weight coefficient at middle position g, wgSpan be wg∈ [0,1], AgRepresent that position g is corresponding in synthetic aperture in the Ω of region
Pixel value in radar SAR image, v represent the variance of the corresponding pixels in synthetic aperture radar SAR image of region Ω；
3rd step, according to the following formula, calculate the response of each pixel comparison value operator in synthetic aperture radar SAR image：
<mrow>
<mi>R</mi>
<mo>=</mo>
<mn>1</mn>
<mo>-</mo>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>{</mo>
<mrow>
<mfrac>
<msub>
<mi>&mu;</mi>
<mi>a</mi>
</msub>
<msub>
<mi>&mu;</mi>
<mi>b</mi>
</msub>
</mfrac>
<mo>,</mo>
<mfrac>
<msub>
<mi>&mu;</mi>
<mi>b</mi>
</msub>
<msub>
<mi>&mu;</mi>
<mi>a</mi>
</msub>
</mfrac>
</mrow>
<mo>}</mo>
</mrow>
</mrow>
Wherein, R represents the response of each pixel comparison value operator in synthetic aperture radar SAR image, and min { } represents to ask most
Small Value Operations, a and b represent the numbering of any two different zones in template, μ respectivelyaAnd μbRespectively represent region a and region b with
The average of respective pixel in synthetic aperture radar SAR image；
4th step, according to the following formula, response of each pixel to correlation operator in calculating synthetic aperture radar SAR image：
<mrow>
<mi>C</mi>
<mo>=</mo>
<msqrt>
<mfrac>
<mn>1</mn>
<mrow>
<mn>1</mn>
<mo>+</mo>
<mn>2</mn>
<mo>&CenterDot;</mo>
<mfrac>
<mrow>
<msubsup>
<mi>v</mi>
<mi>a</mi>
<mn>2</mn>
</msubsup>
<mo>+</mo>
<msubsup>
<mi>v</mi>
<mi>b</mi>
<mn>2</mn>
</msubsup>
</mrow>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>&mu;</mi>
<mi>a</mi>
</msub>
<mo>+</mo>
<msub>
<mi>&mu;</mi>
<mi>b</mi>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mfrac>
</mrow>
</mfrac>
</msqrt>
</mrow>
Wherein, C represents that each pixel is to the response of correlation operator, v in synthetic aperture radar SAR imageaAnd vbRepresent respectively
The variance of the corresponding pixels in synthetic aperture radar SAR image of region a and region b,Represent square root functions；
5th step, according to the following formula, merge the response and synthetic aperture of pixel comparison value operator in synthetic aperture radar SAR image
Pixel is to the response of correlation operator in radar SAR image, calculates in synthetic aperture radar SAR image each pixel to each
The response of template：
<mrow>
<mi>F</mi>
<mo>=</mo>
<msqrt>
<mfrac>
<mrow>
<msup>
<mi>R</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mi>C</mi>
<mn>2</mn>
</msup>
</mrow>
<mn>2</mn>
</mfrac>
</msqrt>
</mrow>
Wherein, F represents that each pixel is to the response of each template in synthetic aperture radar SAR image；
6th step, template of template of the selection with maximum response as pixel in synthetic aperture radar SAR image, and will most
Big intensity of the response as the pixel, the direction using the direction of the template with maximum response as the pixel, is closed
Into the sideline response diagram and directional diagram of aperture radar SAR image；
7th step, using the template selected by each pixel in synthetic aperture radar SAR image, obtain synthetic aperture radar SAR figures
The gradient map of picture；
8th step, according to the following formula, the sideline response diagram for normalizing to [0,1] and the gradient map for normalizing to [0,1] are melted
Close, obtain intensity map：
<mrow>
<mi>I</mi>
<mo>=</mo>
<mfrac>
<mrow>
<mi>x</mi>
<mi>y</mi>
</mrow>
<mrow>
<mn>1</mn>
<mo>-</mo>
<mi>x</mi>
<mo>-</mo>
<mi>y</mi>
<mo>+</mo>
<mn>2</mn>
<mi>x</mi>
<mi>y</mi>
</mrow>
</mfrac>
</mrow>
Wherein, I represents the intensity level in intensity map, and x represents the value in the response diagram of sideline, and y represents the value in gradient map；
9th step, using non-maxima suppression method, intensity map is detected, obtain suggestion sketch；
10th step, the pixel for suggesting that there is maximum intensity in sketch is chosen, will suggest that the pixel in sketch with the maximum intensity connects
Logical pixel connects to form suggestion line segment, obtains suggestion sketch map；
11st step, according to the following formula, calculate the code length gain CLG for suggesting sketch line in sketch map：
<mrow>
<mi>C</mi>
<mi>L</mi>
<mi>G</mi>
<mo>=</mo>
<munderover>
<mi>&Sigma;</mi>
<mi>t</mi>
<mi>m</mi>
</munderover>
<mo>&lsqb;</mo>
<mfrac>
<msubsup>
<mi>A</mi>
<mi>t</mi>
<mn>2</mn>
</msubsup>
<msubsup>
<mi>A</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mn>0</mn>
</mrow>
<mn>2</mn>
</msubsup>
</mfrac>
<mo>+</mo>
<mi>l</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>A</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mn>0</mn>
</mrow>
<mn>2</mn>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mfrac>
<msubsup>
<mi>A</mi>
<mi>t</mi>
<mn>2</mn>
</msubsup>
<msubsup>
<mi>A</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mn>1</mn>
</mrow>
<mn>2</mn>
</msubsup>
</mfrac>
<mo>-</mo>
<mi>l</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>A</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mn>1</mn>
</mrow>
<mn>2</mn>
</msubsup>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
</mrow>
Wherein, CLG represents to suggest the code length gain of sketch line in sketch map, and m represents pixel in current sketch line neighborhood
Number, t represent the numbering of pixel in current sketch line neighborhood, AtThe observation of t-th of pixel in current sketch line neighborhood is represented,
At,0Represent in the case where current sketch line can not represent the hypothesis of structural information, the estimate of t-th of pixel in the sketch line neighborhood,
Ln () represents the log operations using e the bottom of as, At,1Represent in the case where current sketch line can represent the hypothesis of structural information, the element
Retouch the estimate of t-th of pixel in line neighborhood；
12nd step, given threshold T, T span are 5~50, select CLG>T suggestion sketch line is as in final sketch map
Sketch line, obtain input synthetic aperture radar SAR image corresponding to sketch map.
3. the SAR image segmentation method according to claim 1 based on ridge ripple deconvolution network and sparse classification, its feature
It is：Division synthetic aperture radar SAR image comprising the following steps that for different semantic regions described in step (2)：
1st step, according to the concentration class of sketch line segment in the sketch map of synthetic aperture radar SAR image, sketch line is divided into table
Show the aggregation sketch line of aggregation atural object and represent the sketch line of border, line target and isolated target；
2nd step, according to the statistics with histogram of sketch line segment concentration class, choose the sketch line segment work that concentration class is equal to optimal concentration class
For seed line-segment sets { Ek, k=1,2 ..., m }, wherein EkAny bar sketch line segment in seed line-segment sets is represented, k is kind of a sub-line
The label of Duan Jizhong any bar sketch line segments, m are the total number of line segment, and { } represents set operation；
3rd step, if seed line-segment sets { Ek, k=1,2 ..., m in sketch line segment EkIt is not added to some line-segment sets
Close, then with sketch line segment EkNew line segment aggregate is solved for basic point is recursive；
4th step, actionradius are the circular primitive in the optimal concentration class section upper bound, first the line segment in line segment aggregate are carried out swollen
It is swollen, the line segment aggregate ecto-entad after expansion is corroded, the aggregation zone in units of sketch point is obtained in sketch map；
5th step, the length of each sketch line in the sketch line for representing border, line target and isolated target is calculated, according to length
Degree is ranked up to these sketch lines from long to short, the sketch line set after being sorted；
6th step, the initial value of counter is arranged to 1, counter α threshold value E value is arranged to 21；
7th step, judges whether counter is less than threshold value, if it is, performing the 8th step, otherwise, performs the 13rd step；
8th step, 3 seed points are selected on the α root sketch lines after sequence, sketch line is divided into 4 deciles with this 3 seed points
Line segment, if some seed point is precisely the end points of sketch line, the midpoint of sketch line segment where the seed point is moved to；
9th step, in α root sketch lines both sides, using each seed point on α root sketch lines as starting point, along with sketch line
Between angle be that 180 directions of 1~180 degree stretch out, represent border, line target to being run into during extension and isolate
The sketch line of target, the border of aggregation zone, appointing in three kinds of the border situation for the closed area that other sketch line completions obtain
Meaning is a kind of, stops extension, obtains the ray using seed point as starting point, and each seed point respectively produces 180 in α root sketch lines both sides
Bar ray, these caused rays are sorted according to the size at ray and sketch wire clamp angle；
10th step, the length of all rays is calculated, generate first ray cluster with first ray, penetrated for remaining 179
Line, the length ratio of the length of each of which ray previous bar ray adjacent thereto is judged whether between 1.25~1.5,
If the ray the cluster then ray added where its previous bar ray, otherwise, a new ray cluster is generated with the ray；
11st step, the ray comprising ray cluster of the number of rays less than 5 and length mutation is modified；
12nd step, the end points of current sketch line and the ray terminal with its arest neighbors on locus are connected, obtains seed point
Ray sealing ring；
13rd step, merge the ray sealing ring of 3 seed points, obtain the ray sealing ring of sketch line, and close using the ray
Circle obtains the sketch line of completion, obtains the closed area of current sketch line, and counter α value is increased into 1, performs the 7th step；
14th step, to representing border, line target and the sketch line of isolated target and the sketch line of its completion, with each sketch
The geometry window that size is 5 × 5 is constructed centered on each sketch point of line and obtains structural region；
15th step, using the part removed in sketch map beyond aggregation zone and structural region as can not sketch region；
16th step, by the aggregation zone in sketch map, structural region and can not sketch region correspond to synthetic aperture radar SAR figure
As upper, the aggregation zone, structural region and homogenous region of synthetic aperture radar SAR image are obtained.
4. the SAR image segmentation method according to claim 1 based on ridge ripple deconvolution network and sparse classification, its feature
It is：4 layers of ridge ripple deconvolution network RDN's of construction one described in step (3a) comprises the following steps that：
1st step, input layer is arranged to by the 1st layer of ridge ripple deconvolution network RDN；
2nd step, warp lamination is arranged to by the 2nd layer of ridge ripple deconvolution network RDN, warp lamination includes 97 × 7 sizes
The characteristic pattern of wave filter and 9 37 × 37 sizes, the wave filter groups of 97 × 7 sizes is into a wave filter group；
3rd step, warp lamination is arranged to by the 3rd layer of ridge ripple deconvolution network RDN, warp lamination includes 45 7 × 7 sizes
The characteristic pattern of wave filter and 45 43 × 43 sizes, the wave filter groups of 45 7 × 7 sizes is into a wave filter group；
4th step, warp lamination is arranged to by the 4th layer of ridge ripple deconvolution network RDN, warp lamination includes 100 7 × 7 sizes
Wave filter and 100 49 × 49 sizes characteristic pattern, the wave filter groups of 100 7 × 7 sizes is into a wave filter group.
5. the SAR image segmentation method according to claim 1 based on ridge ripple deconvolution network and sparse classification, its feature
It is：Described in step (3b) with ridge ripple function respectively to 3 warps in 4 layers of ridge ripple deconvolution network RDN having constructed
What the wave filter group of lamination was initialized comprises the following steps that：
1st step, discretization is carried out to the parameter of continuous ridge ripple function, obtains the discretization parameter of ridge ripple function, the continuous ridge
Wave function formula is as follows：
<mrow>
<mi>A</mi>
<mo>=</mo>
<msup>
<mi>a</mi>
<mrow>
<mo>-</mo>
<mn>1</mn>
<mo>/</mo>
<mn>2</mn>
</mrow>
</msup>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<msub>
<mi>x</mi>
<mn>1</mn>
</msub>
<mi>c</mi>
<mi>o</mi>
<mi>s</mi>
<mi>&theta;</mi>
<mo>+</mo>
<msub>
<mi>x</mi>
<mn>2</mn>
</msub>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mi>&theta;</mi>
<mo>-</mo>
<mi>b</mi>
</mrow>
<mi>a</mi>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
Wherein, A represents continuous ridge ripple function, and a represents the scale parameter of continuous ridge ripple function, a span for a ∈ (0,3],
Discretization represents to belong to symbol, ψ () expression wavelet functions, x at intervals of 1, ∈1And x2Warp lamination wave filter group is represented respectively
In wave filter in pixel abscissa and ordinate, θ represents the directioin parameter of continuous ridge ripple function, and θ span is θ
∈ [0, π), for discretization at intervals of π/18, b represents the displacement parameter of continuous ridge ripple function, when directioin parameter θ θ ∈ [0, pi/2)
In the range of value when, b span is b ∈ [0, n × (sin θ+cos θ)], when directioin parameter θ θ ∈ [pi/2, π) in the range of
During value, b span is b ∈ [n × cos θ, n × sin θ], and n represents displacement parameter b threshold parameter, n span
For n ∈ (0,1], sin represent SIN function, cos represent cosine function, b discretization is at intervals of 1；
2nd step, appoint the ridge ripple function parameter value for taking 9 groups of discretizations, using the continuous ridge ripple function in the 1st step, to ridge ripple warp
The 2nd layer of product network RDN wave filter group is initialized；
3rd step, appoint the ridge ripple function parameter value for taking 45 groups of discretizations, using the continuous ridge ripple function in the 1st step, to ridge ripple warp
The 3rd layer of product network RDN wave filter group is initialized；
4th step, appoint the ridge ripple function parameter value for taking 100 groups of discretizations, it is anti-to ridge ripple using the continuous ridge ripple function in the 1st step
Convolutional network RDN the 4th layer of wave filter group is initialized.
6. the SAR image segmentation method according to claim 1 based on ridge ripple deconvolution network and sparse classification, its feature
It is：4 layers of ridge ripple deconvolution network RDN's of training one described in step (3c) comprises the following steps that：
1st step, intensive sliding window sampling is carried out to aggregation zone and homogenous region respectively, sampling window size is respectively 31 × 31 pictures
Element and 17 × 17 pixels, obtain aggregation zone and the sample of homogenous region sampling；
2nd step, respectively by aggregation zone and the sample of homogenous region, it is input in 4 layers of ridge ripple deconvolution network RDN；
3rd step, the value of characteristic pattern and wave filter group in ridge ripple deconvolution network RDN is fixed, pass through and solve an one-dimensional optimization
Problem, obtain the optimal value of auxiliary variable in ridge ripple deconvolution network RDN；
4th step, the value of auxiliary variable and wave filter group in ridge ripple deconvolution network RDN is fixed, pass through and solve a linear system
Optimization problem, obtain the optimal value of characteristic pattern in ridge ripple deconvolution network RDN；
5th step, the value of characteristic pattern and auxiliary variable in ridge ripple deconvolution network RDN is fixed, by using gradient descent method, is obtained
The optimal value of ridge ripple deconvolution network RDN median filter groups.
7. the SAR image segmentation method according to claim 1 based on ridge ripple deconvolution network and sparse classification, its feature
It is：The method of sparse classification described in step (4b) and step (5b) comprises the following steps that：
1st step, appoint and take an aggregation zone or homogenous region is set to project aggregation zone or projects homogenous region, appoint take one with
Project aggregation zone or project the different aggregation zone in homogenous region or homogenous region, be set to be projected aggregation zone or be projected
Homogenous region；
2nd step, the wave filter group of last layers of the ridge ripple deconvolution network RDN of projection aggregation zone will be represented, be set to include 100
The projection aggregation zone wave filter group of individual wave filter, the ridge ripple deconvolution network RDN of aggregation zone last layers will be projected
Wave filter group is set to be projected aggregation zone wave filter group comprising 100 wave filters；Or the ridge that projection homogenous region will be represented
The wave filter group of last layer of ripple deconvolution network is set to the projection homogenous region wave filter group comprising 100 wave filters, will be by
The wave filter group for projecting ridge ripple deconvolution network RDN last layers of homogenous region is set to being projected comprising 100 wave filters
Homogenous region wave filter group；
3rd step, according to projection formula, by any one wave filter projected in aggregation zone wave filter group to being projected accumulation regions
All wave filters in the wave filter group of domain are projected, and obtain 100 groups of projection values；Or according to projection formula, homogeneous area will be projected
Any one wave filter in the wave filter group of domain is projected to all wave filters being projected in the wave filter group of homogenous region, is obtained
To 100 groups of projection values；
Described projection formula is as follows：
<mrow>
<mi>d</mi>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>F</mi>
<mn>1</mn>
</msub>
<mo>*</mo>
<msub>
<mi>F</mi>
<mn>2</mn>
</msub>
</mrow>
<mrow>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>F</mi>
<mn>1</mn>
</msub>
<mo>|</mo>
<mo>|</mo>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>F</mi>
<mn>2</mn>
</msub>
<mo>|</mo>
<mo>|</mo>
</mrow>
</mfrac>
</mrow>
Wherein, d represents that projection aggregation zone wave filter or projection homogenous region wave filter project to and be projected aggregation zone filtering
Device or the projection value for being projected homogenous region wave filter, d span is d ∈ [0,1], and ∈ represents to belong to symbol, F1Represent
Project aggregation zone wave filter or projection homogenous region wave filter, F2Expression is projected aggregation zone wave filter or is projected homogeneous
Region filters, * represent dot product operations, | | | | represent modulus operation；
4th step, the minimum value of 100 groups of each group of projection values is added, then divided by 100, obtain representing two aggregation zones or two
Architectural feature similitude estimates between individual homogenous region.
8. the SAR image segmentation method according to claim 1 based on ridge ripple deconvolution network and sparse classification, its feature
It is：Described in step (6) structural region is split comprise the following steps that：
1st step, using watershed algorithm, structural region is divided into super-pixel；
2nd step, in the sketch map of synthetic aperture radar SAR image, parallel and distance is less than to two sketch lines of 7 pixels
It is defined as first kind line target sketch line, the super-pixel between first kind line target sketch line is merged, as the first kind
Line target；
3rd step, in the initial sketch map of synthetic aperture radar SAR image, sketch line both sides are belonged to the sketch of the same area
Line is defined as the second class line target sketch line, and the second class line target sketch line both sides are respectively expanded to a pixel as the second class line mesh
Mark, using other sketch lines as the sketch line for portraying border；
4th step, it is to each super-pixel in addition to the super-pixel that line target and border are covered, adjacent thereto and gray scale is equal
Super-pixel of the difference of value less than 25 merges, until two super-pixel of the difference in the absence of adjacent and gray average less than 25 are
Only；
5th step, by each super-pixel after merging in the 4th step, it is incorporated into and the difference of the super-pixel gray value average is minimum
And in the structural region less than 25, obtain the result after splitting to structural region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510675676.5A CN105374033B (en) | 2015-10-19 | 2015-10-19 | SAR image segmentation method based on ridge ripple deconvolution network and sparse classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510675676.5A CN105374033B (en) | 2015-10-19 | 2015-10-19 | SAR image segmentation method based on ridge ripple deconvolution network and sparse classification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105374033A CN105374033A (en) | 2016-03-02 |
| CN105374033B true CN105374033B (en) | 2017-11-21 |
Family
ID=55376205
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510675676.5A Active CN105374033B (en) | 2015-10-19 | 2015-10-19 | SAR image segmentation method based on ridge ripple deconvolution network and sparse classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105374033B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529570B (en) * | 2016-10-14 | 2019-06-18 | 西安电子科技大学 | Image classification method based on depth ridge ripple neural network |
| CN106846322B (en) * | 2016-12-30 | 2019-06-21 | 西安电子科技大学 | The SAR image segmentation method learnt based on curve wave filter and convolutional coding structure |
| CN106683102B (en) * | 2016-12-30 | 2019-07-23 | 西安电子科技大学 | SAR image segmentation method based on ridge ripple filter and convolutional coding structure learning model |
| CN106611423B (en) * | 2016-12-30 | 2019-05-21 | 西安电子科技大学 | SAR image segmentation method based on ridge ripple filter and deconvolution structural model |
| CN106952269B (en) * | 2017-02-24 | 2019-09-20 | 北京航空航天大学 | The reversible video foreground object sequence detection dividing method of neighbour and system |
| CN107341813B (en) * | 2017-06-15 | 2019-10-25 | 西安电子科技大学 | SAR image segmentation method based on Structure learning and sketch characteristic inference network |
| CN107292268A (en) * | 2017-06-23 | 2017-10-24 | 西安电子科技大学 | The SAR image semantic segmentation method of quick ridge ripple deconvolution Structure learning model |
| CN108334851B (en) * | 2018-02-12 | 2021-08-03 | 西安电子科技大学 | Rapid polarization SAR image segmentation method based on anisotropic property |
| CN108445466B (en) * | 2018-05-04 | 2021-09-28 | 西安电子科技大学 | Radar adjacent target segmentation method based on rain erosion algorithm |
| CN109034374A (en) * | 2018-07-03 | 2018-12-18 | 苏州中科启慧软件技术有限公司 | The relative depth sequence estimation method of convolutional network is intensively connected to using multi-scale |
| CN109145832A (en) * | 2018-08-27 | 2019-01-04 | 大连理工大学 | Polarimetric SAR image semisupervised classification method based on DSFNN Yu non local decision |
| CN109344837B (en) * | 2018-10-22 | 2022-03-04 | 西安电子科技大学 | SAR image semantic segmentation method based on deep convolutional network and weak supervised learning |
| CN110188682B (en) * | 2019-05-30 | 2021-01-19 | 西安电子科技大学 | Optical remote sensing image target detection method based on geometric structure double-path convolution network |
| CN110443821B (en) * | 2019-07-09 | 2023-05-05 | 成都理工大学 | Image-based water body detection method and device |
| CN115223050B (en) * | 2022-04-28 | 2023-08-18 | 湖北工程学院 | Polarized satellite image analysis method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104318246A (en) * | 2014-10-20 | 2015-01-28 | 西安电子科技大学 | Depth self-adaption ridgelet network based polarimetric SAR (Synthetic Aperture Radar) image classification |
| CN104408458A (en) * | 2014-12-09 | 2015-03-11 | 西安电子科技大学 | Ray completion region graph and characteristic learning-based SAR (synthetic aperture radar) image segmentation method |
| CN104408731A (en) * | 2014-12-09 | 2015-03-11 | 西安电子科技大学 | Region graph and statistic similarity coding-based SAR (synthetic aperture radar) image segmentation method |
-
2015
- 2015-10-19 CN CN201510675676.5A patent/CN105374033B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104318246A (en) * | 2014-10-20 | 2015-01-28 | 西安电子科技大学 | Depth self-adaption ridgelet network based polarimetric SAR (Synthetic Aperture Radar) image classification |
| CN104408458A (en) * | 2014-12-09 | 2015-03-11 | 西安电子科技大学 | Ray completion region graph and characteristic learning-based SAR (synthetic aperture radar) image segmentation method |
| CN104408731A (en) * | 2014-12-09 | 2015-03-11 | 西安电子科技大学 | Region graph and statistic similarity coding-based SAR (synthetic aperture radar) image segmentation method |
Non-Patent Citations (2)
| Title |
|---|
| Topographic distance and watershed lines;Fernand Meyer;《Signal Processing》;19941231;第113-15页 * |
| 基于多层反卷积网络的SAR图像分类;王云艳等;《武汉大学学报·信息科学版》;20151008;第40卷(第10期);第1371-1376页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105374033A (en) | 2016-03-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105374033B (en) | SAR image segmentation method based on ridge ripple deconvolution network and sparse classification | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| CN105608456B (en) | A kind of multi-direction Method for text detection based on full convolutional network | |
| CN105608692B (en) | Polarization SAR image segmentation method based on deconvolution network and sparse classification | |
| CN105427313B (en) | SAR image segmentation method based on deconvolution network and adaptive inference network | |
| CN104966085B (en) | A kind of remote sensing images region of interest area detecting method based on the fusion of more notable features | |
| CN108009509A (en) | Vehicle target detection method | |
| CN106611423B (en) | SAR image segmentation method based on ridge ripple filter and deconvolution structural model | |
| CN107341813B (en) | SAR image segmentation method based on Structure learning and sketch characteristic inference network | |
| CN106611421A (en) | SAR image segmentation method based on feature learning and sketch line constraint | |
| Zheng et al. | Large-scale oil palm tree detection from high-resolution remote sensing images using faster-rcnn | |
| CN106683102B (en) | SAR image segmentation method based on ridge ripple filter and convolutional coding structure learning model | |
| CN106611420A (en) | SAR image segmentation method based on deconvolution network and sketch direction constraint | |
| CN103761526B (en) | Urban area detecting method based on feature position optimization and integration | |
| CN103400151A (en) | Optical remote-sensing image, GIS automatic registration and water body extraction integrated method | |
| CN105869178A (en) | Method for unsupervised segmentation of complex targets from dynamic scene based on multi-scale combination feature convex optimization | |
| CN106846322B (en) | The SAR image segmentation method learnt based on curve wave filter and convolutional coding structure | |
| CN104240256A (en) | Image salient detecting method based on layering sparse modeling | |
| CN103729842A (en) | Fabric defect detection method based on local statistical characteristics and overall significance analysis | |
| CN104573685A (en) | Natural scene text detecting method based on extraction of linear structures | |
| CN108257154A (en) | Polarimetric SAR Image change detecting method based on area information and CNN | |
| CN103440510A (en) | Method for positioning characteristic points in facial image | |
| CN104966285A (en) | Method for detecting saliency regions | |
| CN107403434A (en) | SAR image semantic segmentation method based on two-phase analyzing method | |
| CN108305260A (en) | Detection method, device and the equipment of angle point in a kind of image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |