CN106096586A - The extra large background modeling of high resolution remote sensing ocean imagery and the method and system of suppression - Google Patents

Abstract

The present invention provides the extra large background modeling of a kind of high resolution remote sensing ocean imagery and the method for suppression, and wherein, described method includes: cutting step, classifying step, calculation procedure, modeling procedure and suppression step.The present invention also provides for the extra large background modeling of a kind of high resolution remote sensing ocean imagery and the system of suppression.The technical scheme that the present invention provides, by marine background first suppresses this mode detected target the most again, can improve detection accuracy significantly and reduce false alarm rate.

Description

The extra large background modeling of high resolution remote sensing ocean imagery and the method and system of suppression

Technical field

The present invention relates to technical field of image processing, particularly relate to the extra large background modeling of a kind of high resolution remote sensing ocean imagery Method and system with suppression.

Background technology

In optical satellite remote sensing ocean imagery, not only include target information, but also contain target ambient ocean The sea water background information on surface, sea water background is affected by multiple natural causes such as sea wind-force, wind direction, surge, ambient humidities, Display form in the picture is varied, such as stormy waves, surge, the spray, whirlpool, foam etc..Distant in order to effectively detect Target in sense ocean imagery, to sea water background modeling and to carry out the suppression of sea water background be a kind of preferably approach.

At present, the sea water background modeling for optical satellite remote sensing ocean imagery proposed in prior art and the side of suppression Method mainly has following several:

One, method based on mathematical morphology: the method uses structural motif to find the repeatability district, space that characteristic is similar Territory, i.e. sea water background area, and these regions are removed from image, thus extract target area.Utilize mathematical morphology Method when detecting, its result depends on the selection of structural motif, and how selecting preferable structural motif is the method Key, be also difficult point.

Two, method based on image space gray-scale statistical distributed model: first the method selects one can describe ocean The probabilistic model (such as Gauss model, K distributed model etc.) of remote sensing images Sea background spatial gradation statistical distribution characteristic, secondly root According to the spatial gradation of Sea background, the parameter of this distributed model is estimated, finally utilize this sea water spatial context gray scale to unite Meter distributed model, carries out model probability judgement to each pixel gray value in the Marine remote sensing image containing target, thus It is partitioned into Ship Target region.The method for Sea background tranquiler in the case of, it is the most suitable that sea clutter can select Distributed model carry out matching.But, for the image that background clutter is complex, its distributed model often cannot be the most true Fixed, thus affect the accuracy of target area segmentation.

Three, method based on fractal model: the method carries out fractional dimension first with fractal theory and technology to image Multi-resolution decomposition, secondly both is split in the difference of fractal dimension according to Sea background region and target area, Thus detect and extract target area.But by the shadow of background complexity, random noise, image quality etc. in real image Ringing, single yardstick or constant fractal dimension are difficult to distinguish Sea background and target area.

Four, the method for view-based access control model significance model: the method first passes through feature extraction, significance calculates and significantly schemes Merge and generate visual saliency map, in the notable figure generated, secondly find more significantly visual object, and this region is carried Take out, it is achieved the detection of target area.The method introduces various features, it is possible to by target area and Sea background preferably Separated, but owing to feature is more, their the most good assessment method of rationally selection.

In sum, the most existing background model based on image space intensity profile, when Sea background is complex Shi Wufa describes the background clutter in image well, it is impossible to intend the Sea background of satellite remote sensing ocean imagery well Close, so that the object detection method of existing satellite remote sensing ocean imagery based on spatial gradation distributed model exists false-alarm The problem that rate is high, detection accuracy is low.

Summary of the invention

In view of this, it is an object of the invention to provide extra large background modeling and the suppression of a kind of high resolution remote sensing ocean imagery Method and system, it is intended to solve the mesh for satellite remote sensing ocean imagery based on spatial gradation distributed model in prior art There is the problem that false alarm rate is high, detection accuracy is low in mark detection method.

The present invention proposes the extra large background modeling of a kind of high resolution remote sensing ocean imagery and the method for suppression, it is characterised in that Described method includes:

Cutting step: remote sensing images are carried out pretreatment and piecemeal cutting, and each subimage block cut out is carried out Rough sort, selects the subimage block set that can analyze ocean imagery；

Classifying step: all subimage blocks in the subimage block set that can analyze ocean imagery are categorized further, as pure Sea water image block subclass and non-full sea water image block subclass；

Calculation procedure: each subimage block in the subimage block set that can analyze ocean imagery is carried out two-dimensional discrete Fourier transform, obtains the spectrogram of correspondence, and is calculated the amplitude spectrogram of correspondence by spectrogram；

Modeling procedure: determine one piece of subimage block to be analyzed in the subimage block set can analyze ocean imagery, with this Centered by subimage block to be analyzed, by finding out owning around this subimage block to be analyzed in full sea water image block subclass Full sea water image block builds sea background magnitude spectrum gaussian probability illustraton of model；

Suppression step: utilize described amplitude spectrum gaussian probability illustraton of model to build sea background rejects trap, and this is treated point Analysis subimage block carries out sea background suppression.

Preferably, described method also includes:

Circulation step: determine next block subimage block to be analyzed, and walk by repeating described modeling procedure and described suppression Suddenly carry out new subimage block to be analyzed is carried out sea background suppression, until completing the extra large background suppression of all subimage blocks.

Preferably, described suppression step specifically includes:

Utilize the described sea background magnitude spectrum gaussian probability illustraton of model that described modeling procedure constructs, calculate this son to be analyzed The filtering mahalanobis distance figure of image block；

Described filtering mahalanobis distance G-Design is utilized to obtain the preferable sea background rejects trap of this subimage block to be analyzed；

Described preferable sea background rejects trap is utilized to design Gauss sea background rejects trap further；

Utilize described Gauss sea background rejects trap that this subimage block to be analyzed carries out frequency domain filtering, and by after filtering Spectral image by two-dimensional inverse Fourier transform obtain through background suppression subimage block.

On the other hand, the present invention also provides for the extra large background modeling of a kind of high resolution remote sensing ocean imagery and the system of suppression, Described system includes:

Cutting module, for remote sensing images being carried out pretreatment and piecemeal cutting, and each subimage block that will cut out Carry out rough sort, select the subimage block set that can analyze ocean imagery；

Sort module, for classifying further to all subimage blocks in the subimage block set that can analyze ocean imagery For full sea water image block subclass and non-full sea water image block subclass；

Computing module, for carrying out two dimension to each subimage block in the subimage block set that can analyze ocean imagery Discrete Fourier transform (DFT), obtains the spectrogram of correspondence, and is calculated the amplitude spectrogram of correspondence by spectrogram；

MBM, for determining one piece of subimage block to be analyzed in the subimage block set can analyze ocean imagery, Centered by this subimage block to be analyzed, by finding out around this subimage block to be analyzed in full sea water image block subclass All full sea water image blocks build sea background magnitude spectrum gaussian probability illustraton of model；

Suppression module, is used for utilizing described amplitude spectrum gaussian probability illustraton of model to build sea background rejects trap, and to this Subimage block to be analyzed carries out sea background suppression.

Preferably, described system also includes:

Loop module, is used for determining next block subimage block to be analyzed, and by repeating described modeling procedure and described pressing down Step processed carries out sea background suppression to new subimage block to be analyzed, until completing the extra large background suppression of all subimage blocks.

Preferably, described suppression module specifically for:

Utilize the described sea background magnitude spectrum gaussian probability illustraton of model that described MBM constructs, calculate this son to be analyzed The filtering mahalanobis distance figure of image block；

Described filtering mahalanobis distance G-Design is utilized to obtain the preferable sea background rejects trap of this subimage block to be analyzed；

Described preferable sea background rejects trap is utilized to design Gauss sea background rejects trap further；

Utilize described Gauss sea background rejects trap that this subimage block to be analyzed carries out frequency domain filtering, and by after filtering Spectral image by two-dimensional inverse Fourier transform obtain through background suppression subimage block.

The technical scheme that the present invention provides preferably solves the sea of the satellite remote sensing ocean imagery that prior art is used Face background model causes Marine remote sensing image target owing to cannot describe and suppress Sea background clutter in image well The problem that testing result is unstable, the present invention is this by first suppress marine background to detect target the most again Mode, can improve detection accuracy significantly and reduce false alarm rate.

Accompanying drawing explanation

Fig. 1 is the extra large background modeling method flow with suppression of high resolution remote sensing ocean imagery in an embodiment of the present invention Figure；

Fig. 2 a is the subimage block exemplary plot to be analyzed provided in an embodiment of the present invention；

Fig. 2 b is amplitude spectrogram corresponding with Fig. 2 a in an embodiment of the present invention；

Fig. 3 is the concrete detail flowchart of the suppression step shown in Fig. 1 in an embodiment of the present invention；

Fig. 4 a is the preferable sea background rejects trap schematic diagram provided in an embodiment of the present invention；

Fig. 4 b is the regional area enlarged drawing of Fig. 4 a in an embodiment of the present invention；

Fig. 4 c is the graphics of Fig. 4 b in an embodiment of the present invention；

Fig. 5 a is the Gauss sea background rejects trap schematic diagram provided in an embodiment of the present invention；

Fig. 5 b is the regional area enlarged drawing of Fig. 5 a in an embodiment of the present invention；

Fig. 5 c is the graphics of Fig. 5 b in an embodiment of the present invention；

Fig. 6 is to use the Gauss sea background rejects trap shown in Fig. 5 a to shown in Fig. 2 a in an embodiment of the present invention Subimage block to be analyzed be filtered after result schematic diagram；

Fig. 7 is the extra large background modeling system 10 with suppression of high resolution remote sensing ocean imagery in an embodiment of the present invention Internal structure schematic diagram.

Detailed description of the invention

In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, right The present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, and It is not used in the restriction present invention.

In order to solve the problem that prior art exists, the sea of the optical satellite high resolution remote sensing ocean imagery that the present invention proposes Background modeling with the main thought of the method for suppression is: remote sensing ocean imagery is set up the frequency domain statistics mould of the extra large background of local Type, sets up background rejects trap on this basis and suppresses sea background.The process of implementing includes: cutting step: right Remote sensing images carry out pretreatment and piecemeal cutting, and each subimage block cut out is carried out rough sort, select and can analyze sea The subimage block set of ocean image；Classifying step: to all subimage blocks in the subimage block set that can analyze ocean imagery It is categorized further, as full sea water image block subclass and non-full sea water image block subclass；Calculation procedure: to analyzing ocean Each subimage block in the subimage block set of image carries out two dimension discrete fourier transform, obtains the spectrogram of correspondence, And the amplitude spectrogram of correspondence is calculated by spectrogram；Modeling procedure: determine in the subimage block set can analyze ocean imagery One piece of subimage block to be analyzed, centered by this subimage block to be analyzed, by finding out this in full sea water image block subclass All full sea water image blocks around subimage block to be analyzed build sea background magnitude spectrum gaussian probability illustraton of model；Suppression step Rapid: to utilize described amplitude spectrum gaussian probability illustraton of model to build sea background rejects trap, and this subimage block to be analyzed is carried out Sea background suppression.

The extra large background modeling of a kind of high resolution remote sensing ocean imagery that the present invention provides and the method for suppression are by first to sea Ocean background carries out this mode suppressing to detect target the most again, can improve detection accuracy and reduction significantly False alarm rate.

Below by the method to the extra large background modeling of a kind of high resolution remote sensing ocean imagery provided by the present invention with suppression It is described in detail.

Refer to Fig. 1, for the extra large background modeling of high resolution remote sensing ocean imagery in an embodiment of the present invention and suppression Method flow diagram.

In step sl, cutting step: remote sensing images are carried out pretreatment and piecemeal cutting, and each height that will cut out Image block carries out rough sort, selects the subimage block set that can analyze ocean imagery.

In the present embodiment, remote sensing images are carried out the pretreatment such as sea land segmentation, spissatus region detection；After pretreatment Remote sensing images carry out piecemeal cutting；By each subimage block rough sort of cutting out for ocean imagery subgraph set Sa can be analyzed And ocean imagery subgraph set Sn can not be analyzed.

In the present embodiment, the extra large land of remote sensing images is split refer to by the land in remote sensing images and region, island from Splitting in ocean imagery, so detection to targets in ocean just has only to carry out in sea area.In present embodiment In, it is possible to use Global Sea Surface water front data base coordinates the methods such as automatic study to process sea land segmentation problem, it addition, obtaining In Marine remote sensing image, often there are some sea areas by spissatus covering, are cannot in the spissatus region of these remote sensing images Carry out targets in ocean detection.Can be by the spissatus district in remote sensing images by the algorithm of more already present spissatus detections Territory effectively detects.Will not be described in great detail about sea land dividing method and spissatus detection method.In the present invention, it is assumed that it is the most logical Cross the segmentation of land, sea and spissatus detection method has been partitioned into the land in pending remote sensing images and island from ocean imagery Region, and spissatus region.

In the present embodiment, owing to the breadth of High spatial resolution remote sensing is the biggest, the scope of covering is shown Greatly, pixel count is many, data volume is big.In Marine remote sensing image, marine background has a similarity at regional area, and at a distance of than Two regions farther out, the similarity of background can diminish.In order to set up marine background model accurately, effectively suppress background, Detection target, needs large format remote sensing images are carried out piecemeal process.The size of general sub-block elects 256 256 or 512 as 512 sizes are proper.When image block, the uniform piecemeal in no overlap region can be carried out, it is also possible to carry out there is crossover region The uniform piecemeal in territory.For the detection of target of being more convenient for, use between horizontal neighboring sub-block and between vertical neighboring sub-block It is the most suitable that the mode of 50% of all overlapping carries out image block.

In the present embodiment, in a certain subimage of piecemeal, if land area and spissatus region occupied area it With when exceeding this subimage gross area certain proportion r1, this subgraph referred to as can not analyze ocean subgraph.All can not analyze ocean The image collection of subgraph composition referred to as can not analyze ocean imagery subgraph set Sn.In actual use, above-mentioned proportionality coefficient R1 typically can need to elect as 50%, or 75% according to precision.Subimage to all piecemeals, rejects owning in set Sn After subimage block, the set of sub-images that remaining subimage block is constituted referred to as can analyze ocean imagery subgraph set Sa.At this In bright, follow-up all methods are both for analyzing what the image in ocean imagery subgraph set Sa was carried out.

In step s 2, classifying step: all subimage blocks in the subimage block set that can analyze ocean imagery are entered One step is categorized as full sea water image block subclass and non-full sea water image block subclass.

In the present embodiment, to each piece of further rough segmentation of subimage can analyzed in ocean imagery subgraph set Sa Class, can analyze ocean imagery subgraph set Sa and be divided into two subclass: i.e. full sea water image block subclass Sw and non-pure Sea water image block subclass So.

In the present embodiment, in step s 2, the subimage block can analyzed in ocean imagery subgraph set Sa is probably The image of full sea water, it is also possible to comprise the image block of the targets such as naval vessel.In the present invention, the marine background owing to being set up is The statistical model of a kind of spectrum domain, if comprising the target that occupied area is bigger in subimage block, then can affect the ocean imagery back of the body The accuracy of scape modeling；And if target occupied area is less, then the accuracy of ocean imagery background modeling will not be produced significantly Impact.Target occupied area described herein is less refers to that target occupied area in subimage block is less than subimage in the present invention The 1% of the block gross area.Have at present some technology can preferably complete the subimage block in set Sa is carried out full sea water with The rough sort of non-full sea water, such as sorting technique etc. based on the non-unimodal detection of rectangular histogram, can analyze ocean imagery from above-mentioned Subgraph set Sa preferably distinguishes the subimage block wherein containing bigger target.These contain the subimage block of bigger target Constitute non-full sea water image block subclass So, gather in Sa except other image block of So then constitutes full sea water image block Set Sw.Concrete division methods repeats no more.

It is pointed out that in above-mentioned affiliated full sea water image block subclass Sw, may containing some target areas relatively Little non-full sea water subimage block.Non-full sea water image block therein then generally comprises: containing part land or island Image block, containing bigger target to be detected image block and include the image block etc. in the spissatus district of part.

In the later step of the present invention, it is assumed that obtained and can analyze ocean imagery subgraph set Sa, and this set Two subclass: full sea water image block subclass Sw and non-full sea water image block subclass So.

In step s3, calculation procedure: to each subimage block in the subimage block set that can analyze ocean imagery Carry out two dimension discrete fourier transform, obtain the spectrogram of correspondence, and calculated the amplitude spectrogram of correspondence by spectrogram.

In the present embodiment, to each piece of piecemeal subimage f in block image set Sa_kCarry out two dimension Fourier to become Change, obtain each piece of spectrogram F corresponding to subimage block_k, and the amplitude spectrogram A calculated by spectrogram_k。

The size assuming each piece of subimage in block image set Sa is N N, and kth subimage therein is f_k(x, y), (x y) is the pixel space coordinates of image, 0≤x≤N-1,0≤y≤N-1.Then its two-dimensional discrete space Fourier The spectrogram F of leaf transformation_k(u, v), 0≤u≤N-1,0≤v≤N-1 is calculated by formula (1):

$F_k(u,v)=\frac{1}{N}\underset{u=0}{\overset{N-1}{\Σ}}\underset{v=0}{\overset{N-1}{\Σ}}{f}_k(x,y)e^{-j\frac{2\mathrm{\π}}{N}(ux+vy)}---\left(1\right)$

Corresponding amplitude spectrum A_k(u, v) is calculated by formula (2):

$A_k(u,v)=\left|F_k(u,v)\right|=\sqrt{R_k^2(u,v)+I_k^2(u,v)}---\left(2\right)$

Wherein, R_k(u, v), I_k(u v) is respectively F_k(u, the figure of real part composition v) and the figure of imaginary part composition.

In step s 4, modeling procedure: determine one piece of son to be analyzed in the subimage block set can analyze ocean imagery Image block, centered by this subimage block to be analyzed, by finding out this subimage to be analyzed in full sea water image block subclass All full sea water image blocks around block build sea background magnitude spectrum gaussian probability illustraton of model.

In the present embodiment, in image collection Sa, determine one piece of image block f to be analyzed_i.With image block f_iIn for The heart, finds out at f in full sea water image subset closes Sw_iNeighbouring all full sea water image blocks, constitute subclass Sw_i, use Sw_iIn Image block build sea background amplitude spectrum gaussian probability illustraton of model G_i。

Assume that one piece of image to be analyzed block in block image set Sa is f_i(x, y), its size is N N, its two dimension The spectrogram of discrete space Fourier transform is F_i(u, v), amplitude spectrum is A_i(u, v), 0≤u≤N-1,0≤v≤N-1.

Fig. 2 a is the subimage block example to be analyzed that the embodiment of the present invention is given, and Fig. 2 b is the amplitude spectrum of its correspondence Figure.

With image block f_iCentered by, find out at f in full sea water image subset Sw_iNeighbouring all full sea water image blocks, structure Become subclass Sw_i.F described above_iNear refer to image block f_iCentered by a rectangle or circle shaped neighborhood region, cut with aforementioned Length of side N of the image block cut is unit, and the radius of neighbourhood can choose 2N to 5N.Select circle shaped neighborhood region in embodiments of the present invention, and The radius of neighbourhood elects 3N as.Assume Sw_iContaining J block full sea water image block in subclass, wherein jth full sea water subimage block is f_j (x, y), 0≤x≤N-1,0≤y≤N-1, the spectrogram of its two-dimensional discrete spatial Fourier transform is F_j(u, v), amplitude spectrogram For A_j(u, v), 0≤u≤N-1,0≤v≤N-1.

Further, full sea water image block subclass Sw is utilized_iIn all amplitude spectrogram A_j(u v), builds image to be analyzed f_i(x, y) the amplitude spectrum gaussian probability illustraton of model G of the extra large background in marine site, place_i, the probability density function of this model is represented by:

$p_i(u,v)=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_i(u,v)}\exp (-\frac{{\[A_i(u,v)-m_i(u,v)\]}^2}{2{\[{\mathrm{\σ}}_i(u,v)\]}^2})---\left(3\right)$

Wherein,

$m_i(u,v)=\frac{1}{J}\underset{j=1}{\overset{J}{\Σ}}{A}_j(u,v)---\left(4\right)$

${\[{\mathrm{\σ}}_i(u,v)\]}^2=\frac{1}{J}\underset{j=1}{\overset{J}{\Σ}}{\[A_j(u,v)-m_i(u,v)\]}^2---\left(5\right)$

Above-mentioned m_i(u, v) with [σ_i(u,v)]²It is image to be analyzed block f respectively_i(x, y) near full sea water image block subset Close Sw_iIn the average figure of image block amplitude spectrum and variogram, claim m_i(u, v) with [σ_i(u,v)]²For image to be analyzed block f_i(x, Y) extra large background magnitude spectrum gaussian probability illustraton of model G_i.By above-mentioned formula (4) and formula (5), at each frequency of image block (u, v) place establishes the Gauss distribution statistical model of an extra large background to rate point, and its average and variance are respectively m_i(u, v) with [σ_i (u,v)]²。

In step s 5, suppress step: utilize described amplitude spectrum gaussian probability illustraton of model to build sea background rejects trap, And this subimage block to be analyzed is carried out sea background suppression.

In the present embodiment, image block f is used_iAmplitude spectrogram A_iAnd the amplitude spectrum probabilistic model G of sea background_iStructure Build sea background rejects trap B_i, use B_iImage is carried out background suppression.

Specifically, as it is shown on figure 3, step S5 comprises the following steps again:

In step s 51, the described sea background magnitude spectrum gaussian probability illustraton of model that described modeling procedure constructs, meter are utilized Calculate the filtering mahalanobis distance figure of this subimage block to be analyzed.

In the present embodiment, the image block f that step S4 calculates is utilized_iBackground probability model G_i, calculate figure to be analyzed As block f_iFiltering mahalanobis distance figure Q (u, v).

Filtering mahalanobis distance figure Q (u, computing formula v) is formula (6):

$Q(u,v)=\frac{{\[A_i(u,v)-m_i(u,v)\]}^2}{{\[{\mathrm{\σ}}_i(u,v)\]}^2}---\left(6\right)$

In step S52, described filtering mahalanobis distance G-Design is utilized to obtain the preferable sea back of the body of this subimage block to be analyzed Scape rejects trap.

In the present embodiment, step S51 calculated image to be analyzed block f is utilized_iFiltering mahalanobis distance figure Q (u, v), design obtains image to be analyzed block f_iPreferable sea background rejects trap Ω_i(u,v)。

In the present embodiment, image block f_iPreferable sea background rejects trap Ω_i(u v) can be calculated by formula (7) Arrive,

${\mathrm{\Ω}}_i(u,v)=\left\{\begin{array}{cc}1,& Q(u,v)>T\\ 0,& Q(u,v)\≤T\end{array}\right.---\left(7\right)$

Wherein, the T in formula (7) is constant, referred to as a decision threshold.The value of T typically can with value for interval [1, 9] real number between.In embodiments of the present invention, value T=9.

In step S53, described preferable sea background rejects trap is utilized to design the background suppression filtering of Gauss sea further Device.

In the present embodiment, the preferable sea background rejects trap Ω designed is utilized_i(u v), designs height further This sea background rejects trap B_i(u,v)。

In the present embodiment, preferable sea background rejects trap Ω_i(u v) can produce " ring " effect in actual applications Should, thus affect the inhibition of marine background.The present invention utilizes the gaussian kernel function preferable sea background to obtaining in step S52 Rejects trap smooths, and obtains overcoming the Gauss sea background rejects trap B of " ring " effect_i(u,v).Concrete meter Calculate shown in formula such as formula (8):

B_i(u, v)=Ω_i(u,v)*H_d(u,v) (8)

Wherein, in formula (8), the symbol * in formula represents the two-dimensional convolution computing in Digital Image Processing.H_d(u, V) represent with (u, v) centered by the Gaussian smooth function template that size is (2d+1) × (2d+1), d is positive integer.At this In inventive embodiments, take d=1.

Gaussian smooth function template H_d(u, v) can be calculated by formula (9):

$H_d(u,v)=\frac{1}{W}{e}^{-\frac{u^2+v^2}{2{\mathrm{\λ}}^2}},-d\≤u\≤d,-d\≤v\≤d---\left(9\right)$

Wherein, in formula (9), (u, v) value is scope-d≤u≤d, the integer in-d≤v≤d；Parameter lambda is to control The constant of smoothed intensity, typically can take λ=d.In embodiments of the present invention, λ=1 is taken；W be one relevant with smooth function Normalized constant, is calculated by formula (10):

$W=\underset{u=-d}{\overset{d}{\Σ}}\underset{v=-d}{\overset{d}{\Σ}}{e}^{-\frac{u^2+v^2}{2{\mathrm{\λ}}^2}},-d\≤u\≤d,-d\≤v\≤d---\left(10\right)$

In step S54, utilize described Gauss sea background rejects trap that this subimage block to be analyzed is carried out frequency domain filter Ripple, and filtered spectral image is obtained the subimage block through background suppression by two-dimensional inverse Fourier transform.

In the present embodiment, sea background rejects trap B is utilized_iCarry out image to be analyzed block f_iFrequency domain filtering, and will Filtered spectral image passes through two-dimensional inverse Fourier transform, obtains the subimage block f through background suppression_ti。

In the present embodiment, as previously shown, subimage block f to be analyzed_i(x, Fourier spectrum figure y) is F_i(u, v), Gauss sea background rejects trap is B_i(u, v), shown in background suppression filtering computing formula such as formula (11):

F_ti(u, v)=F_i(u,v)B_i(u,v) (11)

Calculate the subimage block f after background suppression_ti(x, shown in the such as formula of two-dimensional inverse Fourier transform y) (12):

$f_{ti}(x,y)=\frac{1}{N}\underset{x=0}{\overset{N-1}{\Σ}}\underset{y=0}{\overset{N-1}{\Σ}}{F}_{ti}(u,v)e^{j\frac{2\mathrm{\π}}{N}(ux+vy)}---\left(12\right)$

Fig. 4, Fig. 5, Fig. 6 give the example that the embodiment of the present invention provides.Wherein, Fig. 4 a is the image block calculated The preferable sea water background rejects trap in Fig. 2 a marine site, Fig. 4 b is the regional area enlarged drawing at the white box place in Fig. 4 a, Fig. 4 c is the graphics of Fig. 4 b.Fig. 5 a is the Gauss sea water background rejects trap in the image block Fig. 2 a marine site calculated, Fig. 5 b Being the regional area enlarged drawing at white box place in Fig. 5 a, Fig. 5 c is Fig. 5 b graphics.Fig. 6 is to use Fig. 5 a wave filter pair Fig. 2 a is transformed into the result figure of spatial domain after carrying out background suppression filtering.

In addition, the side of extra large background modeling and the suppression of a kind of high-resolution satellite remote sensing ocean imagery that the present invention provides Method, further comprises:

Circulation step: determine next block subimage block to be analyzed, and walk by repeating described modeling procedure and described suppression Suddenly carry out new subimage block to be analyzed is carried out sea background suppression, until completing the extra large background suppression of all subimage blocks.

The extra large background modeling of a kind of high resolution remote sensing ocean imagery that the present invention provides and the method for suppression, preferably solve The Sea background model of the satellite remote sensing ocean imagery that prior art is used is owing to cannot describe well and suppress image In Sea background clutter and cause the unstable problem of Marine remote sensing image object detection results, the present invention is by first to ocean Background carries out this mode suppressing to detect target the most again, can improve detection accuracy significantly and reduce void Alert rate.

What the specific embodiment of the invention also provided for the extra large background modeling of a kind of high resolution remote sensing ocean imagery and suppression is System 10, specifically includes that

Cutting module 11, for remote sensing images being carried out pretreatment and piecemeal cutting, and each subimage that will cut out Block carries out rough sort, selects the subimage block set that can analyze ocean imagery；

Sort module 12, for dividing further all subimage blocks in the subimage block set that can analyze ocean imagery Class is full sea water image block subclass and non-full sea water image block subclass；

Computing module 13, for carrying out two to each subimage block in the subimage block set that can analyze ocean imagery Dimension discrete Fourier transform (DFT), obtains the spectrogram of correspondence, and is calculated the amplitude spectrogram of correspondence by spectrogram；

MBM 14, for determining one piece of subimage to be analyzed in the subimage block set can analyze ocean imagery Block, centered by this subimage block to be analyzed, by finding out this subimage block to be analyzed week in full sea water image block subclass The all full sea water image blocks enclosed build sea background magnitude spectrum gaussian probability illustraton of model；

Suppression module 15, is used for utilizing described amplitude spectrum gaussian probability illustraton of model to build sea background rejects trap, and right This subimage block to be analyzed carries out sea background suppression.

The extra large background modeling of a kind of high resolution remote sensing ocean imagery that the present invention provides and the system 10 of suppression, by the most right Marine background carries out this mode suppressing to detect target the most again, can improve detection accuracy and fall significantly Low false alarm rate.

Refer to Fig. 7, show in an embodiment of the present invention the extra large background modeling of high resolution remote sensing ocean imagery and press down The structural representation of the system 10 of system.

In the present embodiment, the extra large background modeling of high resolution remote sensing ocean imagery and the system 10 of suppression, mainly include Cutting module 11, sort module 12, computing module 13, MBM 14 and suppression module 15.

Cutting module 11, for remote sensing images being carried out pretreatment and piecemeal cutting, and each subimage that will cut out Block carries out rough sort, selects the subimage block set that can analyze ocean imagery.

In the present embodiment, what the concrete cutting method of cutting module 11 referred in abovementioned steps S1 relevant retouches State, the most do not do repeated description at this.

Sort module 12, for dividing further all subimage blocks in the subimage block set that can analyze ocean imagery Class is full sea water image block subclass and non-full sea water image block subclass.

In the present embodiment, the concrete sorting technique of sort module 12 refers to the associated description in abovementioned steps S2, Repeated description is not the most done at this.

Computing module 13, for carrying out two to each subimage block in the subimage block set that can analyze ocean imagery Dimension discrete Fourier transform (DFT), obtains the spectrogram of correspondence, and is calculated the amplitude spectrogram of correspondence by spectrogram.

In the present embodiment, the circular of computing module 13 refers to the associated description in abovementioned steps S3, Repeated description is not the most done at this.

MBM 14, for determining one piece of subimage to be analyzed in the subimage block set can analyze ocean imagery Block, centered by this subimage block to be analyzed, by finding out this subimage block to be analyzed week in full sea water image block subclass The all full sea water image blocks enclosed build sea background magnitude spectrum gaussian probability illustraton of model.

In the present embodiment, the concrete modeling method of MBM 14 refers to the associated description in abovementioned steps S4, Repeated description is not the most done at this.

Suppression module 15, is used for utilizing described amplitude spectrum gaussian probability illustraton of model to build sea background rejects trap, and right This subimage block to be analyzed carries out sea background suppression.

In the present embodiment, suppression module 15 specifically for:

Utilize the described sea background magnitude spectrum gaussian probability illustraton of model that described MBM constructs, calculate this son to be analyzed The filtering mahalanobis distance figure of image block；

Described filtering mahalanobis distance G-Design is utilized to obtain the preferable sea background rejects trap of this subimage block to be analyzed；

Described preferable sea background rejects trap is utilized to design Gauss sea background rejects trap further；

Utilize described Gauss sea background rejects trap that this subimage block to be analyzed carries out frequency domain filtering, and by after filtering Spectral image by two-dimensional inverse Fourier transform obtain through background suppression subimage block.

In the present embodiment, the concrete suppressing method of suppression module 15 refers to the associated description in abovementioned steps S5, Repeated description is not the most done at this.

In addition, the system of extra large background modeling and the suppression of a kind of high resolution remote sensing ocean imagery that the present invention provides 10, further comprise:

Loop module, is used for determining next block subimage block to be analyzed, and by repeating described modeling procedure and described pressing down Step processed carries out sea background suppression to new subimage block to be analyzed, until completing the extra large background suppression of all subimage blocks.

The extra large background modeling of a kind of high resolution remote sensing ocean imagery that the present invention provides and the system 10 of suppression, preferably solve The Sea background model of the optical satellite remote sensing ocean imagery that prior art of having determined is used is owing to cannot describe well and press down Sea background clutter in imaged and cause the unstable problem of Marine remote sensing image object detection results, the present invention is by first Marine background is suppressed this mode detected target the most again, can improve significantly detection accuracy and Reduce false alarm rate.

It should be noted that in above-described embodiment, included unit is to carry out dividing according to function logic, But it is not limited to above-mentioned division, as long as being capable of corresponding function；It addition, the specific name of each functional unit is also Only to facilitate mutually distinguish, it is not limited to protection scope of the present invention.

It addition, one of ordinary skill in the art will appreciate that all or part of step realizing in the various embodiments described above method The program that can be by completes to instruct relevant hardware, and corresponding program can be stored in an embodied on computer readable storage and be situated between In matter, described storage medium, such as ROM/RAM, disk or CD etc..

The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all essences in the present invention Any amendment, equivalent and the improvement etc. made within god and principle, should be included within the scope of the present invention.

Claims (6)

1. the extra large background modeling of a high resolution remote sensing ocean imagery and the method for suppression, it is characterised in that described method includes:

Cutting step: remote sensing images are carried out pretreatment and piecemeal cutting, and each subimage block cut out is carried out rough segmentation Class, selects the subimage block set that can analyze ocean imagery；

Classifying step: all subimage blocks in the subimage block set that can analyze ocean imagery are categorized further, as full sea water Image block subclass and non-full sea water image block subclass；

Calculation procedure: each subimage block in the subimage block set that can analyze ocean imagery is carried out two-dimensional discrete Fourier Leaf transformation, obtains the spectrogram of correspondence, and is calculated the amplitude spectrogram of correspondence by spectrogram；

Modeling procedure: determine one piece of subimage block to be analyzed in the subimage block set can analyze ocean imagery, treats point with this Centered by analysis subimage block, by finding out all pure sea around this subimage block to be analyzed in full sea water image block subclass Water images block builds sea background magnitude spectrum gaussian probability illustraton of model；

Suppression step: utilize described amplitude spectrum gaussian probability illustraton of model to build sea background rejects trap, and to this son to be analyzed Image block carries out sea background suppression.

2. the method for the extra large background modeling of high resolution remote sensing ocean imagery as claimed in claim 1 and suppression, it is characterised in that Described method also includes:

Circulation step: determine next block subimage block to be analyzed, and come by repeating described modeling procedure and described suppression step New subimage block to be analyzed is carried out sea background suppression, until completing the extra large background suppression of all subimage blocks.

3. the method for the extra large background modeling of high resolution remote sensing ocean imagery as claimed in claim 1 and suppression, it is characterised in that Described suppression step specifically includes:

Utilize the described sea background magnitude spectrum gaussian probability illustraton of model that described modeling procedure constructs, calculate this subimage to be analyzed The filtering mahalanobis distance figure of block；

Described filtering mahalanobis distance G-Design is utilized to obtain the preferable sea background rejects trap of this subimage block to be analyzed；

Described preferable sea background rejects trap is utilized to design Gauss sea background rejects trap further；

Utilize described Gauss sea background rejects trap that this subimage block to be analyzed is carried out frequency domain filtering, and by filtered frequency Spectrogram picture obtains the subimage block through background suppression by two-dimensional inverse Fourier transform.

4. the extra large background modeling of a high resolution remote sensing ocean imagery and the system of suppression, it is characterised in that described system includes:

Cutting module, for remote sensing images carry out pretreatment and piecemeal cutting, and is carried out each subimage block cut out Rough sort, selects the subimage block set that can analyze ocean imagery；

Sort module, for being categorized further, as pure to all subimage blocks in the subimage block set that can analyze ocean imagery Sea water image block subclass and non-full sea water image block subclass；

Computing module, for carrying out two-dimensional discrete to each subimage block in the subimage block set that can analyze ocean imagery Fourier transform, obtains the spectrogram of correspondence, and is calculated the amplitude spectrogram of correspondence by spectrogram；

MBM, for determining one piece of subimage block to be analyzed, with this in the subimage block set can analyze ocean imagery Centered by subimage block to be analyzed, by finding out owning around this subimage block to be analyzed in full sea water image block subclass Full sea water image block builds sea background magnitude spectrum gaussian probability illustraton of model；

Suppression module, is used for utilizing described amplitude spectrum gaussian probability illustraton of model to build sea background rejects trap, and treats this point Analysis subimage block carries out sea background suppression.

5. the system of the extra large background modeling of high resolution remote sensing ocean imagery as claimed in claim 4 and suppression, it is characterised in that Described system also includes:

Loop module, is used for determining next block subimage block to be analyzed, and walks by repeating described modeling procedure and described suppression Suddenly carry out new subimage block to be analyzed is carried out sea background suppression, until completing the extra large background suppression of all subimage blocks.

6. the system of the extra large background modeling of high resolution remote sensing ocean imagery as claimed in claim 4 and suppression, it is characterised in that Described suppression module specifically for:

Utilize the described sea background magnitude spectrum gaussian probability illustraton of model that described MBM constructs, calculate this subimage to be analyzed The filtering mahalanobis distance figure of block；

Described filtering mahalanobis distance G-Design is utilized to obtain the preferable sea background rejects trap of this subimage block to be analyzed；

Described preferable sea background rejects trap is utilized to design Gauss sea background rejects trap further；

Utilize described Gauss sea background rejects trap that this subimage block to be analyzed is carried out frequency domain filtering, and by filtered frequency Spectrogram picture obtains the subimage block through background suppression by two-dimensional inverse Fourier transform.