CN103886285B - Optical remote sensing image Ship Detection under priori geography information auxiliary - Google Patents
Optical remote sensing image Ship Detection under priori geography information auxiliary Download PDFInfo
- Publication number
- CN103886285B CN103886285B CN201410086835.3A CN201410086835A CN103886285B CN 103886285 B CN103886285 B CN 103886285B CN 201410086835 A CN201410086835 A CN 201410086835A CN 103886285 B CN103886285 B CN 103886285B
- Authority
- CN
- China
- Prior art keywords
- detection
- remote sensing
- image
- region
- point
- 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
Landscapes
- Image Processing (AREA)
Abstract
A kind of optical remote sensing image Ship Detection under priori geography information auxiliary, including set up the harbour priori geographical information library related to coastline, the remote sensing image to be detected after being corrected；Carry out region segmentation and extract boundary line, seashore line vector is obtained according to seashore line vector storehouse, carry out coastline change detection；The extra large isolated water area in land and region of pulling in shore；For water area, naval vessel suspected target detection first is carried out based on many vision significances, then the machine learning method based on multiple features detects naval vessel in naval vessel suspected target；For region of pulling in shore, carry out the detection of global conspicuousness and obtain initial suspicious region, carrying out image segmentation further according to shape information obtains final suspicious region, pulled in shore Ship Target using multiple features constraint detection afterwards.The present invention is started with from various features and is solved complicated optical remote sensing image ship detection problem using priori geography information, vision significance, machine learning, evidence theory.
Description
Technical field
The present invention relates to Remote Sensing Image Processing Technology field, more particularly, to a kind of remote sensing shadow based on priori geography information
As the high accuracy reliability detection method on naval vessel in complex scene.
Background technology
Naval vessel has very important existing as marine transportation carrier and important military target, its automatic detection with identification
Sincere justice, finds and relief, fisherman monitoring, illegal immigrant, defendance territory, anti-drug, naval vessel illegal dumping greasy dirt on naval vessel
Detection with marine transportation manage etc. aspect suffer from being widely applied.Optical remote sensing image is due to covering amplitude broad, revisiting
Cycle is short, imaging resolution are high, abundant in content, the features such as meet mankind's intuitivism apprehension, make the most effective of naval vessel detection
Means.
Naval vessel detection is carried out on high-resolution complexity optical image there is following problem：（1）In complicated marine environment bar
Under part, sea presented on satellite remote-sensing image rambling fish scale light, large area retroreflective regions, irregular movement it is abundant
Texture wave etc., middle-size and small-size Ship Target may be hidden in the background clutter of complexity, so as to influence the detection of Ship Target
With identification.（2）Sea situation (stormy waves), meteorological (cloud and mist), water colour etc. cause remote sensing image Zhong Hai lands characteristic not on Ship Target image
The interference such as stabilization, periphery island is very more, and background complicated and changeable makes target too poor with the separability of background, and naval vessel detection is difficult
Degree is big.（3）The species of high-resolution remote sensing image is more, yardstick differs, and target detail can be variant in image, and Ship Target
Due to being differed greatly in class itself（Aircraft carrier, warship, merchant ship, fishing boat etc.）, cause appearance " generic different characteristic " and " different with characteristic
The phenomenon of classification ", so that Ship Target validity feature extracts difficult, to quick effective detection and the identification band of remote sensing target
Challenge is newly carried out.
For the Ship Target Detection in high-resolution optical image, the naval vessel that pulled in shore to marine vessel and harbour is generally included
The class of detection two.Because sea natural background contrasts obvious with naval vessel in gamma characteristic, detection is relatively easy to, therefore mesh
Preceding most of documents conduct a research mainly for surface vessel detection.For the naval vessel that harbour is stopped, due to its background area bag
Include sea and man-made target harbour, the gray scale on offshore naval vessel, textural characteristics and shore facilities are very close, and because the two is frequent
Adhesion and the influence of shade so that pulling in shore, naval vessel automatic detection difficulty is big, and document is less.
The algorithm of existing Ship Target Detection generally carries out extra large land separation first；And naval vessel mesh is utilized on this basis
Mark and the difference of marine background, obtain the doubtful candidate region comprising Ship Target from marine site；The last profit in candidate region
With naval vessel feature and other disturbing factors（Cloud, wave, clutter, island etc.）Further discriminate between, carry out the confirmation of Ship Target with
Classification.Typical method is included based on gray-scale statistical characteristics, based on fractal theory, view-based access control model attention model three major types.It is based on
The method of gray-scale statistical characteristics is mainly using water body and the gray-scale statistical otherness feature of Ship Target（Gray scale, image information entropy,
Morphology contrast, partial statistics variance etc.）Image Segmentation is carried out, so as to obtain Ship Target candidate region.Such method one
As be applied to that sea is more tranquil, texture is uniform and the relatively low situation of water body gray scale.And in the case of complicated for sea, such as sea
The disturbing factor such as wave, cloud cover or water body gray scale is brighter, noise, shade, adds the black and white polarity of Ship Target, same naval vessel
The gray feature of target different parts is also inconsistent, and the method is also easy to produce more false dismissal and false-alarm.Side based on fractal theory
Method has certain difference principle using the fractal dimension of natural scene and Ship Target, causes to use fractal theory according to this difference
Naval vessel detection is carried out with technology to be possibly realized, its testing result stabilization, better than rim detection, Threshold segmentation, but when background compares
When complicated, when being such as subject to cloud and mist to disturb, the reduction of background self-similarity is larger with Fractal model fit error, algorithm detection effect
Rate is than relatively low.In recent years due to visual attention method by human visual system can quickly focus on area-of-interest characteristic and quilt
It is incorporated into the extraction of Ship Target candidate region, the model has taken into account part by simulating the optic nerve mechanism of human brain
The regularity of randomness and the overall situation, testing result is connective preferably, has good Shandong to noise, fuzzy, contrast and brightness
Rod.But it is sensitive for image dimensional variation, and " conspicuousness target " on different scale differs greatly, when image is covering
Area is very wide, and during jumbo remote sensing image comprising atural objects such as land, island, naval vessel is no longer notable mesh relative to view picture image
Mark, now the method for view-based access control model attention model is no longer valid.
Although the above method all solves the problems, such as that naval vessel is detected to varying degrees, there is problems with：1）
The existing method for extracting Ship Target feature is often processed just for specific image, small range image blocks, the spy of extraction
Levy limitation big, detection efficiency is low, it is impossible to suitable for the naval vessel detection on complicated sea situation image on a large scale；2）Due to image matter
Wave, cloud and mist, island and naval vessel multifarious interference in itself under amount, complex background, cause existing method to detect Ship Target
False-alarm and loss are high, and testing result reliability is not high；3）Inshore ship detection concern is less, even if some methods make use of elder generation
Testing geography information carries out extra large land separation, but is mainly used in rough estimate sea land position and carries out surface vessel detection.Actual conditions
In pull in shore naval vessel because gray scale, textural characteristics and shore facilities are very close, and due to the two frequent adhesion and the shadow of shade
Ring so that Inshore ship detection is increasingly complex.Generally speaking existing Ship Detection universality, reliability have much room for improvement,
It is not high especially for complex scene precision.Therefore need urgently to find a kind of detection efficiency is high, universality is high and take into account sea with
Pull in shore the detection method on naval vessel.
The content of the invention
In order to overcome prior art defect, the present invention to propose the complicated remote sensing under a kind of auxiliary based on priori geography information
Ship Detection in image.
The present invention proposes the optical remote sensing image Ship Detection under a kind of priori geography information auxiliary, including following
Step：
Step 1, sets up the harbour priori geographical information library related to coastline, including multiple dimensioned harbour control point image
Storehouse and seashore line vector storehouse；
Step 2, the remote sensing image to be detected after being corrected using RPC parameters, or according to remote sensing image to be detected
Geographical coordinate, takes corresponding control point image and is matched from harbour image database for control point, the remote sensing to be detected after being corrected
Image；
Step 3, to correction after remote sensing image to be detected carry out region segmentation extract boundary line, according to seashore line vector storehouse
Seashore line vector is obtained, coastline change detection is carried out；
Step 4, using the seashore line vector obtained according to seashore line vector storehouse, realizes that extra large land separates, and obtains water area
With region of pulling in shore；
Step 5, for water area, first carries out naval vessel suspected target detection based on many vision significances, then based on more special
The machine learning method levied detects naval vessel in naval vessel suspected target；
Step 6, for region of pulling in shore, the region as detection zone of pulling in shore after first step 4 sea land is separated, in detection zone
Global conspicuousness detection is carried out in domain, initial suspicious region is obtained；Image segmentation is carried out further according to shape information, obtains final
Suspicious region；Pulled in shore Ship Target using multiple features constraint detection afterwards.
And, step 3 includes following sub-step,
Step 3.1, the coastline of superposition correspondence remote sensing image geographic coordinate range to be detected, including following sub-step,
Step 3.1.1, one or more in remote sensing image geographic coordinate range to be detected is taken from seashore line vector storehouse
Point string is used as initial priori line of vector；
Step 3.1.2, takes each point string the first two point and does extended line respectively with most latter two point, and with remote sensing shadow to be detected
The edge of picture forms intersection point；
Step 3.1.3, initial priori line of vector is added using former and later two intersection points as beginning and end, obtains new sea
Water front line of vector；
Step 3.2, detection, including following sub-step are changed to coastline,
Step 3.2.1, to correction after remote sensing image to be detected carry out region segmentation, extract boundary line；
Step 3.2.2, boundary line is matched with step 3.1.3 gained line of vectors, calculates boundary line with coastline
Coincidence factor；
Step 3.2.3, if coincidence factor reaches corresponding predetermined threshold value, into step 4, otherwise stops flow, and prompting is carried out
Coastline updates.
And, in step 5, carrying out naval vessel suspected target detection based on many vision significances includes following sub-step,
Step 5.1.1, the water area after step 4 sea land is separated is carried out complete as detection zone in detection zone
Office's conspicuousness detection, the global conspicuousness at detection zone midpoint (i, j) uses Sg(i, j) is represented,
Step 5.1.2, carries out local conspicuousness detection in detection zone, and the part at detection zone midpoint (i, j) is significantly
Property uses Sl(i, j) is represented；
Step 5.1.3, the comprehensive notable figure of calculating is as follows,
Definition standard function N (s) is as follows,
N (s)=(s-min (s))/(max (s)-min (s))
In formula, s represents the global or local significance value of every bit in region, and min (s), max (s) represent region respectively
The minimum of interior conspicuousness, maximum；
The comprehensive notable figure S of definitioncIt is as follows,
In formula, N (Sl)、N(Sg) represent respectively according to local, the global saliency value after normalized function N (S) standardization.
And, the implementation that global conspicuousness detection is carried out in detection zone is as follows,
If detection zone has N number of pixel, the RGB average values in detection zone are as follows,
Wherein, r (i, j), g (i, j), b (i, j) are the face at (i, j) place in detection zone on the image to be detected after correcting
Color characteristic value；
Detection zone internal image is carried out into Gaussian smoothing, the new color feature value r of point (i, j) in detection zone is obtainedG
(i,j),gG(i,j),bG(i,j)；
The global significance at detection zone midpoint (i, j) is expressed as follows,
In formula, | | | | represent L2Norm.
And, in step 6, image segmentation is carried out according to shape information, obtain the implementation of final suspicious region such as
Under,
First, Hausdorff distance matchings are carried out to the region being partitioned into using shape information；Whether judge target afterwards
Stick together, if it has not, directly being pulled in shore Ship Target using multiple features constraint detection, such as occur, carried out using morphology operations
Target Segmentation, is then pulled in shore Ship Target using multiple features constraint detection.
And, in step 6, pulled in shore Ship Target using multiple features constraint detection, including calculate the confidence level on the naval vessel that pulls in shore
It is as follows,
In formula, RjIt is the confidence level of evidence j.ρjIt is the degree of membership of evidence, J is evidence total number；
According to the advance confidential interval for dividing, refusal interval, indeterminacy section, it is believed that confidence level is in the right of confidential interval
As if ship, it is incorporated to object set Bs, what confidence level was in refusal interval is other atural objects, is incorporated to object set BN。
And, if J=4, according to default area features evidence confidence level RiArea, textural characteristics evidence confidence level Ri3, chain
Code feature evidence confidence level RiT, pull in shore apart from evidence confidence level RiLWith individual features degree of membership ρArea、ρ3、ρT、ρL, calculating pulls in shore
The confidence level on naval vessel,
A) area and girth feature degree of membership ρAreaDetection is as follows,
ρArea=80*area/perimeter^2
Wherein area represents imagery coverage, and perimeter represents image girth；
B) textural characteristics degree of membership ρ3Detection is as follows,
S in formulaiIt is pixel gray value, p (si) be region shared by the pixel with the gray value ratio, its average
ForL is the maximum of image intensity value；
C) chain code feature degree of membership ρTDetection is as follows,
(1) the difference chain code d of the Freeman chain codes on the border in region to be checked is obtained;
(2) according to d, the angle sequence θ on the border in region to be checked is obtained, if θiIt is the folder at i-th point in angle sequence θ
Angle,
Wherein, n is the number of corner point on curve, and the span of i is 0 to n-1, diIt is i-th -1 o'clock to i-th point
Distance, di+1It is i-th point to i+1 point of distance, ziIt is the distance of the i-th -1 point to i+1 point；
(3) degree of membership is obtained
Wherein, θmiIt is the angle at i-th point on image edge curve to be checked, θliIt is the folder at i-th point of template curve
Angle.
D) pull in shore to detect as follows from degree of membership away from feature,
Wherein Ax+By+C=0 is the line of vector line segment closest with suspected target,（X, y）It is suspected target center of mass point
Coordinate.
And, after step 2 obtains the remote sensing image to be detected after correcting, water body index judgement is carried out, if waiting after correcting
Detection remote sensing image includes more than 99% sea, then be directly entered step 6, otherwise enter step 3.
Contrast prior art, beneficial features of the invention are as follows：
1）The foundation of priori geographical information library：According to detection image there is geographical coordinate feature and geography information to have permanent
Stability feature, sets up the precision vector line database at multiple dimensioned image database for control point and coastline library particularly emphasis harbour；
2）Separated using priori geography information sea land：On the basis of carrying out registration to image using image database for control point, adjust
The vector taken in vector storehouse is overlapped with image to be checked, by judging the set and situation of line of vector and image, carries out coastline
Change detection, realizes that precisely sea land separates, and improves naval vessel accuracy of detection；
3）ROI suspicious regions detection based on many vision notable features：Doubted using many vision significance fast searching naval vessels
Like region, the separation reduction missing inspection of target-background can be well carried out, while removing the false-alarms such as the cloud of large area, wave, island
Interference；
4）The integrated monomer naval vessel detection of multiple features：The distinctive geometry in integrated application naval vessel, texture, shape, V-shape structure
Etc. feature as basis for estimation, Ship Target step-sizing is carried out with reference to evidence theory, realization is pulled in shore and ShipTargets
Real-time detection；
5）Feedback and renewal：Coastline such as changes, and updates coastline；The image collection of new detection and renewal control point shadow
As in storehouse；Ship Target and other jamming targets are respectively added in positive negative example base according to testing result, and train renewal
Grader.
Brief description of the drawings
Fig. 1 is flow chart of the invention.
Specific embodiment
Complicated sea remote sensing image Ship Detection of the present invention based on geographical prior information is using the notable of naval vessel
Feature and coastline geography information feature, extra large land separation is carried out to remote sensing image, obtains sea image and bank range image, right
Sea image is pre-processed, and carries out the detection on single naval vessel to the sub-image block with the presence of doubtful naval vessel further according to image entropy.
Embodiment flow can realize automatic running using computer software technology, as shown in figure 1, specifically including following step
Suddenly：
Specific implementation process is as follows：
Step 1, the satellite-remote-sensing image possessed using user and GOOGLEARTH etc. set up multiple dimensioned harbour control
Point Image Database, seashore line vector storehouse is set up using the open geographic information such as OPENSTREETMAP source, and the geographical letter of priori is constituted together
Breath storehouse.
Typically using SPOT, QUICKBIRD, third satellite image is provided.During specific implementation, can obtain in global range
The latitude and longitude coordinates at built harbour, set up harbour information table.Global harbour data are downloaded on GOOGLEARTH, while collection is every
The coastline vector data of individual harbour region.Longitude and latitude scope according to harbour belonging country and place sets up harbour image
Material database, i.e. harbour image database for control point.
Coastline data are divided into two kinds, port area coastline data and nature coastline data.For harbour data, can
Obtained in the way of to take artificial precise acquisition in advance；For nature coastline data, data source is increased income for OPENSTREETMAP
Data.Downloading process is as follows：
A) using OpenStreetMap API (XAPI), region is downloaded in selection, builds a range boundary frame, then build
One http api URL, vector initial data is downloaded from OSM servers.The form type of API is as follows：
http://www.overpass-api.de/api/xapi?map?bbox=113.8623,30.19262,
114.85107,30.87394
B) initial data of the vector to downloading is parsed.
C) vector data form needed for being created using third-party storehouse, by vector needed for parsing the data obtained write-in.
D) the nature coastline data of download are interrupted by longitude and latitude grid, according to GeoHash principles, is with longitude and latitude
Basis is numbered to every group of coastline.Harbour data are also numbered according to above-mentioned criterion afterwards, and replace respective range
Interior coastline data.
Step 2, using to be detected remote sensing image of the image RPC parameters to after correcting, or according to remote sensing image to be detected
Geographical coordinate, the remote sensing image to be detected matched after being corrected with existing accurate control points harbour Image Database.
First determine whether the remote sensing image of input（Image i.e. to be detected）Whether there is RPC parameters, have and then use RPC parameters pair
Image carries out geometric correction.If there is no RPC parameters, remote sensing image is positioned according to image database for control point.
In embodiment, the remote sensing image of input then directly carries out geometry using existing software if RPC parameters to image
Correct.For the image to be detected of no RPC parameters, registration is carried out using Image Database Memory Reference image.With specific reference to input
The approximate range of remote sensing image takes out corresponding control point image as reference picture from Image Database, is carried out with input image
Sift is matched, and obtains the image coordinate of same point on two width images.Then the pixel coordinate root with reference to the point on image
Geographical coordinate is converted into according to the starting point geographical coordinate with reference to image and geographical resolution ratio, with reference to the picture of the point on image to be detected
Plain coordinate is corrected.
Further to improve efficiency, after obtaining the remote sensing image to be detected after correcting, water body index first can be carried out to it and sentenced
It is disconnected.If it includes more than 99% sea, the surface vessel detection of step 6 is directly entered, otherwise then enters step 3, it is right to match
The coastline answered.
Step 3, due to the trend of the times sex chromosome mosaicism in coastline, using existing correspondence seashore line vector on the basis of Image Matching
Superposition, and carry out coastline change detection.To improve efficiency, can advanced row water body index judgement.
Step 3 in embodiment is specifically included：
Step 3.1, the coastline of superposition correspondence remote sensing image scope to be detected.
During specific implementation, it is divided into following steps：
Step 3.1.1, indexes according to line of vector longitude and latitude, is taken in remote sensing image geographic coordinate range from seashore line vector storehouse
One or more interior point string is used as initial priori line of vector.
Step 3.1.2, because the corresponding image capturing range in each coastline is whole harbour, and remote sensing image is tended not to
Just include whole port area, therefore take each point string the first two point and do extended line respectively with most latter two point, and with it is to be checked
The edge for surveying remote sensing image forms intersection point.
Step 3.1.3, initial priori line of vector is added using former and later two intersection points as beginning and end, gained it is new
Coastline line of vector just can be corresponding with image capturing range.
Step 3.2, detection is changed to coastline, judges whether it changes.
Step 3.2.1, to correction after remote sensing image to be detected carry out region segmentation, extract its boundary line.
Step 3.2.2, boundary line is matched with step 3.1.3 gained line of vectors, calculates its boundary line and coastline
Coincidence factor.
Step 3.2.3, if coincidence factor reaches corresponding predetermined threshold value（Such as 70%）, then be considered as set and.If set and, automatically
Into next step, i.e. step 4.If do not cover and, can stop this flow, point out user to carry out artificial semi-automatic extraction update section
The coastline divided, updates to carry out correspondence coastline.
Step 4, using line of vector when being matched in step 3.2.2, realizes that accurate sea land separates.
When extra large land separates, need to judge coastline which side be ocean, which side is land, therefore, united in gatherer process
One regulation, by the direction of origin-to-destination（Gather direction）, the left side is sea, and the right is land.In system operation, build first
It is vertical with correct after remote sensing image size to be detected, the consistent completely black mask image of coordinate, judge afterwards the starting point in coastline with
Terminal falls on which bar side of image, the angle point point insertion point string on selection image correspondence land, line of vector is turned into one comprising land
The curve of the closing in ground region, it is white that then be partially filled with sea by newly-built mask image, completes extra large land and separates, and obtains sea
Region and region of pulling in shore.
Step 5, for the ShipTargets of water area, using many visions are aobvious etc., feature combination machine learning is examined
Survey：The marine vessel detection time-division is thick, subdivision of reservoir screening strategy, to improve detection efficiency.It is aobvious first with many visions across the sea
Work property fast searching naval vessel suspected target, and texture, shape and the geometry feature of suspected target are extracted on this basis, adopt
The further detection that Ship Target is carried out with SVM methods confirms.
During specific implementation, including following sub-step：
Step 5.1, the naval vessel suspected target detection based on many vision significances.Using the overall situation first in large area marine site
Conspicuousness detects that being detected through global conspicuousness carries out target-background separation, will divide with Ship Target Sea background on a large scale
Leave and.Secondly using local conspicuousness detection, it is special to have in prominent subrange（It is rare）Gray scale, the target in direction, suppression
Make large-scale cloud and wave region.Finally two class notable figures are integrated and globally consistent sex factor is added, this process subtracts significantly
The influence of partly cloudy and clutter, can also highlight for the naval vessel under thin obnubilation lid, so as to preferably extract the doubtful mesh in naval vessel
Mark.Detailed process is as follows：
Step 5.1.1, the water area after step 4 sea land is separated is carried out global aobvious as detection zone in region
The detection of work property：
Frequency tuning (Frequency-tuned, FT) the significantly inspection that Achanta et al. is proposed is used for reference in global conspicuousness detection
Survey method, it is significance that each characteristic mean in image RGB color and the difference after image Gaussian smoothing are used herein.The party
Method simply easily realizes, can extract more complete well-marked target, and its internal consistency is good.Simultaneously as cloud, wave and clutter are bright
Degree characteristic high, causes to be disturbed comprising substantial amounts of these classes in testing result.It is calculated as follows：
Region（Have N number of pixel）Interior color characteristic represents with the RGB average values of whole region respectively, its global average
Respectively
（1）
R (i, j), g (i, j), b (i, j) are the color characteristic at (i, j) place in water area on image to be detected after correcting
Value.
The influence of noise and clutter is reduced, intra-zone image is carried out into Gaussian smoothing, obtain new color feature value, point
The color feature value of (i, j) is expressed as：rG(i,j),gG(i,j),bG(i,j)；
The global significance of each point is expressed as in region：
（2）
In formula, | | | | represent L2The RGB average values of pixel RGB values after norm, i.e. Gaussian smoothing and whole region it
Between Euclidean distance.
Step 5.1.2, carries out local conspicuousness detection in region：
The clutter of cloud, wave and large area has global conspicuousness higher, but because it is distributed on a large scale, it is local notable
Property relative target can be small it is many.Local conspicuousness detection is using Harel et al. based on figure（Graph-based visual
Saliency, GBVS）Conspicuousness detection method, the method protrudes notable portion by the way that the characteristic pattern of Itti et al. is normalized to
Point.The method tends to having prominent edge part to produce highly significant value, therefore right rather than whole object is equably protruded
There is very strong inhibitory action in large-scale cloud, clutter and wave.The method can quickly calculate the local conspicuousness of image, and symbol
Close the biological vision characteristic of the mankind.The local conspicuousness that every bit will be calculated in region uses Sl(i, j) is represented.
Step 5.1.3, calculates comprehensive notable figure：
In order to merge global and local notable figure, the notable figure that will be obtained first is standardized.To definition standard letter
Number N (s):
N (s)=(s-min (s))/(max (s)-min (s))（3）
In formula, s represents the significance value of every bit in region（Global or local）, min (s), max (s) represent area respectively
The minimum of conspicuousness, maximum in domain.
Its fusion of the Gaussian hybrid function pair of 2D is used herein.Function is defined as follows comprehensive notable figure ScFor：
（4）
In formula, N (Sl)、N(Sg) represent respectively according to local, the global saliency value after normalized function N (S) standardization.If
Respectively when local consistent with global conspicuousness, its conspicuousness fraction can be very high after fusion, also illustrates that for the significance value of certain point
It is very notable in the picture.Therefore the comprehensive conspicuousness after merging can protrude the global and local significant doubtful mesh in naval vessel
Mark, further suppresses the global conspicuousness very interference region such as high cloud, wave, completes the preliminary extraction of the doubtful Ship Target that pulls in shore.Tool
When body is implemented, suitable threshold value can automatically be chosen by Otsu methods, as the S at water area midpoint (i, j)cMore than threshold value
When judge that the point belongs to the doubtful Ship Target that pulls in shore.
Step 5.2, using the machine learning method based on multiple features in well-marked target（Doubtful Ship Target）Middle detection warship
Ship.The well-marked target that step 5.1 is detected not necessarily be exactly need detection Ship Target, it is also possible to shipform and greatly
Small similar target, such as island, wave, the cloud of fritter and some other naval targets, it is therefore desirable to carried out to it
Further recognition and verification.Other naval targets such as the profile and texture on naval vessel and island, wave exist more obvious poor
Different, this is also exactly identified the foundation of differentiation to these well-marked targets.Here by extract the texture of well-marked target, geometry and
Shape facility, and naval vessel detection further carried out by SVM classifier.
During specific implementation, those skilled in the art can the texture of sets itself well-marked target, geometry and Shape Feature Extraction
Mode.Embodiment is comprised the following steps that：
Step 5.2.1, texture feature extraction
LBP（Local Binary Pattern, local binary patterns）Be one kind for describing image Local textural feature
Operator, it has the significant advantage such as rotational invariance and gray scale consistency, has for the goal description under different illumination
Good robustness.Herein using the equivalent formulations of the propositions such as Ojala（Uniform Pattern）LBP carries out well-marked target
Textural characteristics are described, and ultimately form the LBP texture feature vectors in whole well-marked target region（58 dimensions）.
Step 5.2.2, Extraction of Geometrical Features
To each well-marked target for extracting, extract respectively as follows：
1）Area:Using the sum of all pixels in well-marked target as target area.
2）The length-width-ratio of minimum enclosed rectangle:It is the important parameter on naval vessel, the major axis of minimum enclosed rectangle is naval vessel mesh
Target main shaft.
3）Rectangular degree:The area that the region contour of well-marked target is surrounded and minimum enclosed rectangle area ratio.
4）Degree of compacting:The girth of well-marked target square and area ratio, describe well-marked target profile complexity.
Geometric properties 4 are tieed up totally, there is provided architectural feature is described.
Step 5.2.3, Shape Feature Extraction
The Shape context SC proposed using Belongie et al.（Shape Contexts）Feature is described, this feature
Ignore the deformation such as rotation, scaling of figure, be a kind of method that simple and robust finds the uniformity between figure.SC is used herein
Feature carries out the shape facility description of well-marked target, there is provided Scale invariant shape SC features, ultimately forms whole well-marked target area
The SC characteristic vectors of domain profile（64 dimensions）.
Step 5.2.4, the Ship Target Detection based on SVM.
After textural characteristics, geometric properties and Shape Feature Extraction is completed, one 126 characteristic vector of dimension is obtained,
Classified using SVM methods.When being realized in Ship Target Detection problem with SVM, it is only necessary to which certain well-marked target is naval vessel
Also it is the class of non-naval vessel two.And before test sample is input into SVM, SVM is entered using substantial amounts of sample data in advance
Row training.Therefore, it can use naval vessel, the class sample of non-naval vessel two in advance（Positive and negative samples）The characteristic vector for extracting 126 dimensions constitutes sample
Eigen storehouse is simultaneously trained to SVM, is test with each well-marked target extracted from the water area of image to be detected
Sample, be input into SVM classifier, you can obtain well-marked target be naval vessel be also non-naval vessel result.
In sum, step 5.1 realizes that the ROI suspicious regions based on many vision notable features detect that step 5.2 is realized many
The integrated monomer naval vessel detection of feature.
Step 6, for the Ship Target that pulls in shore, carries out image segmentation and obtains doubtful mesh using significance and shape information
Mark, extracts naval vessel suspected target multiple feature and detects degree of membership, and the confidence level on naval vessel is belonged to reference to evidential probability detection.
Step 6.1, Image Segmentation is carried out using global significance, obtains initial suspicious region
Region as detection zone of pulling in shore after step 4 sea land is separated, carries out global conspicuousness detection in region.With
Global conspicuousness detection implementation in step 5.1.1 is identical, has been described in detail in steps of 5, and here is omitted.One
As detect after, the global conspicuousness S of most of point (i, j) in region of pulling in shoreg(i, j) is 0, and other are set to category for the point of non-zero value
In initial suspicious region.
Step 6.2, fine Ship Target segmentation is carried out using shipform prior information
In specific implementation, first, Hausdorff distance matchings are carried out to the region being partitioned into using shape information.Afterwards
Judge whether target sticks together, if it has not, being transferred to next step step 6.3, such as occur, target point is carried out using morphology operations
Cut, subsequently into step 6.3.
Step 6.2.1, the suspicious region initial to step 6.1 gained carries out Hausdorff distance matchings
Marginal point in definition template image is represented with T, and the marginal point searched in image is expressed as into E.So, mould
Hausdorff distances between the edge point set E of the edge point set T of project picture and image to be searched can be expressed as:H(T,E)=
Max (h (T, E), h (E, T)), wherein, function| | | | represent L2Norm, i.e. Gaussian smoothing
Euclidean distance between rear pixel RGB values and the RGB average values of whole region.Template be naval vessel stem " V " font structure with
And parallel lines, if its distance is less than empirical value, it is considered as suspected target, otherwise exclude.
Step 6.2.2：Judge whether target sticks together, Target Segmentation is carried out using morphology operations in adhesion.
Specific dividing method is as follows：First, concave point is extracted using Freeman chain codes difference, to prevent over-segmentation, splits string
Chord length must be smaller than certain value, i.e. d<=P/ π, wherein d are distance between two points, and P is girth, this guarantees only splitting apart from d
Less than or equal to the point of the imaginary circular diameter with P as girth.Secondly, the string arc ratio of concave point cannot be less than 1.5.On border between 2 points
, than referred to as string arc ratio, in the case of adhesion, object boundary constitutes arc for the length of arc and the distance between them.Constitute target most
Minimum condition is that arc should be major arc, i.e., it should be greater than 1.5 times of string.Meet above-mentioned two condition d<The string arc ratio of=P/ π and concave point
Cannot be less than 1.5, i.e. target to stick together, concave point can be attached, so as to split to adhesion target, obtain final
Suspicious region.
Step 6.3, the feature of the gained suspicious region of extraction step 6.2 is pulled in shore Ship Target using multiple features constraint detection.
Embodiment extracts girth, area, texture and the chain code feature in region to be detected in remote sensing image, and according to difference
Membership function detect the degree of membership of each feature, give each feature evidence probability ρ by the way of confidence level is distributedArea、ρ3、
ρT、ρL, the degree of membership of result is extracted with reference to different characteristic, calculate the confidence level on the naval vessel that pulls in shore.
Specific embodiment is as follows：
A) area is detected with girth feature degree of membership：ρArea=80*area/perimeter^2;（6）
Wherein area represents imagery coverage, and perimeter represents image girth（80 is constant）
B) textural characteristics degree of membership detection（Third moment）：（7）
S in formulaiIt is pixel gray value, p (si) be region shared by the pixel with the gray value ratio, its average
ForL is the maximum of image intensity value, and acquiescence value is 255.
C) bow feature degree of membership (chain code feature degree of membership) detection：
(4) the difference chain code d of the Freeman chain codes on the border in region to be checked is obtained.
(5) according to d, the angle sequence θ on the border in region to be checked is obtained, if θiIt is the folder at i-th point in angle sequence θ
Angle,
Wherein, n is the number of corner point on curve.The span of i is 0 to n-1, diIt is i-th -1 o'clock to i-th point
Distance, di+1It is i-th point to i+1 point of distance, ziIt is the distance of the i-th -1 point to i+1 point.
(6) degree of membership ρ is obtainedT：（8）
Wherein θmiIt is the angle at i-th point on image edge curve to be checked, θliIt is the folder at i-th point of template curve
Angle.
D) pull in shore to be detected apart from degree of membership：（9）
Wherein Ax+By+C=0 is the line of vector line segment closest with suspected target,（X, y）It is suspected target center of mass point
Coordinate.
According to evidence theory principle, each feature evidence of suspicious region is given by way of distributing confidence level, specific real
Those skilled in the art can as the case may be pre-set confidence level when applying.The area that to detect, texture, shape facility enter
Row combines to calculate the confidence level of suspicious region.Think confidence level be in confidential interval pair as if ship, be incorporated to object set Bs, put
What reliability was in refusal interval is other atural objects, is incorporated to object set BN, confidential interval, refusal is interval, indeterminacy section can be by skill
Art personnel are according to the previously given division of actual conditions.Confidence level PJudgeFormula it is as follows：
（10）
Wherein, RjIt is the confidence level of evidence j.ρjIt is the degree of membership of evidence, if not detecting corresponding evidence, ρjTake
0.J is evidence total number, the confidence level P for finally calculatingJudge.The evidence of embodiment is 4, i.e. J=4, corresponding RjIt is area
Feature evidence confidence level RiArea, textural characteristics evidence confidence level Ri3, chain code feature evidence confidence level RiT, pull in shore to be put apart from evidence
Reliability RiL, respectively such as a）、b）、c）、d）Each feature evidence degree of membership ρ is obtained using corresponding membership functionArea、ρ3、ρT、ρL。
In sum, step 6.1,6.2 realize that the ROI suspicious regions based on many vision notable features are detected, step 6.3 is real
The integrated monomer naval vessel detection of existing multiple features.
Step 5,6 can be carried out parallel.During specific implementation, periodically vector database and image database for control point can also be entered
Row updates.Sample Storehouse is updated according to naval vessel testing result.
Seashore line vector, Image Database update：Because of natural feature change, the reason such as harbour reconstruction is likely to result in coastline change
Change.According to the testing result of step 3.2, if coastline changes, link OPENSTREETMAP increases income geographical information library, under
The coastline data of correspondence block are carried, in storage to database.Periodically downloaded and reminded user to supplement people automatically simultaneously
Work gathered data, realizes that seashore line vector and Image Database update.
Sample Refreshment：Ship Target and other jamming targets are respectively added in positive negative example base according to testing result,
And train renewal grader.
Specific embodiment described herein is only to the spiritual explanation for example of the present invention.Technology neck belonging to of the invention
The technical staff in domain can be made various modifications or supplement to described specific embodiment or be replaced using similar mode
Generation, but without departing from spirit of the invention or surmount scope defined in appended claims.
Claims (6)
1. the optical remote sensing image Ship Detection under a kind of priori geography information is aided in, it is characterised in that including following step
Suddenly：
Step 1, sets up the harbour priori geographical information library related to coastline, including multiple dimensioned harbour image database for control point and
Seashore line vector storehouse；
Step 2, the remote sensing image to be detected after being corrected using RPC parameters, or according to the geography of remote sensing image to be detected
Coordinate, takes corresponding control point image and is matched from harbour image database for control point, the remote sensing image to be detected after being corrected；
Step 3, to correction after remote sensing image to be detected carry out region segmentation extract boundary line, according to seashore line vector storehouse obtain
Seashore line vector, carries out coastline change detection；
Step 4, using the seashore line vector obtained according to seashore line vector storehouse, realizes that extra large land separates, and obtains water area and leans on
Land region domain；
Step 5, for water area, first carries out naval vessel suspected target detection based on many vision significances, then based on multiple features
Machine learning method detects naval vessel in naval vessel suspected target；
It is described to carry out the detection of naval vessel suspected target based on many vision significances and include following sub-step,
Step 5.1.1, the water area after step 4 sea land is separated is carried out global aobvious as detection zone in detection zone
Work property detection, the global conspicuousness at detection zone midpoint (i, j) uses Sg(i, j) is represented,
Step 5.1.2, carries out local conspicuousness detection in detection zone, and the local conspicuousness at detection zone midpoint (i, j) is adopted
Use Sl(i, j) is represented；
Step 5.1.3, the comprehensive notable figure of calculating is as follows,
Definition standard function N (s) is as follows,
N (s)=(s-min (s))/(max (s)-min (s))
In formula, s represents the global or local significance value of every bit in region, and min (s), max (s) are represented in region and shown respectively
Minimum, the maximum of work property；
The comprehensive notable figure S of definitioncIt is as follows,
In formula, N (Sl)、N(Sg) represent respectively according to local, the global saliency value after normalized function N (S) standardization；
Step 6, for region of pulling in shore, the region as detection zone of pulling in shore after first step 4 sea land is separated, in detection zone
Global conspicuousness detection is carried out, initial suspicious region is obtained；Image segmentation is carried out further according to shape information, final doubting is obtained
Like region；Pulled in shore Ship Target using multiple features constraint detection afterwards；
Wherein, the implementation that global conspicuousness detection is carried out in detection zone is as follows,
If detection zone has N number of pixel, the RGB average values in detection zone are as follows,
Wherein, r (i, j), g (i, j), b (i, j) are the color spy at (i, j) place in detection zone on the image to be detected after correcting
Value indicative；
Detection zone internal image is carried out into Gaussian smoothing, the new color feature value r of point (i, j) in detection zone is obtainedG(i,j),
gG(i,j),bG(i,j)；
The global significance at detection zone midpoint (i, j) is expressed as follows,
In formula, | | | | represents L2Norm.
2. the optical remote sensing image Ship Detection under priori geography information is aided according to claim 1, its feature exists
In：Step 3 includes following sub-step,
Step 3.1, the coastline of superposition correspondence remote sensing image geographic coordinate range to be detected, including following sub-step,
Step 3.1.1, the one or more point string in remote sensing image geographic coordinate range to be detected is taken from seashore line vector storehouse
As initial priori line of vector；
Step 3.1.2, takes each point string the first two point and does extended line respectively with most latter two point, and with remote sensing image to be detected
Edge forms intersection point；
Step 3.1.3, initial priori line of vector is added using former and later two intersection points as beginning and end, obtains new coastline
Line of vector；
Step 3.2, detection, including following sub-step are changed to coastline,
Step 3.2.1, to correction after remote sensing image to be detected carry out region segmentation, extract boundary line；
Step 3.2.2, boundary line is matched with step 3.1.3 gained line of vectors, and calculating boundary line overlaps with coastline
Rate；
Step 3.2.3, if coincidence factor reaches corresponding predetermined threshold value, into step 4, otherwise stops flow, points out to carry out seashore
Line updates.
3. the optical remote sensing image Ship Detection under priori geography information is aided according to claim 1, its feature exists
In：In step 6, image segmentation is carried out according to shape information, the implementation for obtaining final suspicious region is as follows,
First, Hausdorff distance matchings are carried out to the region being partitioned into using shape information；Judge whether target occurs afterwards
Adhesion, if it has not, directly being pulled in shore Ship Target using multiple features constraint detection, is such as occurred, and target is carried out using morphology operations
Segmentation, is then pulled in shore Ship Target using multiple features constraint detection.
4. the optical remote sensing image Ship Detection under priori geography information is aided according to claim 1, its feature exists
In：In step 6, pulled in shore Ship Target using multiple features constraint detection, including calculate the naval vessel that pulls in shore confidence level it is as follows,
In formula, RprfIt is the confidence level of evidence prf, ρprfIt is the degree of membership of evidence, J is evidence total number；
According to the advance confidential interval for dividing, refusal be interval, indeterminacy section, it is believed that confidence level be in confidential interval pair as if
Ship, is incorporated to object set Bs, what confidence level was in refusal interval is other atural objects, is incorporated to object set BN。
5. the optical remote sensing image Ship Detection under priori geography information is aided according to claim 4, its feature exists
In：If J=4, according to default area features evidence confidence level RiArea, textural characteristics evidence confidence level Ri3, chain code feature evidence
Confidence level RiT, pull in shore apart from evidence confidence level RiLWith individual features degree of membership ρArea、ρ3、ρT、ρL, calculate the confidence on the naval vessel that pulls in shore
Degree,
A) area and girth feature degree of membership ρAreaDetection is as follows,
ρArea=80*area/perimeter^2
Wherein area represents imagery coverage, and perimeter represents image girth；
B) textural characteristics degree of membership ρ3Detection is as follows,
S in formulalIt is pixel gray value, p (sl) be region shared by the pixel with the gray value ratio, its average isL is the maximum of image intensity value；
C) chain code feature degree of membership ρTDetection is as follows,
(1) the difference chain code d of the Freeman chain codes on the border in region to be checked is obtained；
(2) according to d, the angle sequence θ on the border in region to be checked is obtained, if θeIt is the angle at e-th point in angle sequence θ,
Wherein, n is the number of corner point on curve, and the span of e is 0 to n-1, deFor the e-1 o'clock to e-th point away from
From de+1It is e-th point to the e+1 distance of point, zeIt is the e-1 o'clock to the e+1 distance of point；
(3) degree of membership is obtained
Wherein, θmeIt is the angle at e-th point on image edge curve to be checked, θleIt is the angle at e-th point of template curve；
D) pull in shore to detect as follows from degree of membership away from feature,
Wherein Ax+By+C=0 is the line of vector line segment closest with suspected target, and (x, y) is the seat of suspected target center of mass point
Mark.
6. the optical remote sensing image naval vessel detection side under priori geography information is aided according to claim 1 or 2 or 3 or 4 or 5
Method, it is characterised in that：After step 2 obtains the remote sensing image to be detected after correcting, water body index judgement is carried out, if waiting after correcting
Detection remote sensing image includes more than 99% sea, then be directly entered step 5, otherwise enter step 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410086835.3A CN103886285B (en) | 2014-03-11 | 2014-03-11 | Optical remote sensing image Ship Detection under priori geography information auxiliary |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410086835.3A CN103886285B (en) | 2014-03-11 | 2014-03-11 | Optical remote sensing image Ship Detection under priori geography information auxiliary |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103886285A CN103886285A (en) | 2014-06-25 |
| CN103886285B true CN103886285B (en) | 2017-06-09 |
Family
ID=50955168
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410086835.3A Active CN103886285B (en) | 2014-03-11 | 2014-03-11 | Optical remote sensing image Ship Detection under priori geography information auxiliary |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103886285B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106651880A (en) * | 2016-12-27 | 2017-05-10 | 首都师范大学 | Method for detecting marine moving target of thermal infrared remote sensing image based on multi-feature fusion |
Families Citing this family (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104516950A (en) * | 2014-12-10 | 2015-04-15 | 百度在线网络技术（北京）有限公司 | Inquiring method and device of interest points |
| CN104965199B (en) * | 2015-07-28 | 2017-11-17 | 中国人民解放军海军航空工程学院 | Radar video moving target Fusion Features decision method |
| CN105138992A (en) * | 2015-08-28 | 2015-12-09 | 大连海事大学 | Coastline detection method based on regional active outline model |
| CN105303526B (en) * | 2015-09-17 | 2017-11-07 | 哈尔滨工业大学 | A kind of Ship Target Detection method based on coastline data and spectrum analysis |
| CN105551049B (en) * | 2015-12-28 | 2018-08-03 | 清华大学 | Polarimetric SAR image major port detection method based on water areas of harbor extraction |
| CN105893957B (en) * | 2016-03-30 | 2019-03-22 | 上海交通大学 | View-based access control model lake surface ship detection recognition and tracking method |
| CN105868794B (en) * | 2016-04-19 | 2019-03-12 | 哈尔滨工业大学 | A kind of Ship Target fuzzy recognition method based on ISAR image |
| CN106295503A (en) * | 2016-07-25 | 2017-01-04 | 武汉大学 | The high-resolution remote sensing image Ship Target extracting method of region convolutional neural networks |
| CN107704865A (en) * | 2017-05-09 | 2018-02-16 | 北京航空航天大学 | Fleet Targets Detection based on the extraction of structure forest edge candidate region |
| CN107909002B (en) * | 2017-08-07 | 2021-01-19 | 北京航空航天大学 | Sea-land segmentation method of infrared remote sensing image based on coastline matching |
| CN109726616A (en) * | 2017-10-30 | 2019-05-07 | 中电科海洋信息技术研究院有限公司 | A kind of detection of naval vessel and recognition methods and device |
| CN107977621A (en) * | 2017-11-29 | 2018-05-01 | 淮海工学院 | Shipwreck identification model construction method, device, electronic equipment and storage medium |
| CN108121949B (en) * | 2017-12-04 | 2018-11-23 | 交通运输部规划研究院 | A kind of harbour Ore stockpile recognition methods based on remote sensing scene classification |
| CN108305257B (en) * | 2017-12-27 | 2020-08-04 | 华中科技大学 | Seabed tunnel remote sensing detection positioning method under background thermal radiation constraint |
| CN108288268B (en) * | 2018-01-23 | 2020-07-24 | 华中科技大学 | Method for detecting ship landing in high-resolution remote sensing image based on projection analysis |
| CN108229433B (en) * | 2018-02-01 | 2021-10-26 | 中国电子科技集团公司第十五研究所 | Method for detecting ship landing on shore based on straight-line segment detection and shape characteristics |
| CN108536140B (en) * | 2018-02-26 | 2022-03-25 | 臻迪科技股份有限公司 | Unmanned ship autonomous navigation system and method |
| CN108537116B (en) * | 2018-03-05 | 2021-09-24 | 中国地质大学（武汉） | Multi-scale feature-based coastline secondary type extraction method and system |
| CN109087319B (en) * | 2018-08-17 | 2021-07-02 | 北京华航无线电测量研究所 | Mask manufacturing method and system |
| CN109359654B (en) * | 2018-09-18 | 2021-02-12 | 北京工商大学 | Image segmentation method and system based on frequency tuning global saliency and deep learning |
| CN109815807B (en) * | 2018-12-18 | 2020-12-08 | 浙江大学 | Method for detecting ship in shore based on edge line analysis and aggregation channel characteristics |
| CN109727265B (en) * | 2018-12-28 | 2022-03-22 | 北京行易道科技有限公司 | Target segmentation method and device |
| CN110021024B (en) * | 2019-03-14 | 2023-12-26 | 华南理工大学 | Image segmentation method based on LBP and chain code technology |
| CN110428425B (en) * | 2019-07-02 | 2023-03-07 | 中国航空工业集团公司雷华电子技术研究所 | Sea-land separation method of SAR image based on coastline vector data |
| CN110807424B (en) * | 2019-11-01 | 2024-02-02 | 深圳市科卫泰实业发展有限公司 | Port ship comparison method based on aerial image |
| CN112329613B (en) * | 2020-11-03 | 2021-10-15 | 珠海大横琴科技发展有限公司 | Sea wave influence launching detection method and device, electronic equipment and storage medium |
| CN114022778B (en) * | 2021-10-25 | 2023-04-07 | 电子科技大学 | SAR (synthetic Aperture Radar) berthing ship detection method based on significance CNN (CNN) |
| CN114742854B (en) * | 2022-04-02 | 2023-06-02 | 西安电子科技大学 | SAR image sea-land segmentation method based on scene priori knowledge and region combination |
| CN114764821B (en) * | 2022-04-19 | 2023-07-28 | 极米科技股份有限公司 | Moving object detection method, moving object detection device, electronic equipment and storage medium |
| CN114998740B (en) * | 2022-06-13 | 2023-07-21 | 中国电子科技集团公司第五十四研究所 | Airport linear feature extraction method based on line segment distribution |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096824A (en) * | 2011-02-18 | 2011-06-15 | 复旦大学 | Multi-spectral image ship detection method based on selective visual attention mechanism |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6778469B1 (en) * | 2003-02-12 | 2004-08-17 | Science Applications International Corporation | Harbor fence |
-
2014
- 2014-03-11 CN CN201410086835.3A patent/CN103886285B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096824A (en) * | 2011-02-18 | 2011-06-15 | 复旦大学 | Multi-spectral image ship detection method based on selective visual attention mechanism |
Non-Patent Citations (3)
| Title |
|---|
| Context modeling combined with motion analysis for moving ship detection in port surveillance;Xinfeng Bao 等;《Journal of Electronic Imaging》;20131231;第22卷(第4期);第1-16页 * |
| SAR图像港口区域舰船检测新方法;陈琪 等;《电子与信息学报》;20110930;第33卷(第9期);第2132页-2137页 * |
| 复杂港口背景下舰船目标检测方法研究;况小琴;《中国优秀硕士学位论文全文数据库信息科技辑》;20120715;摘要、第4页-第48页 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106651880A (en) * | 2016-12-27 | 2017-05-10 | 首都师范大学 | Method for detecting marine moving target of thermal infrared remote sensing image based on multi-feature fusion |
| CN106651880B (en) * | 2016-12-27 | 2020-04-14 | 首都师范大学 | Offshore moving target detection method based on multi-feature fusion thermal infrared remote sensing image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103886285A (en) | 2014-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103886285B (en) | Optical remote sensing image Ship Detection under priori geography information auxiliary | |
| CN105654091B (en) | Sea-surface target detection method and device | |
| CN104951799B (en) | A kind of SAR remote sensing image oil spilling detection recognition method | |
| Chaudhuri et al. | Automatic building detection from high-resolution satellite images based on morphology and internal gray variance | |
| CN104217196B (en) | A kind of remote sensing image circle oil tank automatic testing method | |
| CN102867196B (en) | Based on the complicated sea remote sensing image Ship Detection of Gist feature learning | |
| CN104463877B (en) | A kind of water front method for registering based on radar image Yu electronic chart information | |
| CN104361582B (en) | Method of detecting flood disaster changes through object-level high-resolution SAR (synthetic aperture radar) images | |
| CN109427055B (en) | Remote sensing image sea surface ship detection method based on visual attention mechanism and information entropy | |
| CN108694393A (en) | A kind of certificate image text area extraction method based on depth convolution | |
| CN107122777A (en) | A kind of vehicle analysis system and analysis method based on video file | |
| CN109598241A (en) | Satellite image marine vessel recognition methods based on Faster R-CNN | |
| Marinelli et al. | A novel approach to 3-D change detection in multitemporal LiDAR data acquired in forest areas | |
| CN106803070A (en) | A kind of port area Ship Target change detecting method based on remote sensing images | |
| CN103218831A (en) | Video moving target classification and identification method based on outline constraint | |
| CN107154044B (en) | Chinese food image segmentation method | |
| CN103020975A (en) | Wharf and ship segmentation method combining multi-source remote sensing image characteristics | |
| CN106709426A (en) | Ship target detection method based on infrared remote sensing image | |
| CN103778627A (en) | Sea oil spill detection method based on SAR image | |
| Kalkan et al. | Comparison of support vector machine and object based classification methods for coastline detection | |
| CN107704865A (en) | Fleet Targets Detection based on the extraction of structure forest edge candidate region | |
| CN109978848A (en) | Method based on hard exudate in multiple light courcess color constancy model inspection eye fundus image | |
| CN106780504B (en) | It is a kind of based on distance keep level set beef image in carry on the back long flesh automatic division method | |
| CN102175993A (en) | Radar scene matching feature reference map preparation method based on satellite SAR (synthetic aperture radar) images | |
| Mello et al. | Automatic image segmentation of old topographic maps and floor plans |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |