CN108256425B - A method of harbour container is extracted using Remote Spectra efficient information rate - Google Patents
A method of harbour container is extracted using Remote Spectra efficient information rate Download PDFInfo
- Publication number
- CN108256425B CN108256425B CN201711312966.9A CN201711312966A CN108256425B CN 108256425 B CN108256425 B CN 108256425B CN 201711312966 A CN201711312966 A CN 201711312966A CN 108256425 B CN108256425 B CN 108256425B
- Authority
- CN
- China
- Prior art keywords
- harbour container
- characteristic
- extracting
- information rate
- remote
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of methods for extracting harbour container using Remote Spectra efficient information rate, include the following steps: that being primarily based on textural characteristics carries out target detection, pass through the constant building background model of LBP texture gray scale, prospect probability graph is extracted from background model, and is carried out difference and obtained initial target model；Secondly several regions not overlapped are divided an image into according to different characteristic on initial target model；Furthermore by the spectral signature of existing harbour container and shape feature quantitative description, and quantitative characteristic library is formed after correcting by covariance matrix；Last remote sensing features comparison identification, successively carries out Characteristic Contrast, and successively identify harbour container in the region of division；It carries out carrying out digitization while Fuzzy Selection, keeps marginal information, rapid comparison and identification are carried out by spectral signature and shape feature, self-correction is carried out on the basis of this and improves accuracy rate.
Description
Technical field
The present invention relates to remote sensing recognition technical fields, specially a kind of to extract harbour collection using Remote Spectra efficient information rate
The method of vanning.
Background technique
Harbour is as seaborne important component, under the main trend of economic globalization, increasingly by
The attention of people, and container plays the role of clearly in sea transport.So-called container be can load packaging or
Non-packed cargo is transported, and convenient for carrying out a kind of composition tool of handling with mechanical equipment.In operation, it generally requires
The practical dynamic of harbour container is obtained, and with the development of remote sensing technology, remote sensing images are applied into the inspection in harbour container
It is a kind of current trend in survey.
In existing container remote sensing monitoring technology, main problem is that the characteristics of being not based on container is established accordingly
Remote-sensing monitoring method, it finds that data volume to be treated is huge in actual application, and the extraction for image
Very inaccurate, the efficiency of processing is very low, often will do it many useless processing.In the research to container, in fact may be used
With discovery, container itself is that have whole world unified standard, then during remote sensing monitoring, it should based on above-mentioned spy
Point identifies container, but by retrieval discovery, in current technology scheme, for this point, there are no clear
Method is identified and is extracted to realize efficient container.
And in the identification of the container of application, also it is primarily present following problem:
(1) through edge logos in the identification of container, actual data processing amount is still very big, therefore,
In remote sensing images, container is directly based upon edge and is handled since head is very little, during digitization, calculates
It will be very huge for measuring, and be not easy to subsequent processing；
(2) it is very few to know another characteristic, the information after not making full use of remote sensing images or digitization, for container
Recognition accuracy is not high, and in identification, can't be corrected by self-test, in order to improve accuracy of identification, generally requires
Special correcting algorithm, and this process summarizes, and will greatly improve the calculation amount of data.
Summary of the invention
In order to overcome the shortcomings of that prior art, the present invention provide a kind of utilization Remote Spectra efficient information rate extraction port
The method of mouth container carries out carrying out digitization while Fuzzy Selection, keeps marginal information, passes through spectral signature and shape is special
Sign carries out rapid comparison and identification, and self-correction is carried out on the basis of this and improves accuracy rate, can effectively solve that background technique mentions
Out the problem of.
The technical solution adopted by the present invention to solve the technical problems is: a kind of to be extracted using Remote Spectra efficient information rate
The method of harbour container, includes the following steps:
S100, target detection is carried out based on textural characteristics, by the constant building background model of LBP texture gray scale, from background
Prospect probability graph is extracted in model, and is carried out difference and obtained initial target model；
S200, image segmentation divide an image into several according to different characteristic on initial target model and do not overlap
Region；
S300, setting quantitative characteristic library, by the spectral signature of existing harbour container and shape feature quantitative description, and
Quantitative characteristic library is formed after correcting by covariance matrix；
S400, remote sensing features comparison identification, successively carry out Characteristic Contrast, and successively identify harbour collection in the region of division
Vanning.
As a preferred technical solution of the present invention, in the step s 100, LBP textural characteristics construct the tool of background model
Body algorithm are as follows:
S101, textural characteristics assignment, if radius is the Joint Distribution T=t (g of P pixel on the annular field of Rc,
g0..., gP-1), wherein Joint Distribution T is the textural characteristics of image, gcThe gray value at local domain center, gP(P=0,
1 ..., P-1) respective radius be R annulus on P Along ent gray value；
S102, it is based on the constant texture feature extraction of gray scale, due to gcAnd gPIndependently of each other, then T ≈ t (gc)(g0-gc, g1-
gc..., gP-1-gc), wherein t (gc) be whole image intensity profile；
S103, difference extract the data value of initial target, textural characteristics are carried out with the assignment of specific value, then T ≈ t (s
(g0-gc), s (g1-gc) ..., s (gP-1-gc)),
Wherein s (x) is sign function, is specifically had
S104, data value is modeled, Joint Distribution T is pressed into the pixel sequence on annular field and constitutes 0/1 sequence,
By to s (gP-gc) assign the binomial factor 2P, obtain local binary model
As a preferred technical solution of the present invention, in step s 200, the feature of foundation includes gray scale, spatial texture
And geometric characteristic.
As a preferred technical solution of the present invention, in image segmentation, need to be tracked boundary and vector quantization, and
The tracking and vector quantization specific steps on boundary are as follows:
S201, several vertex (x is set1, y1), (x2, y2) ..., (xn, yn) and distance definition threshold value t；
S202, any two vertex is chosen as the beginning and end for dividing boundary, be denoted as (x1, y1) and (xn, yn), and
It is directly connected to constitute polyline；
S203, other vertex are calculated to the vertical range d between polylinei(i=2,3 ..., n-1), in all di> t's
Point concentrates search max (di) corresponding vertex (xm, ym), original initial starting point and emphasis are connect with this vertex respectively, generated
New polyline；
S204, approaching for next round constantly is carried out to every new polyline according to the calculating process of step S203, until
The distance of all points to corresponding polyline is less than the threshold value t being previously set, that is, terminates.
As a preferred technical solution of the present invention, the spectral signature of harbour container includes mean value and mean square deviation, shape
Shape feature includes area, length-width ratio, length, width, boundary length, shape index, density and symmetry.
As a preferred technical solution of the present invention, in step S300, covariance matrix is according in step S200
Vector quantization, specific algorithm are as follows:
Covariance matrix
Wherein X and Y is the x of the initial target model all pixels, the vector of y-coordinate composition, Var (X), Var (Y) respectively
It is the variance of X and Y respectively, Cov (XY) is X, the covariance between Y.
As a preferred technical solution of the present invention, in quantitative characteristic library, to the spectral information and sky of selecting object
Between information be weighted, as heterogeneity index f, specific algorithm are as follows:
F=ω hcolor+(1-ω)hshape, wherein hcolorIt is spectrum heterogeneity, hshapeIt is special heterogeneity, ω is definition
Spectrum weight (0 < ω < 1).
As a preferred technical solution of the present invention, the spectrum heterogeneity hcolorSpecific algorithm are as follows:
hcolor=∑CωC﹒ σC；
Special heterogeneity hshapeCompactness h including objectcmpactWith the slickness h of objectsmooth, the specific calculation of the two
Method is respectively as follows:
The compactness of objectThe slickness of object
Wherein ωCFor the weight factor of each wave band, σCFor the grey scale variance of each wave band, C is wave band number, and l is pair
The actual boundary of elephant is long, and n is the overall pixel number of object, and b is the most short side of the object boundary rectangle.
As a preferred technical solution of the present invention, in step S400, the specific algorithm of Characteristic Contrast are as follows:
S401, using normalization Euclidean distance criterion calculate feature vector between similarity distance, and according to similitude away from
It is indexed from KD tree is established；
S402, the lookup that match point is carried out using BBF algorithm, i.e., be directed to certain query point using priority query, searched for whole
What is recorded in a KD tree queue is all KD root vertex and tree node, extracts the node of highest priority；
S403, it repeats the above steps until queue is for sky or beyond time restriction, wherein BBF algorithm can by Priority Queues
With terminal inquiry at any time, and returns retrieved the corresponding node of highest priority in real time.
It needs to carry out double based on KD tree index and BBF algorithm Rapid matching as a preferred technical solution of the present invention
To match check, specific algorithm are as follows:
Wherein Zi, Zj, Qk, Qj, QlIt is characteristic matching point pair, Q, Z are respectively that two characteristic matching points divide collection, ql and qk
Not Wei the two characteristic matching points to the element of concentration, α, which is characterized, concentrates a closest ratio with time adjacency.
Compared with prior art, the beneficial effects of the present invention are:
The present invention is based on the Fuzzy Selections that textural characteristics carry out target, and in the treatment process of texture, pass through background
Remote sensing images quickly can be carried out digitization by model and difference, and edge can also be kept to believe during digitization
Breath is convenient for subsequent processing, and is acted on during the treatment by image segmentation, and remote sensing images are divided into several parts,
And then quantitative characteristic library is established, and rapid comparison and identification can be carried out by spectral signature and shape feature, it is enterprising on this basis
Row self-correction improves accuracy rate.
Detailed description of the invention
Fig. 1 is overall structure diagram of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Embodiment:
As shown in Figure 1, the present invention provides a kind of method for extracting harbour container using Remote Spectra efficient information rate,
Include the following steps:
Step S100, target detection is carried out based on textural characteristics, by the constant building background model of LBP texture gray scale, from
Prospect probability graph is extracted in background model, and is carried out difference and obtained initial target model.
The specific algorithm of LBP textural characteristics building background model are as follows:
Step S101, textural characteristics assignment, if radius is the Joint Distribution T=t of P pixel on the annular field of R
(gc, g0..., gP-1), wherein Joint Distribution T is the textural characteristics of image, gcThe gray value at local domain center, gP(P=0,
1 ..., P-1) respective radius be R annulus on P Along ent gray value.
Among the above, different (P, R) is combined, specific LBP operator i.e. specific model are ignorant of, because
This needs to establish the LBP operator based on different (P, R) combination according to the actual situation.In order to keep textural characteristics constant for gray scale
Property, with the gray value g of P Along ent on annular fieldP(P=0,1 ..., P-1) subtracts center gray value gc, then above-mentioned
Joint Distribution T conversion are as follows:
T=t (gc, g0-gc..., gP-1-gc)。
Step S102, it is based on the constant texture feature extraction of gray scale, due to gcAnd gPIndependently of each other, by the joint after conversion point
Cloth T carries out approximate factorization, then T ≈ t (gc)(g0-gc, g1-gc..., gP-1-gc), wherein t (gc) divide for the gray scale of whole image
Cloth.
Due to t (gc) intensity profile of whole image is described, therefore no shadow is distributed for the local grain of image
It rings, carries out difference convenient for subsequent step.
Step S103, difference extracts the data value of initial target, textural characteristics is carried out with the assignment of specific value, then T ≈ t
(s(g0-gc), s (g1-gc) ..., s (gP-1-gc)),
Wherein s (x) is sign function, is specifically had
In above-mentioned steps, it is necessary first to it is clear that, the first result for carrying out difference to Joint Distribution should are as follows:
T=t (g0-gc, g1-gc..., gP-1-gc), due to regardless of how changing in the picture, center pixel and annular field
On the relative size of grey scale pixel value be that will not change, this is the self attributes of image, immutable, therefore, can be used
The sign function of the interpolation of imago element and field pixel describes image to replace specific data.
Step S104, data value is modeled, Joint Distribution T is pressed into the pixel sequence on annular field and constitutes 0/1
Sequence, by s (gP-gc) assign the binomial factor 2P, obtain local binary model
In step S104, essence is exactly that the local space texture structure of pixel is expressed as a unique decimal system
Number, the decimal number i.e. above-mentioned LBPP,RNumber.
And it is further, specific LBP textural characteristics are handled, so that each pixel in image all has uniquely
Corresponding LBP characteristic value has just obtained the LBP textural characteristics of image, in LBP textural characteristics figure, at image border
The influence in LBP textural characteristics hand field is smaller, and the gray value of original pixels is remained for the pixel of image border, convenient for subsequent
Operation.
Step S200, image segmentation divides an image into several mutually not according to different characteristic on initial target model
The region of overlapping.
In above-mentioned steps, the feature of foundation includes gray scale, spatial texture and geometric characteristic.
Therefore, need to comprehensively consider in partitioning algorithm is spectrum and two factors of spatial information, be it is a kind of from lower and
On region merging technique, in the comparison in later period, and need according to identical comparison principle progress Characteristic Contrast.
It should be further noted that in image segmentation, need to be tracked boundary and vector quantization, and the tracking on boundary
With vector quantization specific steps are as follows:
S201, several vertex (x is set1, y1), (x2, y2) ..., (xn, yn) and distance definition threshold value t；
S202, any two vertex is chosen as the beginning and end for dividing boundary, be denoted as (x1, y1) and (xn, yn), and
It is directly connected to constitute polyline；
S203, other vertex are calculated to the vertical range d between polylinei(i=2,3 ..., n-1), in all di> t's
Point concentrates search max (di) corresponding vertex (xm, ym), original initial starting point and emphasis are connect with this vertex respectively, generated
New polyline；
S204, approaching for next round constantly is carried out to every new polyline according to the calculating process of step S203, until
The distance of all points to corresponding polyline is less than the threshold value t being previously set, that is, terminates.
In above-mentioned segmentation, it is generally desirable to determine above-mentioned partitioning parameters, but in remote sensing image, it can root
It homogenizes, and homogenizes according to remote sensing images, need to select to homogenize the factor, in the present embodiment, homogenizing step can be into
Row can also be without this is because, will also be calculated in subsequent comparison by the essential parameter of equal prime factor.
But in order to better understand the technical program, or it is necessary to the selections to equal prime factor to be illustrated: homogeneous
The factor is made of form factor and color factor, the tightness factor and the smoothness factor these two pair factor.Since color factor exists
Belong to most important reference information in information extraction so color factor occupies very big weight to the setting of parameter；In order to avoid
Object shapes it is imperfect have an adverse effect to precision introduce form factor；The effect of the smoothness factor be improve it is convenient smooth
Imaged object；The target whether effect of the tightness factor is to discriminate between compact-sized.
Step S300, quantitative characteristic library is set, the spectral signature of existing harbour container and shape feature quantification are retouched
It states, and forms quantitative characteristic library after correcting by covariance matrix.
Wherein the spectral signature of harbour container includes mean value and mean square deviation, and shape feature includes area, length-width ratio, length
Degree, width, boundary length, shape index, density and symmetry.
As shown in following table, work and specific calculation formula that feature quantitative description asks for progress are carried out:
But during actually calculating, it should be appreciated that harbour container belongs to made Target, in remote sensing images,
Artificial target is generally to have the shape of comparison rule, during carrying out quantitative description, it is only necessary to according to vector
The covariance that each point coordinate forms after change is verified.
Covariance matrix
Wherein X and Y is the x of the initial target model all pixels, the vector of y-coordinate composition, Var (X), Var (Y) respectively
It is the variance of X and Y respectively, Cov (XY) is X, the covariance between Y.
In order to be best understood from this step, need that the specific example enumerated in examples detailed above is combined to be understood.
In quantitative characteristic library, the spectral information and spatial information of selecting object are weighted, as heterogeneity index
F, specific algorithm are as follows:
F=ω hcolor+(1-ω)hshape, wherein hcolorIt is spectrum heterogeneity, hshapeIt is special heterogeneity, ω is definition
Spectrum weight (0 < ω < 1).
The spectrum heterogeneity hcolorSpecific algorithm are as follows:
hcolor=∑CωC﹒ σC；
Special heterogeneity hshapeCompactness h including objectcmpactWith the slickness h of objectsmooth, the specific calculation of the two
Method is respectively as follows:
The compactness of objectThe slickness of object
Wherein ωCFor the weight factor of each wave band, σCFor the grey scale variance of each wave band, C is wave band number, and l is pair
The actual boundary of elephant is long, and n is the overall pixel number of object, and b is the most short side of the object boundary rectangle.
In this step, feature description and comparison should be carried out in conjunction with mentioned-above equal prime factor.
Step S400, remote sensing features comparison identification, successively carries out Characteristic Contrast, and successively identify port in the region of division
Mouth container.
The specific algorithm of Characteristic Contrast are as follows:
S401, using normalization Euclidean distance criterion calculate feature vector between similarity distance, and according to similitude away from
It is indexed from KD tree is established；
S402, the lookup that match point is carried out using BBF algorithm, i.e., be directed to certain query point using priority query, searched for whole
What is recorded in a KD tree queue is all KD root vertex and tree node, extracts the node of highest priority；
S403, it repeats the above steps until queue is for sky or beyond time restriction, wherein BBF algorithm can by Priority Queues
With terminal inquiry at any time, and returns retrieved the corresponding node of highest priority in real time.
Based on KD tree index and BBF algorithm Rapid matching, need to carry out bi-directional matching inspection, specific algorithm are as follows:
Wherein Zi, Zj, Qk, Qj, QlIt is characteristic matching point pair, Q, Z are respectively that two characteristic matching points divide collection, ql and qk
Not Wei the two characteristic matching points to the element of concentration, α, which is characterized, concentrates a closest ratio with time adjacency.
It can be seen that the constraint condition of bi-directional matching algorithm all more than the constraint condition of former algorithm from condition defined above
By force.With the enhancing of constraint condition, on the one hand, in the initial matching stage, more exterior points are detected in the initial matching stage
Come, obtained initial error match point logarithm is reduced, and on the other hand, is rejecting the exterior point stage, these erroneous matchings are to algorithm
It influences to weaken, causes last total match point logarithm and correct match point logarithm to increase, to improve Feature Points Matching
Performance.
In conclusion the main characteristic of the invention lies in that:
(1) the present invention is based on the Fuzzy Selections that textural characteristics carry out target, and in the treatment process of texture, pass through back
Remote sensing images quickly can be carried out digitization by scape model and difference, and edge can also be kept to believe during digitization
Breath is convenient for subsequent processing；
(2) it is acted on during the treatment by image segmentation, remote sensing images is divided into several parts, and then establish and determine
Measure feature library can carry out rapid comparison and identification by spectral signature and shape feature, self-correction is carried out on the basis of this
Improve accuracy rate.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie
In the case where without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims
Variation is included within the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.
Claims (9)
1. a kind of method for extracting harbour container using Remote Spectra efficient information rate, which comprises the steps of:
S100, target detection is carried out based on textural characteristics, by the constant building background model of LBP texture gray scale, from background model
Middle extraction prospect probability graph, and carry out difference and obtain initial target model, in differential process center pixel and neighborhood territory pixel
The sign function of interpolation describes image to replace specific data；
S200, image segmentation determine partitioning parameters first on initial target model, are in subsequent contrast according to remote sensing images
It is no to carry out essential parameter calculating by equal prime factor to choose whether to homogenize by the factor that homogenizes, determining segmentation ginseng
Number divides an image into several areas not overlapped according to different characteristic after determining partitioning parameters and being homogenized
Domain, and this feature includes gray scale, spatial texture and geometry, and is based on partitioning algorithm, spectrum and sky in specific segmentation
Between two factors of information carry out region merging techniques, and according to identical comparison principle progress Characteristic Contrast in comparison；
S300, setting quantitative characteristic library, by the spectral signature of existing harbour container and shape feature quantitative description, in conjunction with equal
The matter factor forms quantitative characteristic library after correcting by covariance matrix to carry out feature description and comparison；
S400, remote sensing features comparison identification, successively carry out Characteristic Contrast, and successively identify harbour packaging in the region of division
Case.
2. a kind of method for extracting harbour container using Remote Spectra efficient information rate according to claim 1, special
Sign is that in the step s 100, LBP textural characteristics construct the specific algorithm of background model are as follows:
S101, textural characteristics assignment, if radius is the Joint Distribution T=t (g of p pixel on the annular neighborhood of Rc, g0...,
gp-1), wherein Joint Distribution T is the textural characteristics of image, gcThe gray value at local neighborhood center, gPRespective radius is the circle of R
The gray value of p Along ent on ring, wherein p=0,1 ..., p-1；
S102, it is based on the constant texture feature extraction of gray scale, due to gcAnd gpIndependently of each other, then T ≈ t (gc)(g0-gc, g1-gc...,
gp-1-gc), wherein t (gc) be whole image intensity profile；
S103, difference extract the data value of initial target, textural characteristics are carried out with the assignment of specific value, then T ≈ t (s (g0-
gc), s (g1-gc) ..., s (gp-1-gc)),
Wherein s (x) is sign function, is specifically had
S104, data value is modeled, Joint Distribution T is pressed into the pixel sequence on annular neighborhood and constitutes 0/1 sequence, is passed through
To s (gp-gc) assign the binomial factor 2p, obtain local binary model
3. a kind of method for extracting harbour container using Remote Spectra efficient information rate according to claim 1, special
Sign is, in image segmentation, needs to be tracked boundary and vector quantization, and the tracking on boundary and vector quantization specific steps are as follows:
S201, several vertex (x is set1, y1), (x2, y2) ..., (xn, yn) and distance definition threshold value t；
S202, any two vertex is chosen as the beginning and end for dividing boundary, be denoted as (x1, y1) and (xn, yn), and it is straight
It connects and connects and composes polyline；
S203, other vertex are calculated to the vertical range d between polylinei, in all diThe point of > t concentrates search max (di) right
Vertex (the x answeredm, ym), original initial starting point and emphasis are connect with this vertex respectively, generate new polyline, wherein i=
2,3 ..., n-1；
S204, approaching for next round constantly is carried out to every new polyline according to the calculating process of step S203, until all
The distance of point to corresponding polyline be less than the threshold value t that is previously set, that is, terminate.
4. a kind of method for extracting harbour container using Remote Spectra efficient information rate according to claim 1, special
Sign is that the spectral signature of harbour container includes mean value and mean square deviation, and shape feature includes area, length-width ratio, length, width
Degree, boundary length, shape index, density and symmetry.
5. a kind of method for extracting harbour container using Remote Spectra efficient information rate according to claim 3, special
Sign is, in step S300, covariance matrix is according to the vector quantization in step S200, specific algorithm are as follows:
Covariance matrix
Wherein X and Y is the x of the initial target model all pixels respectively, and the vector of y-coordinate composition, Var (X), Var (Y) are respectively
It is the variance of X and Y, Cov (XY) is X, the covariance between Y.
6. a kind of method for extracting harbour container using Remote Spectra efficient information rate according to claim 1, special
Sign is, in quantitative characteristic library, is weighted to the spectral information and spatial information of selecting object, as heterogeneity index F,
Its specific algorithm are as follows:
F=ω hcolor+(1-ω)hshape, wherein hcolorIt is spectrum heterogeneity, hshapeIt is special heterogeneity, ω is the light of definition
Weight is composed, wherein 0 < ω < 1.
7. a kind of method for extracting harbour container using Remote Spectra efficient information rate according to claim 6, special
Sign is, the spectrum heterogeneity hcolorSpecific algorithm are as follows:
hcolor=∑CωC﹒ σC；
Special heterogeneity hshapeCompactness h including objectcmpactWith the slickness h of objectsmooth, the specific algorithm point of the two
Not are as follows:
The compactness of objectThe slickness of object
Wherein ωCFor the weight factor of each wave band, σCFor the grey scale variance of each wave band, C is wave band number, and l is object
Actual boundary is long, and n is the overall pixel number of object, and b is the most short side of the object boundary rectangle.
8. a kind of method for extracting harbour container using Remote Spectra efficient information rate according to claim 1, special
Sign is, in step S400, the specific algorithm of Characteristic Contrast are as follows:
S401, the similarity distance between feature vector is calculated using normalization Euclidean distance criterion, and is built according to similarity distance
Vertical KD tree index；
S402, the lookup that match point is carried out using BBF algorithm, i.e., be directed to certain query point using priority query, search for entire KD
What is recorded in tree queue is all KD root vertex and tree node, extracts the node of highest priority；
S403, repeat the above steps until queue be it is empty or beyond time restriction, wherein BBF algorithm can be with by Priority Queues
When terminal inquiry, and return retrieved the corresponding node of highest priority in real time.
9. a kind of method for extracting harbour container using Remote Spectra efficient information rate according to claim 8, special
Sign is, based on KD tree index and BBF algorithm Rapid matching, needs to carry out bi-directional matching inspection, specific algorithm are as follows:
Wherein Zi, Zj, Qk, Qj, QlIt is characteristic matching point pair, Q, Z
Respectively two characteristic matching points are to collection, and ql and qk are respectively element of the two characteristic matching points to concentration, and α is characterized point set
In the closest ratio with time adjacency.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711312966.9A CN108256425B (en) | 2017-12-12 | 2017-12-12 | A method of harbour container is extracted using Remote Spectra efficient information rate |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711312966.9A CN108256425B (en) | 2017-12-12 | 2017-12-12 | A method of harbour container is extracted using Remote Spectra efficient information rate |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108256425A CN108256425A (en) | 2018-07-06 |
| CN108256425B true CN108256425B (en) | 2019-06-25 |
Family
ID=62721081
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711312966.9A Active CN108256425B (en) | 2017-12-12 | 2017-12-12 | A method of harbour container is extracted using Remote Spectra efficient information rate |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108256425B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113688845B (en) * | 2021-08-19 | 2022-12-30 | 南京诺源医疗器械有限公司 | Feature extraction method and device suitable for hyperspectral remote sensing image and storage medium |
| CN114285651A (en) * | 2021-12-27 | 2022-04-05 | 电子科技大学 | DDoS attack detection method under industrial control environment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101710419A (en) * | 2009-10-29 | 2010-05-19 | 中国科学院对地观测与数字地球科学中心 | Automatic intelligent method for detecting insufficiently-segmented regions of high-resolution remote sensing image |
| CN103679675A (en) * | 2013-11-29 | 2014-03-26 | 航天恒星科技有限公司 | Remote sensing image fusion method oriented to water quality quantitative remote sensing application |
| CN104915636A (en) * | 2015-04-15 | 2015-09-16 | 北京工业大学 | Remote sensing image road identification method based on multistage frame significant characteristics |
-
2017
- 2017-12-12 CN CN201711312966.9A patent/CN108256425B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101710419A (en) * | 2009-10-29 | 2010-05-19 | 中国科学院对地观测与数字地球科学中心 | Automatic intelligent method for detecting insufficiently-segmented regions of high-resolution remote sensing image |
| CN103679675A (en) * | 2013-11-29 | 2014-03-26 | 航天恒星科技有限公司 | Remote sensing image fusion method oriented to water quality quantitative remote sensing application |
| CN104915636A (en) * | 2015-04-15 | 2015-09-16 | 北京工业大学 | Remote sensing image road identification method based on multistage frame significant characteristics |
Non-Patent Citations (3)
| Title |
|---|
| 基于改进SIFT算法的图像复制粘贴篡改检测;李昆仑等;《计算机科学》;20160630;第43卷(第6A期);第179-183页 |
| 结合光谱、纹理与形状结构信息的遥感影像分割方法;巫兆聪;《测绘学报》;20130228;第42卷(第1期);第44-50页 |
| 面向对象的高空间分辨率遥感影像信息提取--汶川地震城市震害房屋案例研究;曾涛等;《自然灾害学报》;20101031;第19卷(第5期);第81-87页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108256425A (en) | 2018-07-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Photovoltaic panel extraction from very high-resolution aerial imagery using region–line primitive association analysis and template matching | |
| US8031948B2 (en) | Shape comparison apparatus on contour decomposition and correspondence | |
| CN112084869B (en) | Compact quadrilateral representation-based building target detection method | |
| CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
| CN107330875B (en) | Water body surrounding environment change detection method based on forward and reverse heterogeneity of remote sensing image | |
| CN106503739A (en) | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics | |
| CN105809651B (en) | Image significance detection method based on the comparison of edge non-similarity | |
| US11804025B2 (en) | Methods and systems for identifying topographic features | |
| CN106294705A (en) | A kind of batch remote sensing image preprocess method | |
| Gueguen et al. | Toward a generalizable image representation for large-scale change detection: Application to generic damage analysis | |
| US9595114B2 (en) | Semi automatic target initialization method based on visual saliency | |
| CN102289681A (en) | Method for matching envelope images | |
| Wei et al. | A concentric loop convolutional neural network for manual delineation-level building boundary segmentation from remote-sensing images | |
| CN108256425B (en) | A method of harbour container is extracted using Remote Spectra efficient information rate | |
| Quispe et al. | Automatic building change detection on aerial images using convolutional neural networks and handcrafted features | |
| Mizoguchi et al. | Individual tree species classification based on terrestrial laser scanning using curvature estimation and convolutional neural network | |
| Omidalizarandi et al. | Segmentation and classification of point clouds from dense aerial image matching | |
| Wang et al. | A Crown Guess and Selection Framework for Individual Tree Detection From ALS Point Clouds | |
| Joshi et al. | Automatic rooftop detection using a two-stage classification | |
| Liu et al. | Adaptive algorithm for automated polygonal approximation of high spatial resolution remote sensing imagery segmentation contours | |
| Lukashevich et al. | Building detection on aerial and space images | |
| Sun et al. | Contextual models for automatic building extraction in high resolution remote sensing image using object-based boosting method | |
| Du et al. | Shadow detection in high-resolution remote sensing image based on improved K-means | |
| CN110490069A (en) | A kind of Remote Sensing Target recognition methods based on down-sampled local Differential Binary | |
| CN108257143A (en) | A kind of method based on remote Sensing Interpretation analytical technology extraction container bridge |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |