CN103714344A - Geometrical component and radial gradient angle-based aircraft remote sensing detection method - Google Patents
Geometrical component and radial gradient angle-based aircraft remote sensing detection method Download PDFInfo
- Publication number
- CN103714344A CN103714344A CN201310703176.9A CN201310703176A CN103714344A CN 103714344 A CN103714344 A CN 103714344A CN 201310703176 A CN201310703176 A CN 201310703176A CN 103714344 A CN103714344 A CN 103714344A
- Authority
- CN
- China
- Prior art keywords
- sub
- aircraft
- gradient
- remote sensing
- atomic diagram
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a geometrical component and radial gradient angle-based aircraft remote sensing detection method. The method includes the following steps that: a geometric atomic library is established according to special geometric contours for an aircraft; based on a sparse representation principle, a known aircraft contour is sparsely represented as a combination of geometric atoms; an aircraft star-type component model is constructed with the geometric atoms adopted as components; a sliding window method is adopted to scan a test image; the principal direction of each image block is estimated through a radial-gradient angle voting algorithm, such that the direction of the component model can be adjusted; a total similarity distribution diagram of each position is calculated through using the adjusted model; and whether the aircraft exists in a detected area is judged through maximum posteriori probability reasoning, and the position of the aircraft can be further obtained. According to the geometrical component and radial gradient angle-based aircraft remote sensing detection method, the sparse representation principle is utilized to adaptively select the geometric components, and the principal directions are estimated through the radial-gradient angle voting algorithm which adopts the contours as guidance, and therefore, aircraft remote sensing detection at different directions in a complex background can be realized. The geometrical component and radial gradient angle-based aircraft remote sensing detection method is advantageous in simple algorithm, fast detection speed, better robust and more accurate detection results.
Description
Technical field
The present invention relates to a kind of method that remote sensing detects aircraft.
Background technology
The sixties in 20th century, it is sign that the USSR (Union of Soviet Socialist Republics) of take has been launched its first man made earth satellite in the world, and " remote sensing " starts to enter the mankind's work and life.Remote sensing refers to from the platform away from ground, adopts sensor to survey ground, by collecting, analyzing and process, surveys the signal obtaining, and obtains the relevant information on ground.Along with the develop rapidly of computing machine and sensor devices, remote sensing technology is widely applied, and relates to environmental disaster control, city planning, and a plurality of fields such as resource exploration, produce far-reaching influence to national defense construction and human lives.
Aircraft, as common means of transport, carries out remote sensing to it and detects transport power planning, scheduling controlling and the emergency processing (as searching of wrecked aircraft) etc. that are conducive to improve air transportation.Due to computer technology and sensor devices develop rapidly, the resolution of remotely sensed image can reach panchromatic wave-band 0.61m, and the spatial resolution of colored wave band 2.44m can adopt finer aircraft local feature to describe aircraft.The people such as Zhang are at document 1(Wanceng Zhang, Xian Sun.Object Detection in High-Resolution Remote Sensing Images Using Rotation Invariant Parts Based Model.IEEE Geoscience and Remote Sensing Letters.2013.
http:// ieeexplore.ieee.org/xpls/absall.jsp arnumber=6512596 & tag=1) in, propose the partial model based on invariable rotary histogram of gradients RIHOG, realize the remote sensing images Airplane detection of different directions.The method first people calibrates the parts that training image is concentrated each correcting image; Then to each parts, adopt gradient orientation histogram HOG method statistic to obtain its principal direction, and carry out rotation correction; Finally by the wave filter of SVM method training production part.Each parts that it is concentrated training image need artificial demarcation, lack dirigibility, and demarcate training plan image set and produce huge workload, while increasing modeling, consume; In addition, parts wave filter and principal direction are estimated all to take picture material as guidance, make to detect the easy impact because of color, texture, illumination and complex background and lead to the failure.Summary of the invention
The object of the present invention is to provide a kind of remote sensing based on geometry parts and radial gradient angle to detect the method for aircraft, the method can detect the aircraft of different directions under complex background automatically, and it detects fast, and result is accurate.
The present invention solves its technical matters, and the technical scheme adopting is: the remote sensing based on geometry parts and radial gradient angle detects the method for aircraft, comprises the following steps:
The foundation of A, the star-like partial model of aircraft:
A1, by the known aircraft image rectification in known remote sensing images be head straight up, obtain correcting image;
A2, the pixel value of aircraft part in correcting image is set to 1, background parts pixel value is set to 0, obtains the two-value profile P of known aircraft
b;
A3, the two-value profile P to known aircraft
bcarry out Its Sparse Decomposition, obtain optimum atomic diagram image set
n
optfor optimum atomic diagram is as number, i is optimum atomic diagram picture
index number,
be i optimum atomic diagram picture;
A4, get i optimum atomic diagram picture
in non-zero region obtain i geometric figure
with i geometric figure
as i parts Part
i, with the two-value profile P of known aircraft
bfor root part Root, carry out Star Model association; When associated, i parts Part
ibe shaped as i geometric figure
shape
i parts Part
ito root part Root center (x
r, y
r) relative coordinate position be i geometric figure
center is to atomic diagram picture
the relative coordinate at center
thereby i parts Part
iprincipal direction be i geometric figure
principal direction
wherein, x
r, y
rbe respectively horizontal ordinate and the ordinate of root Root center in correcting image;
A5, repetition A4, by all geometric figures
all as parts Part
icarry out associatedly with root part Root, set up and to obtain the star-like partial model of aircraft;
Airplane detection in B, remote sensing images
B1, remote sensing via satellite obtain the remote sensing images P of surveyed area, get the two-value profile P of size and known aircraft
bidentical detection window carries out point by point scanning in remote sensing images P, obtains subimage block P
sub;
B2, radially-gradient angle Voting Algorithm of employing calculate subimage block P
subprincipal direction ξ
sub;
B3, aircraft Star Model is rotated to heading and subimage block P
subprincipal direction ξ
subunanimously, calculate each parts Part in postrotational aircraft Star Model
iprincipal direction
and with root part Root center (x
r, y
r) relative coordinate
B4, in aircraft Star Model according to i parts Part
ipostrotational principal direction
i parts Part
irotate rear and root part Root center relative coordinate
and the shape of i parts
adopt method of geometry to calculate rear i i the region that parts cover of aircraft Star Model rotation
B5, by i region in aircraft Star Model
interior pixel is set to 1, and other pixel is set to 0, generates i two-value component diagram
that is:
B6, i two-value component diagram of calculating
histograms of oriented gradients, generate i parts wave filter F
i:
In formula, HOG () is histograms of oriented gradients feature extraction operator;
The subimage block P of B6, calculating remote sensing images P
subhistograms of oriented gradients H
sub:
H
sub=HOG(P
sub)
B7, calculating subimage block P
subtotal similarity Score with all parts in postrotational aircraft Star Model:
In formula, < > is inner product operator;
B8, repeat B1~B7, calculate total similarity Score of all detection windows in remote sensing images P, obtain total similarity Score distribution plan S;
C, in total similarity Score distribution plan S, have the total similarity Score that is greater than setting threshold, judge in surveyed area and have aircraft, its corresponding detection window position in remote sensing images is aircraft location.
Compared with prior art, the invention has the beneficial effects as follows:
One, the automatic geometry atomic diagram picture producing of method that the present invention passes through rarefaction representation using the profile of known remote sensing images aircraft is as parts, set up the star-like partial model of aircraft, compare with traditional Star Model, the resulting component representation performance of the inventive method is stronger, not affected by the factors such as color, texture, illumination; Good stability, the testing result of parts are accurate;
Two, the present invention estimates the subimage block principal direction obtaining according to radially-gradient Voting Algorithm, adjusts the direction of the star-like partial model of aircraft, and adopts the parts wave filter after adjusting to detect, thereby realizes the Airplane detection of different directions.
In a word, the inventive method chooses by the method self-adaptation of the profile rarefaction representation of known remote sensing images aircraft how much parts that modeling of aircraft is used, and makes parts have better robustness; By the star-like partial model direction rotation of aircraft, the principal direction to subimage block detects again simultaneously, can realize the Airplane detection of different directions.
In above-mentioned steps A2 to known aircraft two-value profile P
bthe specific practice of carrying out Its Sparse Decomposition is:
1) generate former word bank D how much
A, construct oval former word bank D
e:
Define oval atomic diagram as g
e(x, y) is as follows:
In formula, x, y is respectively that oval atomic diagram is as g
e(x, y) horizontal ordinate and ordinate in correcting image, oval atomic diagram is as g
ethe two-value profile P of the size of (x, y) and known aircraft
bit is identical,
for oval atomic diagram is as g
eelliptic region in (x, y), by following formula, calculated:
In formula, u, v be respectively horizontal ordinate x and ordinate y with oval atomic diagram as g
e(x, y) center is that the point of rotation turns clockwise
after new horizontal ordinate and ordinate, M
pBand N
pBbe respectively the two-value profile P of known aircraft
blength and width; (px, py) is elliptic region
center is to the two-value profile P of known aircraft
bthe relative coordinate at center, px ∈ [M
pB/ 2, M
pB/ 2], py ∈ [N
pB/ 2, N
pB/ 2]; A and b are respectively elliptic region
major axis and minor axis,
for elliptic region
principal direction,
By parameter a, b,
px, py gets the oval atomic diagram that generates all over above all values as g
e(x, y), forms oval former word bank D
e:
B, construct trapezoidal former word bank D
t:
Define trapezoidal atomic diagram as g
t(x, y) is as follows:
In formula, x, y is respectively that trapezoidal atomic diagram is as g
t(x, y) horizontal ordinate and ordinate in correcting image, trapezoidal atomic diagram is as g
tthe two-value profile P of the size of (x, y) and known aircraft
bidentical;
for trapezoidal atomic diagram is as g
ttrapezoid area in (x, y), is surrounded by four straight lines, by following formula, is calculated:
In formula, u, v is respectively that horizontal ordinate x and ordinate y justify atomic diagram as g with ladder
t(x, y) center is that the point of rotation turns clockwise
after horizontal ordinate and ordinate, ub, lb and h are respectively trapezoidal upper base, go to the bottom and high,
f is upper base center and the side-play amount at the center of going to the bottom, regulation upper base center toward left avertence for negative, toward right avertence for just, f ∈ [50,50];
with px, the definition of py and span and oval atomic diagram are as g
e(x, y) is identical;
By parameters u b, lb, h, f,
px, py gets the trapezoidal atomic diagram that generates all over all values as g
t(x, y), forms trapezoidal former word bank D
t:
C, by the former word bank D of all ellipses
ein atomic diagram as g
e(x, y) and trapezoidal former word bank D
tatom image g
t(x, y) merges, and generates the two-value profile P of known aircraft
bthe former word bank D of geometry:
D=D
e∪D
t
D, by all atomic diagram in how much former word bank D as g
e(x, y), g
t(x, y) renames numbering, is expressed as g
j(x, y), j=1,2 ..., N, N be atomic diagram in how much former word bank D as number, j renames atomic diagram after numbering as g
jthe index number of (x, y);
2) set up the optimal model of aircraft profile rarefaction representation:
min||Ω||
0
<g
i,g
j>=0,i≠j
In formula, || ||
0for 0-norm, the element number of expression set, || ||
2for 2-norm, represent that energy, < > are that indicate satisfied condition, σ of inner product operator, s.t. is energy threshold, get the two-value profile P of known aircraft
b15% of area, i.e. σ=15% * Area (P
b), Area () is that area operator, Ω are the subspace in the former word bank D of geometry satisfying condition,
3) adopt quantum genetic algorithm to solve above optimal model, obtain optimum atom image combining
be i optimum atomic diagram picture, N
optfor optimum atomic diagram picture
number, i be optimum atomic diagram picture
index number.
Like this, the parts in model can be chosen adaptively according to the profile feature of different aircrafts, and are applicable to various aircraft, and its applicability is strong, and dirigibility is good; The part count that self-adaptation obtains is less, makes its detection speed faster.
Employing in above-mentioned steps B2 radially-gradient Voting Algorithm calculates subimage block P
subprincipal direction θ
subspecific practice be:
A, extraction subimage block P
subtwo-value profile P with known aircraft
bhistograms of oriented gradients feature H
sub(m, n, γ) and H
b(m, n, γ):
H
sub(m,n,γ)=HOG(P
sub)
H
B(m,n,γ)=HOG(P
B)
In formula, m, n, γ is respectively the horizontal ordinate in histograms of oriented gradients, ordinate and gradient direction;
B, be calculated as follows subimage block P
subtwo-value profile P with known aircraft
bmain gradient direction figure
with
C, by subimage block P
submain gradient direction figure
two-value profile P with known aircraft
bmain gradient direction figure
be converted to polar form, obtain subimage block P
subdirection gradient polar plot PH
subthe two-value profile P of (r, θ) and known aircraft
bdirection gradient polar plot PH
b(r, θ), wherein, r, θ is respectively utmost point footpath coordinate and the polar angle coordinate of direction gradient polar plot;
D, press following formula antithetical phrase image block P
subdirection gradient polar plot PH
sub(r, θ) screens, and obtains candidate direction gradient-pole coordinate diagram C (r, θ):
In formula, inf () is that infimum operator, ε are threshold value, get 0.1-0.2, Φ=θ | PH
b(r, θ) ≠-10} is the two-value profile P of known aircraft
bdirection gradient polar plot PH
bin (r, θ), with subimage block P
subdirection gradient polar plot PH
sub(r, θ) has same pole footpath r, and gradient direction value is not-10 polar angle collection, RGA
sub(r, θ) and RGA
b(r, θ) is respectively subimage block P
subtwo-value profile P with known aircraft
bradially-gradient angle, be calculated as follows:
RGA
sub(r,θ)=|PH
sub(r,θ)-θ|
RGA
B(r,θ)=|PH
B(r,θ)-θ|
E, make gradient direction in candidate direction gradient-pole coordinate diagram C (r, θ) for-10 point is available point, and be calculated as follows out the prediction rotation angle R (k) of available point:
In formula, the index number that k is available point, N
cfor the number of available point,
be k the corresponding polar angle of available point,
two-value profile P for known aircraft
bdirection gradient polar plot PH
bpoint in (r, θ) in radially-gradient angle and candidate direction gradient-pole coordinate diagram
the corresponding polar angle of immediate point, wherein
for candidate direction gradient-pole coordinate diagram mid point
radially-gradient angle;
F, by following formula to the statistics of voting of the rotation angle in the prediction rotation angle R (k) of available point, obtain the histogram His of rotation angle
r:
His
R(λ)=#{k|R(k)=λ}
In formula, # () represents the number of set.
His
rthe direction of middle votes maximum is subimage block P
subprincipal direction ξ
sub, that is:
By above-mentioned voting process, the calculating of principal direction is not subject to the impact of the factors such as color, texture, illumination, can under complex background, obtain precise and stable principal direction; Its computing method speed is fast, than the computation complexity of the method for exhaustion, greatly reduces.
Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.
Accompanying drawing explanation
Fig. 1 a is the correcting image of embodiment emulation experiment 1-3; Fig. 1 b is the two-value profile P of the known aircraft of embodiment emulation experiment 1-3
b; Fig. 1 c is the optimum atomic diagram image set Ω of embodiment emulation experiment 1-3
opt(number in the figure is 1,2 ..., 5 image block is respectively atomic diagram picture
Fig. 2 is the star-like partial model of aircraft of embodiment of the present invention emulation experiment 1-3, wherein Part
i(i=1,2,3,4,5) represent the two-value profile P of known aircraft
b5 how much parts, F
i(i=1,2,3,4,5) represent corresponding parts wave filter.
Fig. 3 a is the remote sensing images P of the surveyed area of emulation experiment 1; Fig. 3 b is total similarity Score distribution plan S that emulation experiment 1 adopts document 1 algorithm to obtain; Fig. 3 c is total similarity Score distribution plan S that emulation experiment 1 adopts the inventive method to obtain.
Fig. 4 a is the remote sensing images P of the surveyed area of emulation experiment 2; Fig. 4 b is total similarity Score distribution plan S that emulation experiment 2 adopts document 1 algorithm to obtain; Fig. 4 c is total similarity Score distribution plan S that emulation experiment 2 adopts the inventive method to obtain.
Fig. 5 a is the remote sensing images P of the surveyed area of emulation experiment 3; Fig. 5 b is total similarity Score distribution plan S that emulation experiment 3 adopts document 1 algorithm to obtain; Fig. 5 c is total similarity Score distribution plan S that emulation experiment 3 adopts the inventive method to obtain.
Embodiment
A kind of embodiment of the present invention is that the remote sensing based on geometry parts and radial gradient angle detects the method for aircraft, comprises the following steps:
The foundation of A, the star-like partial model of aircraft:
A1, by the known aircraft image rectification in known remote sensing images be head straight up, obtain correcting image;
A2, the pixel value of aircraft part in correcting image is set to 1, background parts pixel value is set to 0, obtains the two-value profile P of known aircraft
b;
A3, the two-value profile P to known aircraft
bcarry out Its Sparse Decomposition, obtain optimum atomic diagram image set
n
optfor optimum atomic diagram is as number, i is index number,
be i optimum atomic diagram picture; Concrete steps are as follows:
The described remote sensing based on geometry parts and radial gradient angle detects the method for aircraft, it is characterized in that the two-value profile P to known aircraft in described steps A 2
bthe specific practice of carrying out Its Sparse Decomposition is:
1) generate former word bank D how much
A, construct oval former word bank D
e:
Define oval atomic diagram as g
e(x, y) is as follows:
In formula, x, y is respectively that oval atomic diagram is as g
e(x, y) horizontal ordinate and ordinate in correcting image, oval atomic diagram is as g
ethe two-value profile P of the size of (x, y) and known aircraft
bit is identical,
for oval atomic diagram is as g
eelliptic region in (x, y), by following formula, calculated:
In formula, u, v be respectively horizontal ordinate x and ordinate y with oval atomic diagram as g
e(x, y) center is that the point of rotation turns clockwise
after new horizontal ordinate and ordinate, M
pBand N
pBbe respectively the two-value profile P of known aircraft
blength and width; (px, py) is elliptic region
center is to the two-value profile P of known aircraft
bthe relative coordinate at center, px ∈ [M
pB/ 2, M
pB/ 2], py ∈ [N
pB/ 2, N
pB/ 2]; A and b are respectively elliptic region
major axis and minor axis,
for elliptic region
principal direction,
By parameter a, b,
px, py gets the oval atomic diagram that generates all over above all values as g
e(x, y), forms oval former word bank D
e:
B, construct trapezoidal former word bank D
t:
Define trapezoidal atomic diagram as g
t(x, y) is as follows:
In formula, x, y is respectively that trapezoidal atomic diagram is as g
t(x, y) horizontal ordinate and ordinate in correcting image, trapezoidal atomic diagram is as g
tthe two-value profile P of the size of (x, y) and known aircraft
bidentical;
for trapezoidal atomic diagram is as g
ttrapezoid area in (x, y), is surrounded by four straight lines, by following formula, is calculated:
In formula, u, v is respectively that horizontal ordinate x and ordinate y justify atomic diagram as g with ladder
t(x, y) center is that the point of rotation turns clockwise
after horizontal ordinate and ordinate, ub, lb and h are respectively trapezoidal upper base, go to the bottom and high,
f is upper base center and the side-play amount at the center of going to the bottom, regulation upper base center toward left avertence for negative, toward right avertence for just, f ∈ [50,50];
with px, the definition of py and span and oval atomic diagram are as g
e(x, y) is identical;
By parameters u b, lb, h, f,
px, py gets the trapezoidal atomic diagram that generates all over all values as g
t(x, y), forms trapezoidal former word bank D
t:
C, by the former word bank D of all ellipses
ein atomic diagram as g
e(x, y) and trapezoidal former word bank D
tatom image g
t(x, y) merges, and generates the two-value profile P of known aircraft
bthe former word bank D of geometry:
D=D
e∪D
t
D, by all atomic diagram in how much former word bank D as g
e(x, y), g
t(x, y) renames numbering, is expressed as g
j(x, y), j=1,2 ..., N, N be atomic diagram in how much former word bank D as number, j renames atomic diagram after numbering as g
jthe index number of (x, y);
2) set up the optimal model of aircraft profile rarefaction representation:
min||Ω||
0
<g
i,g
j>=0,i≠j
In formula, || ||
0for 0-norm, the element number of expression set, || ||
2for 2-norm, represent that energy, < > are that indicate satisfied condition, σ of inner product operator, s.t. is energy threshold, get the two-value profile P of known aircraft
b15% of area, i.e. σ=15% * Area (P
b), Area () is that area operator, Ω are the subspace in the former word bank D of geometry satisfying condition,
3) adopt quantum genetic algorithm to solve above optimal model, obtain optimum atom image combining
be i optimum atomic diagram picture, N
optfor optimum atomic diagram picture
number, i be optimum atomic diagram picture
index number.
A4, get i optimum atomic diagram picture
in non-zero region obtain i geometric figure
with i geometric figure
as i parts Part
i, with the two-value profile P of known aircraft
bfor root part Root, carry out Star Model association; When associated, i parts Part
ibe shaped as i geometric figure
shape
i parts Part
ito root part Root center (x
r, y
r) relative coordinate position be i geometric figure
center is to atomic diagram picture
the relative coordinate at center
thereby i parts Part
iprincipal direction be i geometric figure
principal direction
wherein, x
r, y
rbe respectively horizontal ordinate and the ordinate of root Root center in correcting image;
A5, repetition A4, by all geometric figures
all as parts Part
icarry out associatedly with root part Root, set up and to obtain the star-like partial model of aircraft;
Airplane detection in B, remote sensing images
B1, remote sensing via satellite obtain the remote sensing images P of surveyed area, get the two-value profile P of size and known aircraft
bidentical detection window carries out point by point scanning in remote sensing images P, obtains subimage block P
sub;
B2, radially-gradient angle Voting Algorithm of employing calculate subimage block P
subprincipal direction ξ
sub; Concrete steps are as follows:
A, extraction subimage block P
subtwo-value profile P with known aircraft
bhistograms of oriented gradients feature H
sub(m, n, γ) and H
b(m, n, γ):
H
sub(m,n,γ)=HOG(P
sub)
H
B(m,n,γ)=HOG(P
B)
In formula, m, n, γ is respectively the horizontal ordinate in histograms of oriented gradients, ordinate and gradient direction;
B, be calculated as follows subimage block P
subtwo-value profile P with known aircraft
bmain gradient direction figure
with
C, by subimage block P
submain gradient direction figure
two-value profile P with known aircraft
bmain gradient direction figure
be converted to polar form, obtain subimage block P
subdirection gradient polar plot PH
subthe two-value profile P of (r, θ) and known aircraft
bdirection gradient polar plot PH
b(r, θ), wherein, r, θ is respectively utmost point footpath coordinate and the polar angle coordinate of direction gradient polar plot;
D, press following formula antithetical phrase image block P
subdirection gradient polar plot PH
sub(r, θ) screens, and obtains candidate direction gradient-pole coordinate diagram C (r, θ):
In formula, inf () is that infimum operator, ε are threshold value, get 0.1-0.2, Φ=θ | PH
b(r, θ) ≠-10} is the two-value profile P of known aircraft
bdirection gradient polar plot PH
bin (r, θ), with subimage block P
subdirection gradient polar plot PH
sub(r, θ) has same pole footpath r, and gradient direction value is not-10 polar angle collection, RGA
sub(r, θ) and RGA
b(r, θ) is respectively subimage block P
subtwo-value profile P with known aircraft
bradially-gradient angle, be calculated as follows:
RGA
sub(r,θ)=|PH
sub(r,θ)-θ|
RGA
B(r,θ)=|PH
B(r,θ)-θ|
E, make gradient direction in candidate direction gradient-pole coordinate diagram C (r, θ) for-10 point is available point, and be calculated as follows out the prediction rotation angle R (k) of available point:
In formula, the index number that k is available point, N
cfor the number of available point,
be k the corresponding polar angle of available point,
two-value profile P for known aircraft
bdirection gradient polar plot PH
bpoint in (r, θ) in radially-gradient angle and candidate direction gradient-pole coordinate diagram
the corresponding polar angle of immediate point, wherein
for candidate direction gradient-pole coordinate diagram mid point
radially-gradient angle;
F, by following formula to the statistics of voting of the rotation angle in the prediction rotation angle R (k) of available point, obtain the histogram His of rotation angle
r:
His
R(λ)=#{k|R(k)=λ}
In formula, # () represents the number of set.
His
rthe direction of middle votes maximum is subimage block P
subprincipal direction ξ
sub, that is:
B3, aircraft Star Model is rotated to heading and subimage block P
subprincipal direction ξ
subunanimously, calculate each parts Part in postrotational aircraft Star Model
iprincipal direction
and with root part Root center (x
r, y
r) relative coordinate
B4, in aircraft Star Model according to i parts Part
ipostrotational principal direction
i parts Part
irotate rear and root part Root center relative coordinate
and the shape of i parts
adopt method of geometry to calculate rear i i the region that parts cover of aircraft Star Model rotation
B5, by i region in aircraft Star Model
interior pixel is set to 1, and other pixel is set to 0, generates i two-value component diagram
that is:
B6, i two-value component diagram of calculating
histograms of oriented gradients, generate i parts wave filter F
i:
In formula, HOG () is histograms of oriented gradients feature extraction operator;
The subimage block P of B6, calculating remote sensing images P
subhistograms of oriented gradients H
sub:
H
sub=HOG(P
sub)
B7, calculating subimage block P
subtotal similarity Score with all parts in postrotational aircraft Star Model:
In formula, < > is inner product operator;
B8, repeat B1~B7, calculate total similarity Score of all detection windows in remote sensing images P, obtain total similarity Score distribution plan S;
C, in total similarity Score distribution plan S, have the total similarity Score that is greater than setting threshold, judge in surveyed area and have aircraft, its corresponding detection window position in remote sensing images is aircraft location.
Simulation result
Below by emulation experiment, the inventive method and existing document 1 detection method are verified:
Fig. 1 a is the correcting image of emulation experiment 1-3; Fig. 1 b is the two-value profile P of the known aircraft that in emulation experiment 1-3, embodiment method obtains
b; Fig. 1 c is the optimum atomic diagram image set Ω that in emulation experiment 1-3, embodiment method obtains
opt(number in the figure is 1,2 ..., 5 image block is respectively atomic diagram picture
Fig. 2 is the star-like partial model of aircraft that in emulation experiment 1-3, embodiment obtains, wherein Part
i(i=1,2,3,4,5) represent the two-value profile P of known aircraft
b5 how much parts, F
i(i=1,2,3,4,5) represent corresponding parts wave filter.
Fig. 3 a is the remote sensing images P of the surveyed area of emulation experiment 1; Fig. 3 b is total similarity Score distribution plan S that the remote sensing images of Fig. 3 a adopt document 1 algorithm to obtain; Fig. 3 c is total similarity Score distribution plan S that the remote sensing images of Fig. 3 a adopt embodiment of the present invention method to obtain; Aircraft center in Fig. 3 b and Fig. 3 c in the remote sensing images P of white point positional representation surveyed area.
Fig. 4 a is the remote sensing images P of the surveyed area of emulation experiment 2; Fig. 4 b is total similarity Score distribution plan S that the remote sensing images of Fig. 4 a adopt document 1 algorithm to obtain; Fig. 4 c is total similarity Score distribution plan S that the remote sensing images of Fig. 4 a adopt embodiment of the present invention method to obtain; Aircraft center in Fig. 4 b and Fig. 4 c in the remote sensing images P of white point positional representation surveyed area.
Fig. 5 a is the remote sensing images P of the surveyed area of emulation experiment 3; Fig. 5 b is total similarity Score distribution plan S that the remote sensing images of Fig. 5 a adopt document 1 algorithm to obtain; Fig. 5 c is total similarity Score distribution plan S that the remote sensing images of Fig. 5 a adopt embodiment of the present invention method to obtain; Aircraft center in Fig. 5 b and Fig. 5 c in the remote sensing images P of white point positional representation surveyed area.
From above emulation experiment figure, can find out, for the simple Fig. 3 a of background and objective contour Fig. 4 a clearly, adopt document 1 and the inventive method all can obtain accurate testing result; But poor to picture quality, Fig. 5 a that background is more complicated, the testing result that the inventive method obtains is accurate compared with document 1, this is because document 1 adopts gradient orientation histogram feature as parts wave filter, and principal direction distributes to estimate by the direction in statistical gradient direction histogram, because wave filter and principal direction are estimated all to take picture material as guidance, affected by complex background; And how much parts that the inventive method adopts can be resisted the impact of the factors such as color, texture, illumination, the method that principal direction guides by profile is estimated, is conducive to reduce the impact of background on statistics.
Above the simulation experiment result shows, the inventive method can realize the Airplane detection of different directions under complex background, and its positioning precision is high than existing methods, at Remote Sensing Target detection field, has certain feasibility and applicability.
Claims (3)
1. the remote sensing based on geometry parts and radial gradient angle detects a method for aircraft, and its step is as follows:
The foundation of A, the star-like partial model of aircraft:
A1, by the known aircraft image rectification in known remote sensing images be head straight up, obtain correcting image;
A2, the pixel value of aircraft part in correcting image is set to 1, background parts pixel value is set to 0, obtains the two-value profile P of known aircraft
b;
A3, the two-value profile P to known aircraft
bcarry out Its Sparse Decomposition, obtain optimum atomic diagram image set
for optimum atomic diagram is as number, i is optimum atomic diagram picture
index number,
be i optimum atomic diagram picture;
A4, get i optimum atomic diagram picture
in non-zero region obtain i geometric figure
with i geometric figure
as i parts Part
i, with the two-value profile P of known aircraft
bfor root part Root, carry out Star Model association; When associated, i parts Part
ibe shaped as i geometric figure
shape
i parts Part
ito root part Root center (x
r, y
r) relative coordinate position be i geometric figure
center is to atomic diagram picture
the relative coordinate at center
thereby i parts Part
iprincipal direction be i geometric figure
principal direction
wherein, x
r, y
rbe respectively horizontal ordinate and the ordinate of root part Root center in correcting image;
A5, repetition A4, by all geometric figures
all as parts Part
icarry out associatedly with root part Root, set up and to obtain the star-like partial model of aircraft;
Airplane detection in B, remote sensing images
B1, remote sensing via satellite obtain the remote sensing images P of surveyed area, get the two-value profile P of size and known aircraft
bidentical detection window carries out point by point scanning in remote sensing images P, obtains subimage block P
sub;
B2, radially-gradient angle Voting Algorithm of employing calculate subimage block P
subprincipal direction ξ
sub;
B3, aircraft Star Model is rotated to heading and subimage block P
subprincipal direction ξ
subunanimously, calculate each parts Part in postrotational aircraft Star Model
iprincipal direction
and with root part Root center (x
r, y
r) relative coordinate
B4, in aircraft Star Model according to i parts Part
ipostrotational principal direction
i parts Part
irotate rear and root part Root center (x
r, y
r) relative coordinate
and the shape of i parts
adopt method of geometry to calculate rear i i the region that parts cover of aircraft Star Model rotation
B5, by i region in aircraft Star Model
interior pixel is set to 1, and other pixel is set to 0, generates i two-value component diagram
that is:
In formula, x, y is respectively two-value component diagram
horizontal ordinate in correcting image and ordinate;
B6, i two-value component diagram of calculating
histograms of oriented gradients, generate i parts wave filter F
i:
In formula, HOG () is histograms of oriented gradients feature extraction operator;
The subimage block P of B6, calculating remote sensing images P
subhistograms of oriented gradients H
sub:
H
sub=HOG(P
sub)
B7, calculating subimage block P
subtotal similarity Score with all parts in postrotational aircraft Star Model:
In formula, < > is inner product operator;
B8, repeat B1~B7, calculate total similarity Score of all detection windows in remote sensing images P, obtain total similarity Score distribution plan S;
C, in total similarity Score distribution plan S, have the total similarity Score that is greater than setting threshold, judge in surveyed area and have aircraft, its corresponding detection window position in remote sensing images is aircraft location.
2. the remote sensing based on geometry parts and radial gradient angle according to claim 1 detects the method for aircraft, it is characterized in that the two-value profile P to known aircraft in described steps A 2
bthe specific practice of carrying out Its Sparse Decomposition is:
1) generate former word bank D how much
A, construct oval former word bank D
e:
Define oval atomic diagram as g
e(x, y) is as follows:
In formula, x, y is respectively that oval atomic diagram is as g
e(x, y) horizontal ordinate and ordinate in correcting image, oval atomic diagram is as g
ethe two-value profile P of the size of (x, y) and known aircraft
bit is identical,
for oval atomic diagram is as g
eelliptic region in (x, y), by following formula, calculated:
In formula, u, v be respectively horizontal ordinate x and ordinate y with oval atomic diagram as g
e(x, y) center is that the point of rotation turns clockwise
after new horizontal ordinate and ordinate, M
pBand N
pBbe respectively the two-value profile P of known aircraft
blength and width; (px, py) is elliptic region
center is to the two-value profile P of known aircraft
bthe relative coordinate at center, px ∈ [M
pB/ 2, M
pB/ 2], py ∈ [N
pB/ 2, N
pB/ 2]; A and b are respectively elliptic region
major axis and minor axis,
for elliptic region
principal direction,
By parameter a, b,
px, py gets the oval atomic diagram that generates all over above all values as g
e(x, y), forms oval former word bank D
e:
B, construct trapezoidal former word bank D
t:
Define trapezoidal atomic diagram as g
t(x, y) is as follows:
In formula, x, y is respectively that trapezoidal atomic diagram is as g
t(x, y) horizontal ordinate and ordinate in correcting image, trapezoidal atomic diagram is as g
tthe two-value profile P of the size of (x, y) and known aircraft
bidentical;
for trapezoidal atomic diagram is as g
ttrapezoid area in (x, y), is surrounded by four straight lines, by following formula, is calculated:
In formula, u, v is respectively that horizontal ordinate x and ordinate y justify atomic diagram as g with ladder
t(x, y) center is that the point of rotation turns clockwise
after horizontal ordinate and ordinate, ub, lb and h are respectively trapezoidal upper base, go to the bottom and high,
f is upper base center and the side-play amount at the center of going to the bottom, regulation upper base center toward left avertence for negative, toward right avertence for just, f ∈ [50,50];
with px, the definition of py and span and oval atomic diagram are as g
e(x, y) is identical;
By parameters u b, lb, h, f,
px, py gets the trapezoidal atomic diagram that generates all over all values as g
t(x, y), forms trapezoidal former word bank D
t:
C, by the former word bank D of all ellipses
ein atomic diagram as g
e(x, y) and trapezoidal former word bank D
tatom image g
t(x, y) merges, and generates the two-value profile P of known aircraft
bthe former word bank D of geometry:
D=D
e∪D
t
D, by all atomic diagram in how much former word bank D as g
e(x, y), g
t(x, y) renames numbering, is expressed as g
j(x, y), j=1,2 ..., N, N be atomic diagram in how much former word bank D as number, j renames atomic diagram after numbering as g
jthe index number of (x, y);
2) set up the optimal model of aircraft profile rarefaction representation:
min||Ω||
0
<g
i,g
j>=0,i≠j
In formula, || ||
0for 0-norm, the element number of expression set, || ||
2for 2-norm, represent that energy, < > are that indicate satisfied condition, σ of inner product operator, s.t. is energy threshold, get the two-value profile P of known aircraft
b15% of area, i.e. σ=15% * Area (P
b), Area () is that area operator, Ω are the subspace in the former word bank D of geometry satisfying condition,
3. the remote sensing images Airplane detection algorithm based on how much parts and radial gradient angle according to claim 1, is characterized in that, the employing in described step B2 radially-gradient angle Voting Algorithm calculates subimage block P
subprincipal direction ξ
subspecific practice be:
A, extraction subimage block P
subtwo-value profile P with known aircraft
bhistograms of oriented gradients feature H
sub(m, n, γ) and H
b(m, n, γ):
H
sub(m,n,γ)=HOG(P
sub)
H
B(m,n,γ)=HOG(P
B)
In formula, m, n, γ is respectively the horizontal ordinate in histograms of oriented gradients, ordinate and gradient direction;
B, be calculated as follows subimage block P
subtwo-value profile P with known aircraft
bmain gradient direction figure
with
C, by subimage block P
submain gradient direction figure
two-value profile P with known aircraft
bmain gradient direction figure
be converted to polar form, obtain subimage block P
subdirection gradient polar plot PH
subthe two-value profile P of (r, θ) and known aircraft
bdirection gradient polar plot PH
b(r, θ), wherein, r, θ is respectively utmost point footpath coordinate and the polar angle coordinate of direction gradient polar plot;
D, press following formula antithetical phrase image block P
subdirection gradient polar plot PH
sub(r, θ) screens, and obtains candidate direction gradient-pole coordinate diagram C (r, θ):
In formula, inf () is that infimum operator, ε are threshold value, get 0.1-0.2, Φ=θ | PH
b(r, θ) ≠-10} is the two-value profile P of known aircraft
bdirection gradient polar plot PH
bin (r, θ), with subimage block P
subdirection gradient polar plot PH
sub(r, θ) has same pole footpath r, and gradient direction value is not-10 polar angle collection, RGA
sub(r, θ) and RGA
b(r, θ) is respectively subimage block P
subtwo-value profile P with known aircraft
bradially-gradient angle, be calculated as follows:
RGA
sub(r,θ)=|PH
sub(r,θ)-θ|
RGA
B(r,θ)=|PH
B(r,θ)-θ|
E, make gradient direction in candidate direction gradient-pole coordinate diagram C (r, θ) for-10 point is available point, and be calculated as follows out the prediction rotation angle R (k) of available point:
In formula, the index number that k is available point, N
cfor the number of available point,
be k the corresponding polar angle of available point,
two-value profile P for known aircraft
bdirection gradient polar plot PH
bpoint in (r, θ) in radially-gradient angle and candidate direction gradient-pole coordinate diagram
the corresponding polar angle of immediate point, wherein
for candidate direction gradient-pole coordinate diagram mid point
radially-gradient angle;
F, by following formula to the statistics of voting of the rotation angle in the prediction rotation angle R (k) of available point, obtain the histogram His of rotation angle
r:
His
R(λ)=#{k|R(k)=λ}
In formula, # () represents the number of set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310703176.9A CN103714344B (en) | 2013-12-19 | 2013-12-19 | Geometrical component and radial gradient angle-based aircraft remote sensing detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310703176.9A CN103714344B (en) | 2013-12-19 | 2013-12-19 | Geometrical component and radial gradient angle-based aircraft remote sensing detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103714344A true CN103714344A (en) | 2014-04-09 |
CN103714344B CN103714344B (en) | 2017-02-08 |
Family
ID=50407301
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310703176.9A Expired - Fee Related CN103714344B (en) | 2013-12-19 | 2013-12-19 | Geometrical component and radial gradient angle-based aircraft remote sensing detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103714344B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104392432A (en) * | 2014-11-03 | 2015-03-04 | 深圳市华星光电技术有限公司 | Histogram of oriented gradient-based display panel defect detection method |
US10592761B2 (en) | 2017-09-01 | 2020-03-17 | Boe Technology Group Co., Ltd. | Image processing method and device |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2010013236A1 (en) * | 2008-07-29 | 2010-02-04 | Superfish Ltd | Increasing interest point coverage in an image |
CN102004922A (en) * | 2010-12-01 | 2011-04-06 | 南京大学 | High-resolution remote sensing image plane extraction method based on skeleton characteristic |
CN102298698A (en) * | 2011-05-30 | 2011-12-28 | 河海大学 | Remote sensing image airplane detection method based on fusion of angle points and edge information |
CN102945374A (en) * | 2012-10-24 | 2013-02-27 | 北京航空航天大学 | Method for automatically detecting civil aircraft in high-resolution remote sensing image |
CN103246887A (en) * | 2013-03-26 | 2013-08-14 | 中国科学院电子学研究所 | Airport object multithreading detection method based on optical remote sensing images with geometrical characteristics |
-
2013
- 2013-12-19 CN CN201310703176.9A patent/CN103714344B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2010013236A1 (en) * | 2008-07-29 | 2010-02-04 | Superfish Ltd | Increasing interest point coverage in an image |
CN102004922A (en) * | 2010-12-01 | 2011-04-06 | 南京大学 | High-resolution remote sensing image plane extraction method based on skeleton characteristic |
CN102298698A (en) * | 2011-05-30 | 2011-12-28 | 河海大学 | Remote sensing image airplane detection method based on fusion of angle points and edge information |
CN102945374A (en) * | 2012-10-24 | 2013-02-27 | 北京航空航天大学 | Method for automatically detecting civil aircraft in high-resolution remote sensing image |
CN103246887A (en) * | 2013-03-26 | 2013-08-14 | 中国科学院电子学研究所 | Airport object multithreading detection method based on optical remote sensing images with geometrical characteristics |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104392432A (en) * | 2014-11-03 | 2015-03-04 | 深圳市华星光电技术有限公司 | Histogram of oriented gradient-based display panel defect detection method |
US10592761B2 (en) | 2017-09-01 | 2020-03-17 | Boe Technology Group Co., Ltd. | Image processing method and device |
Also Published As
Publication number | Publication date |
---|---|
CN103714344B (en) | 2017-02-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11783569B2 (en) | Method for classifying hyperspectral images on basis of adaptive multi-scale feature extraction model | |
CN111914795B (en) | Method for detecting rotating target in aerial image | |
CN111160440B (en) | Deep learning-based safety helmet wearing detection method and device | |
CN103400151B (en) | The optical remote sensing image of integration and GIS autoregistration and Clean water withdraw method | |
CN103258214B (en) | Based on the Classifying Method in Remote Sensing Image of image block Active Learning | |
CN104463249B (en) | A kind of remote sensing images airfield detection method based on Weakly supervised learning framework | |
Storie et al. | Deep learning neural networks for land use land cover mapping | |
CN103218831B (en) | A kind of video frequency motion target classifying identification method based on profile constraint | |
CN105373135A (en) | Method and system for guiding airplane docking and identifying airplane type based on machine vision | |
CN110163177A (en) | A kind of wind power generation unit blade unmanned plane automatic sensing recognition methods | |
CN107481315A (en) | A kind of monocular vision three-dimensional environment method for reconstructing based on Harris SIFT BRIEF algorithms | |
CN103824049A (en) | Cascaded neural network-based face key point detection method | |
CN103822616A (en) | Remote-sensing image matching method with combination of characteristic segmentation with topographic inequality constraint | |
CN110084165A (en) | The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations | |
CN105550675B (en) | A kind of binocular pedestrian detection method based on optimization polymerization integrating channel | |
CN105869178A (en) | Method for unsupervised segmentation of complex targets from dynamic scene based on multi-scale combination feature convex optimization | |
CN103886589A (en) | Goal-oriented automatic high-precision edge extraction method | |
CN110070025A (en) | Objective detection system and method based on monocular image | |
CN108776777A (en) | The recognition methods of spatial relationship between a kind of remote sensing image object based on Faster RCNN | |
Ma et al. | Crlf: Automatic calibration and refinement based on line feature for lidar and camera in road scenes | |
Siva et al. | Voxel-based representation learning for place recognition based on 3d point clouds | |
CN103971112A (en) | Image feature extracting method and device | |
Chen et al. | Identification of autonomous landing sign for unmanned aerial vehicle based on faster regions with convolutional neural network | |
CN111123962A (en) | Rotor unmanned aerial vehicle repositioning photographing method for power tower inspection | |
CN104318051A (en) | Rule-based remote-sensing automatic extraction system and method of wide-range water body information |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170208 Termination date: 20191219 |
|
CF01 | Termination of patent right due to non-payment of annual fee |