CN103065136A - Method for recognizing collaborative target in SAR (Synthetic Aperture Radar) image based on visual attention mechanism - Google Patents

Abstract

The invention provides a method for recognizing a collaborative target in an SAR (Synthetic Aperture Radar) image based on a visual attention mechanism. The method comprises the following steps of: establishing the scale spatial presentation of an SAR image; extracting a low-layer visual feature map, carrying out semantic translation on a visual task, translating the visual task, decomposing the visual feature map, calculating the weight of each visual feature map, and generating a visual task saliency map relevant to the visual task; selecting a salient target region in the SAR image by utilizing the visual task saliency map to realize the separation of a foreground and a background of the target region in the SAR image; learning samples of various known prototype targets through a collaborative learning algorithm to obtain features and feature spaces of the known prototype targets and generate a known prototype target knowledge base; and recognizing collaboration of targets in the salient target region: recognizing the targets in the salient target region of the selected SAR image based on the known prototype target knowledge base by utilizing a collaboration pattern recognition order parameter dynamic iteration process.

Description

The collaborative target identification method of a kind of SAR image based on vision noticing mechanism

Technical field

The present invention relates to a kind of SAR image object recognition technology field, particularly the collaborative target identification method of a kind of SAR image based on vision noticing mechanism.

Background technology

Synthetic-aperture radar (Synthetic Aperture Radar, SAR) be a kind of active Coherent Imaging RADAR that is operated in microwave region, has remote, quick, large-area high resolution imaging ability, since coming out, the fields such as military surveillance, land mapping, resource detection have been widely used in.The SAR image militarily is mainly used in scouting, the aspects such as detection of dynamic of supervision, Hitting Effect Evaluation and important goal, can obtain the informations such as type, distribution situation and geographic coordinate of multiple military target enemy's depth zone in wide scope.Along with early warning plane, scout the fast development of the associated equipment of the carry SAR imaging device such as unmanned plane, the SAR image has become an important information source of modern battlefield information reconnaissance and surveillance.Information processing demand to the SAR image will be the significant capability of following Information Handling System.From the SAR image of magnanimity, automatically detect the important goal of paying close attention to, and it is identified, become a key issue of SAR image processing field.

Detection and Identification to target in the SAR image are important means of the judgement of realize target attribute and classification of type.Generally can be by artificial interpretation or automated characterization and pattern analysis, the judgement of realize target attribute or classification of type.Artificial interpretation speed is slower, for growing military SAR image, can not satisfy the demand.Common automated characterization and pattern analysis solution route comprise statistical method, structural approach and spectral method etc.For the existence to target judges, these methods often need whole image-region is searched for, processed, but the content of in fact being concerned about only accounts for a part of area very little in the image usually.This overall treatment method had both increased computational complexity, had increased the weight of again the analysis difficulty.In addition, all kinds of algorithms all are from image, do not consider visual task to the directive function of graphical analysis, lack in the research to contact from bottom to top data-driven and the from top to bottom tie of knowledge guidance.All kinds of algorithm intelligent processing method levels are low, differ far away with the various functions of the cognitive process of the mankind in identifying, and its testing result is also unsatisfactory.

Vision selection attention (Visual Selective Attention) mechanism has well solved this problem.Vision selection attention mechanism is a human inherent attribute, it is the essential characteristic that the people processes external environment condition visual information, a gordian technique of particular region of interest that also to be the people select from the bulk information of external world's input is the human visual system with the Nature reciprocal process in the essential characteristic of the processing visual information that develops through very long and complicated evolution.The people can find rapidly zone or the object of " interested " and " meaningful " in magnanimity visual information, in the face of a complex scene or image, people's vision always selects rapidly a few zone to carry out priority processing, and selected zone is called focus-of-attention (Focus of Attention).Vision attention can carry out selectivity processing to each scene areas with different order and dynamics, thereby has avoided calculating waste, has reduced the analysis difficulty.The vision selection attention computation model is that the mathematics of simulating human vision noticing mechanism is realized, for piece image, visual attention computation model can be selected marking area or the object in the complex scene fast.Visual attention computation model is studied widely and is used in vision fields such as figure image intensifying, image segmentation, compression of images, target identifications.

Summary of the invention

Goal of the invention: technical matters to be solved by this invention is for the deficiencies in the prior art, provides a kind of SAR image based on vision noticing mechanism collaborative target identification method.

In order to solve the problems of the technologies described above, the invention discloses the collaborative target identification method of a kind of SAR image based on vision noticing mechanism, may further comprise the steps:

Step 1, the metric space of setting up the SAR image represents; Extract the Low Level Vision characteristic pattern, comprise and extract brightness, contour feature and direction character;

Step 2 is carried out semantic conversion to visual task, and visual task is changed, is decomposed into the visual signature figure corresponding with visual signature in the step 1, calculates the weight of each visual signature figure, generates the visual task relevant with visual task and significantly schemes;

Step 3 utilizes visual task significantly to scheme to select the well-marked target in the SAR image regional, realizes the prospect of SAR objective area in image and separating of background;

Step 4, the study of known prototype target is learnt the sample of various known prototype targets by the Cooperative Study algorithm, obtains known prototype target signature and known prototype target signature space, generates known prototype object knowledge storehouse;

Step 5, the collaborative identification of target in the well-marked target zone: based on known prototype object knowledge storehouse, utilize Synergetic Pattern Recognition order parameter dynamic iteration process, to identifying of target in the SAR image well-marked target zone of selecting.

Step 1 of the present invention comprises that the gaussian pyramid yardstick of setting up the SAR image represents, extracts contour feature and the direction character of different scale images, utilizes the poor operator of central peripheral to obtain brightness vision characteristic pattern, profile visual signature figure and directional vision characteristic pattern.

In the step 2 of the present invention, by normalization brightness vision characteristic pattern, profile visual signature figure and directional vision characteristic pattern, the visual task of semantic expressiveness is converted to the visual signature figure corresponding with the Low Level Vision feature, and then significantly schemes according to the weight generation visual task relevant with visual task of each visual signature figure.

In the step 3 of the present invention, significantly scheme based on visual task, select successively the well-marked target in the SAR image regional according to inhibition of return mechanism.

In the step 4 of the present invention, the proper vector of known prototype target sample is calculated in utilization based on the Cooperative Study algorithm of independent component analysis, make up known prototype target sample feature space, generate known prototype object knowledge storehouse, and calculate the adjoint vector of each known prototype target sample proper vector.

Step 5 of the present invention comprises: the SAR image well-marked target zone of selecting is mapped in the target signature space, generate well-marked target provincial characteristics vector, calculate the initial order parameter in well-marked target zone, if the maximal value of the initial order parameter in well-marked target zone greater than setting threshold, is carried out the identification of well-marked target zone and output recognition result by order parameter dynamic iteration equation; Order parameter dynamic iteration equation is:

${\stackrel{\·}{\mathrm{\ξ}}}_k={\mathrm{\λ}}_k{\mathrm{\ξ}}_k-B\underset{k^{\′}\≠k}{\mathrm{\Σ}}{\mathrm{\ξ}}_{k^{\′}}^2{\mathrm{\ξ}}_k-C\left(\underset{k^{\′}=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{\ξ}}_{k^{\′}}^2\right){\mathrm{\ξ}}_k.$

If the maximal value of the initial order parameter in well-marked target zone less than setting threshold, is learnt unknown object by the Cooperative Study algorithm based on independent component analysis again, and the result that will learn dynamically is increased in the known prototype object knowledge storehouse.

The present invention studies the object detection and recognition relevant with visual task in the SAR image with visual attention computation model with Synergetic Algorithm for Pattern Recognition, can realize according to the visual search task, select fast the well-marked target zone relevant with task in the SAR image, and can identify the target in the well-marked target zone.Simultaneously, the method also has the on-line study ability, and the unknown object in the well-marked target zone in the SAR image is learnt, and expands the object knowledge storehouse.

Beneficial effect: the invention has the advantages that:

1) based on SAR image imaging mechanism, SAR image Low Level Vision feature extracting method is proposed, effectively represent the data characteristic of SAR image;

2) merge low layer view data feature and visual search task, the visual task that generates fusion image data and visual task is significantly schemed, and selects fast well-marked target zone relevant with visual task in the SAR image;

3) by the study of Synergetic Pattern Recognition learning algorithm realization to known prototype target, make up the known target knowledge base;

4) by the dynamic process of order parameter in the Synergetic Algorithm for Pattern Recognition, realize the identification to target in the well-marked target zone, and possess the on-line study ability to unknown object in the well-marked target zone.

Description of drawings

Below in conjunction with the drawings and specific embodiments the present invention is done further to specify, above-mentioned and/or otherwise advantage of the present invention will become apparent.

Fig. 1 is the SAR image object recognition methods process flow diagram based on vision noticing mechanism.

Fig. 2 is the collaborative identification process figure in SAR image well-marked target zone.

Fig. 3 is the SAR image that embodiment contains target to be identified.

Fig. 4 A ~ Fig. 4 E is the learning sample of embodiment five class prototype battlebuses.

Fig. 5 is the well-marked target zone of selecting among the embodiment.

Embodiment

As shown in Figure 1, comprise two parts, the one, the visual task of fusion image data and visual task significantly schemes to generate and the marking area detection, and the 2nd, to study and the well-marked target identification of known mode.

The visual task of fusion image data and visual task significantly schemes to generate and the marking area detection, and the metric space of model SAR image represents, extracts brightness, profile and direction character presentation video information; Next is the conversion to the semantic visual task, the characteristic pattern corresponding with certain visual signature that extracts changed, is decomposed into to visual task, calculate the weight of different characteristic figure, the visual task that the generation task is relevant is significantly schemed, and then the well-marked target that detects in the SAR image is regional, realizes the foreground/background separation of SAR objective area in image.

The study of known models and well-marked target identification to study and the well-marked target identification division of known models, are adopted Synergetic Algorithm for Pattern Recognition, by the study to prototype target training sample, set up prototype target image feature space, calculate the prototype target signature; At last, the well-marked target provincial characteristics is calculated in the well-marked target zone and the target signature space that utilize visual attention computation model to select, and calculate the initial order parameter in well-marked target zone, realize identification to target in the well-marked target zone by order parameter dynamic process.

The present invention specifically may further comprise the steps:

Step 1, the metric space of setting up the SAR image represents; Extract the Low Level Vision characteristic pattern, comprise and extract brightness, contour feature and direction character.

The metric space that the present invention at first utilizes the dimensional Gaussian pyramid model to obtain image represents, by the different scale space representation of image, " central authorities-periphery the is poor " operator by a linearity calculates the Low Level Vision feature." central authorities-periphery is poor " operator is by poor foundation the between high resolving power yardstick and the low resolution yardstick, uses symbol Expression.

By extracting brightness vision characteristic pattern, profile visual signature figure and directional vision characteristic pattern in the SAR image, as the early vision feature in the visual attention computation model, and then computation vision is noted remarkable figure.

Remember that a width of cloth gray scale SAR image is I _SAR, I _SARBe used for producing gaussian pyramid yardstick image I _SAR(σ), wherein σ is scale factor, general value 1,2,3,4, and namely 4 grades of yardsticks represent.

The brightness vision characteristic pattern

The brightness vision characteristic pattern is produced by luminance contrast, by calculating I _SAR(c, s) can represent the brightness vision characteristic pattern.

Wherein c is the high resolving power scale factor, common value 1,2,3, and s is the low resolution scale factor, usually value 4,5,6.Image under the different scale is realized consistance by interpolation, and carries out " central authorities-periphery is poor " and calculate.

Profile visual signature figure

SAR image outline visual signature figure is made of image medium-high frequency message part.At first utilize two-dimentional Teager wave filter to extract the SAR image outline of different scale space representation, remember E _SAR(σ) be contour images under the different scale space representation, then profile visual signature figure is expressed as:

The directional vision characteristic pattern

The V1 district is the front end of Vision information processing, and its mathematical model represents with the Gabor wave filter usually.The Gabor function is the sine and cosine functions of Gaussian function modulation, consists of accordingly its even small echo and strange small echo, and it is the unique function that can obtain spatial domain and frequency-domain combined uncertainty principle lower limit, and good directional selectivity is arranged.Formula (3) is the mathematic(al) representation of two-dimensional Gabor filter.

$G(x,y)=\frac{1}{2\mathrm{\π\α\β}}\exp \{-\mathrm{\π}[\frac{{(x-x_0)}^2}{{\mathrm{\α}}^2}+\frac{{(y-y_0)}^2}{{\mathrm{\β}}^2}\left]\right\}\exp \{i{\mathrm{\ξ}}_{0}x+{\mathrm{\υ}}_{0}y]\}---\left(3\right)$

(x wherein ₀, y ₀) be receptive field centre coordinate in the spatial domain, x ₀Be horizontal ordinate, y ₀Be ordinate; (ξ ₀, υ ₀) be the optimal spatial frequency of wave filter on frequency domain, ξ ₀Be real part, υ ₀Be imaginary part.α and β are respectively the standard deviations of Gaussian function on x and the y direction of principal axis.The present invention gets the Gabor wave filter of four direction and exports as the directional vision characteristic pattern:

N is constant 4, and i represents different directions, and value is 0,1,2,3.

Utilize Gabor pyramid O (σ, θ), can be from image I _SARObtain local direction information, wherein σ is scale factor, θ ∈ 0 °, and 45 °, 90 °, 135 ° }.By the calculating of local direction contrast, directional vision characteristic pattern O (c, s, θ) is encoded into one group:

O(c,s,θ)=|O(c,θ)-O(s,θ)| （4）

Step 2 is carried out semantic conversion to visual task, and visual task is changed, is decomposed into the visual signature figure corresponding with visual signature in the step 1, calculates the weight of each visual signature figure, generates the visual task relevant with visual task and significantly schemes.

Various early sign figure have different dynamic ranges and extraction mechanism because all characteristic patterns are combined, the stronger object of conspicuousness in minority figure only, may by noise or in large spirogram the less object of conspicuousness cover.Therefore, the present invention proposes a characteristic pattern normalization operator N (), to strengthen the less characteristic pattern in remarkable peak, have a large amount of significantly characteristic patterns at peak and weaken.To each characteristic pattern, the operation of this operator comprises: 1) this characteristic pattern of normalization to [0 ..., 1] and in the scope, depend on the amplitude difference of feature with elimination; 2) calculate all local greatly averages except Global maximum

3) use

Take advantage of this characteristic pattern.

Only consider that local maximum can make normalization operator N () that significant zone in the characteristic pattern is compared, and ignore homogeneous area.The difference of Global maximum and all local maximum averages has reflected the difference between most interested zone and average area-of-interest.If this difference is very large, most interested zone will highlight, if this difference is less, shows not contain any zone with peculiar property in the characteristic pattern.

Characteristic pattern is combined into 3 characteristic remarkable descriptions, and namely the gray feature conspicuousness is described

The contour feature conspicuousness is described

Describe with the direction character conspicuousness

Be shown below.

$\tilde{I}=\underset{c}{\mathrm{\⊕}}\underset{s}{\mathrm{\⊕}}N\left(I_{\mathrm{SAR}}(c,s)\right)$

$\tilde{E}=\underset{c}{\mathrm{\⊕}}\underset{s}{\mathrm{\⊕}}N(E_{\mathrm{SAR}}(c,s))---\left(6\right)$

$\tilde{O}=\underset{\mathrm{\θ}}{\mathrm{\Σ}}N(\underset{c}{\mathrm{\⊕}}\underset{s}{\mathrm{\⊕}}N(O(c,s,\mathrm{\θ})))---(7)$

Symbol in the formula

Expression pointwise summation.

Be to embody visual task to the impact of the remarkable figure of vision attention, the introducing task remarkable figure concept of being correlated with.Visual task is generally provided by semantic concept, and in the practical application, task is multifarious, such as search for traffic pattern, river, specific objective etc. in a width of cloth SAR image.The diversity of visual task has increased great difficulty for the main significantly figure of the relevant vision of Task calculates.The present invention is directed to the target in the SAR image, visual task be decomposed into one corresponding with the Low Level Vision feature of extracting, by increasing the weight of this characteristic pattern, generate visual task and significantly scheme, be described as:

$S=\frac{1}{3}({\mathrm{\λ}}_{1}N\widetilde{\left(I\right)}+{\mathrm{\λ}}_{2}N\widetilde{\left(E\right)}+{\mathrm{\λ}}_{3}N\widetilde{\left(O\right)})---\left(8\right)$

λ ₁, λ ₂, λ ₃Be respectively that gray feature is significantly schemed, contour feature is significantly schemed and the remarkable figure weight of direction character, span is 0 ~ 1, satisfies simultaneously λ ₁+ λ ₂+ λ ₃=1.For example, visual task is the white object in the search SAR image, then increases the weight of brightness vision characteristic pattern; If visual task is the square target in the searching image, then increase the weight of contour feature figure; If visual task is the linear target in the searching image, then the weight of augment direction visual signature figure.

In the step 3, utilize visual task significantly to scheme to select the well-marked target in the SAR image regional, realize the prospect of SAR objective area in image and separating of background; Select the well-marked target in the SAR image regional according to the inhibition of return method.

After calculating the remarkable figure of visual task by step 2, the point of usually choosing gray-scale value maximum among the remarkable figure of visual task is blinkpunkt first, centered by this point, goes out first well-marked target with the rectangular selection of former SAR image size 1/16 regional.After first well-marked target zone is selected, the gray-scale value of blinkpunkt first is made as zero, the point of again selecting gray-scale value maximum among the remarkable figure of visual task is blinkpunkt for the second time, adopts the rectangular selection of former SAR image size 1/16 to go out second well-marked target zone.Iterative cycles is selected the well-marked target zone in the former SAR image one by one, generally selects number of times in 20 times.

Step 4, the study of known prototype target is learnt the sample of various known prototype targets by the Cooperative Study algorithm, obtains known prototype target signature and known prototype target signature space, generates known prototype object knowledge storehouse.

To known prototype target sample, calculate the proper vector of known prototype target sample based on the Cooperative Study algorithm of independent component analysis, make up known prototype target sample feature space, generate known prototype object knowledge storehouse, and calculate the adjoint vector of each known prototype target sample proper vector, concrete steps are as follows:

1) X is training sample set, and size is m * N, and wherein N is the number of training set mesarcs target sample, and m is the dimension of prototype target image column vector by rows;

Be the average of training sample set, try to achieve the data of centralization

2) process the correlativity of removing between the input signal by albefaction, reach simultaneously normalized purpose, namely allow

Take advantage of a matrix M to get: Wherein

Λ is by covariance matrix R _XxFront n the diagonal matrix that maximum eigenwert consists of, the row of E are corresponding proper vectors.

3) a linear transformation W by normalized signal Z tries to achieve independent entry

Made up the independent image space of prototype target image by independent entry.And calculate transformation matrix W ^T

4) by

Can be in the hope of corresponding to the projection coefficient of each prototype target image in the independent image space, namely prototype clarification of objective vector forms prototype target signature space U by the prototype target feature vector, has namely made up prototype object knowledge storehouse.

5) obtain simultaneously, the adjoint vector of the prototype target feature vector of prototype target feature vector by finding the solution prototype target feature vector pseudoinverse.

Step 5, the collaborative identification of target in the well-marked target zone: based on known prototype object knowledge storehouse, utilize Synergetic Pattern Recognition order parameter dynamic iteration process, to identifying of target in the SAR image well-marked target zone of selecting.

The Synergetic Pattern Recognition process can be understood as the process of some order parameter competitions.Target to be identified in the SAR image is made as pattern q to be identified, constructs a dynamic process, makes q enter into a pattern v of all prototype patterns through an intermediateness q (t) (wherein, t is time factor) _kThis process can be described as q (0) → q (t) → v _k, the equation that satisfies pattern-recognition is:

$\stackrel{\·}{q}=\underset{k}{\mathrm{\Σ}}{\mathrm{\λ}}_k{v}_k\left(v_k^{+}q\right)-B\underset{k\≠k^{\′}}{\mathrm{\Σ}}{\left(v_{k^{\′}}^{+}q\right)}^2\left(v_k^{+}q\right)v_k-C\left(q^{+}q\right)q+F---\left(9\right)$

Wherein, q is the state vector take input pattern q (0) as initial value; λ _kBe attention parameters, only have when it for timing, pattern just can be identified, and works as λ _kDuring complete equating, namely be called the balance attention parameters, otherwise, uneven attention parameters be called.Attention parameters λ among the present invention _k=1, q is pattern to be identified, q ⁺Quadrature adjoint vector for q; v _kBe prototype pattern;

Be v _kThe quadrature adjoint vector; F is fluctuating force; B and C are prescribed coefficient, are realizing it being that value is 1 usually.

Collaborative recognition methods can be described as the dynamic process of a potential function V, ignores F and transient in the formula (9), and its potential function expression formula is:

$V=-\frac{1}{2}\underset{k=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{\λ}}_k{\left(v_k^{+}q\right)}^2+\frac{B}{4}\underset{k\≠k^{\′}}{\mathrm{\Σ}}{\left(v_k^{+}\right)}^2{\left(v_{k^{\′}}^{+}\right)}^2+\frac{C}{4}{\left(\underset{k=1}{\overset{M}{\mathrm{\Σ}}}{\left(v_k^{+}q\right)}^2\right)}^2---\left(10\right)$

Wherein, M is number of modes to be identified, and k ≠ k represents two different patterns.

If the maximal value of the initial order parameter in well-marked target zone greater than setting threshold, is carried out the identification of well-marked target zone and output recognition result by order parameter dynamic iteration equation, the introducing of order parameter can make the expression of cooperative kinetics simplify.Definition order parameter

Corresponding order parameter kinetics equation is:

${\stackrel{\·}{\mathrm{\ξ}}}_k={\mathrm{\λ}}_k{\mathrm{\ξ}}_k-B\underset{k^{\′}\≠k}{\mathrm{\Σ}}{\mathrm{\ξ}}_{k^{\′}}^2{\mathrm{\ξ}}_k-C\left(\underset{k^{\′}=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{\ξ}}_{k^{\′}}^2\right){\mathrm{\ξ}}_k---\left(11\right)$

Based on the SAR salient region of image identifying of Synergetic Pattern Recognition as shown in Figure 2.To the SAR salient region of image of selecting, be mapped in the prototype target signature space, generate the marking area proper vector, calculate the initial order parameter of marking area with prototype target adjoint vector, if the maximal value of the initial order parameter of marking area greater than setting threshold, identifies and exports recognition result by order parameter dynamic process to marking area.If the maximal value of the initial order parameter of marking area was not learnt before less than setting threshold (be marking area proper vector and arbitrary prototype target of having learnt levy the degree of correlation all very little), the target in this marking area being described.For new unknown pattern, the Cooperative Study algorithm based on independent component analysis that reuses in the step 4 is learnt, and the result that will learn is increased in the prototype object knowledge storehouse.

Embodiment

By an object lesson implementation procedure of the present invention is described.

Selection contains the SAR image of some targets to be identified as input picture, and as shown in Figure 3, size is 1024 * 1024 pixels, and visual task is the battlebus target in the search SAR image.

Described according to step 1,4 grades of metric spaces setting up the SAR image represent, calculate the brightness vision characteristic pattern I of SAR image by " central authorities-periphery is poor " between high resolving power scalogram picture and the low resolution scalogram picture _SAR(1,3) and I _SAR(2,4), profile visual signature figure E _SAR(1,3) and E _SAR(2,4), directional vision characteristic pattern O (1,3) and O (2,4).

Described according to step 2, at first calculate brightness by brightness figure, contour feature figure and direction character figure and significantly scheme

Contour feature is significantly schemed

Significantly scheme with direction character

Analyze again visual task " the battlebus target in the search SAR image ", because battlebus shows well-regulated edge in image, the visual task of search battlebus is converted to the target of well-regulated direction in the searching image, therefore, increase directional vision characteristic pattern weight among the remarkable figure of vision attention, λ is set ₁=0.2, λ ₂=0.2, λ ₃=0.6.Calculating visual task significantly schemes

$S=\frac{1}{3}(0.2\×N\widetilde{\left(I\right)}+0.2\×N\widetilde{\left(E\right)}+0.6\×N\widetilde{\left(O\right)}).$

Described according to step 3, the brightness value of significantly scheming each pixel among the S according to visual task selects visual task significantly to scheme significant point among the S from big to small successively, the control blinkpunkt shifts 13 times in the present embodiment, namely selects visual task significantly to scheme front 13 points of pixel value maximum among the S.Centered by this 13 each point, select marking area with one 100 * 50 rectangle frame, be respectively marking area 1 ~ marking area 13, as shown in Figure 5.

Described according to step 4, known five class prototype battlebus target training samples as shown in Figure 4, every width of cloth prototype battlebus image is 256 * 256 pixel sizes.Utilize the independent component analysis method described in the step 4 that it is learnt.At first be column vector by rows with each prototype battlebus sample conversion, size is that 65536 * 1,5 prototype compositions of sample are the training sample set of 65536 * 5 sizes, and the average of calculation training sample set is utilized each step in the computer program calculation procedure 4.Obtain prototype target feature vector v _k, made up prototype target signature space U, than passing through to calculate prototype target feature vector v _kPseudoinverse obtain prototype target adjoint vector

Described according to step 5, to 13 well-marked target zones selecting, the well-marked target region projection in prototype target signature space U, is calculated respectively the order parameter ξ in well-marked target zone ₁~ ξ _L3, and obtain recognition result by formula (11).Recognition result to marking area 1 is battlebus D, recognition result to marking area 2 is battlebus C, recognition result to marking area 3 is battlebus C, recognition result to marking area 4 is battlebus A, recognition result to marking area zone 5 is battlebus A, recognition result to marking area 6 is battlebus B, recognition result to marking area 7 is battlebus E, recognition result to marking area 8 is battlebus D, recognition result to marking area 9 is battlebus C, is battlebus B to the recognition result of marking area 10, is battlebus C to the recognition result of marking area 11, recognition result to marking area 12 is battlebus E, is battlebus A to the recognition result of marking area 13.

The present invention proposes a kind of SAR image well-marked target based on vision noticing mechanism and detects and recognition methods, and the effect of invention can be summarised as:

1) take the visual performance of the mankind in study, memory, attention, selection and identifying as background, makes up the algorithm flow that is integrated from the on-line study of the study of prototype target, area-of-interest selection and identification, unknown pattern;

2) merge visual task and SAR image Low Level Vision feature, the attention that the generation task is relevant is significantly schemed, and can select fast area-of-interest relevant with task in the image;

3) the prototype target signature effectively be learnt and be expressed to Synergetic Algorithm for Pattern Recognition can, and the target in the marking area is effectively identified, and simultaneously, possesses the on-line study ability, realizes the dynamic learning to unknown pattern;

4) the present invention can be used for the fields such as Intelligent treatment, multi-source image fusion of Artillery Operational Commanding Information system Image Intelligence, improves the Image Intelligence treatment effeciency.

The invention provides the collaborative target identification method of a kind of SAR image based on vision noticing mechanism; method and the approach of this technical scheme of specific implementation are a lot; the above only is preferred implementation of the present invention; should be understood that; for those skilled in the art; under the prerequisite that does not break away from the principle of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.In the present embodiment not clear and definite each ingredient all available prior art realized.

Claims (6)

1. the collaborative target identification method of the SAR image based on vision noticing mechanism is characterized in that, may further comprise the steps:

Step 1, the metric space of setting up the SAR image represents; Extract the Low Level Vision characteristic pattern, comprise and extract brightness, contour feature and direction character;

Step 2 is carried out semantic conversion to visual task, and visual task is changed, is decomposed into the visual signature figure corresponding with visual signature in the step 1, calculates the weight of each visual signature figure, generates the visual task relevant with visual task and significantly schemes;

Step 3 utilizes visual task significantly to scheme to select the well-marked target in the SAR image regional, realizes the prospect of SAR objective area in image and separating of background;

Step 4, the study of known prototype target is learnt the sample of various known prototype targets by the Cooperative Study algorithm, obtains known prototype target signature and known prototype target signature space, generates known prototype object knowledge storehouse;

Step 5, the collaborative identification of target in the well-marked target zone: based on known prototype object knowledge storehouse, utilize Synergetic Pattern Recognition order parameter dynamic iteration process, to identifying of target in the SAR image well-marked target zone of selecting.

2. a kind of SAR image based on vision noticing mechanism according to claim 1 is worked in coordination with target identification method, it is characterized in that, step 1 comprises that the gaussian pyramid yardstick of setting up the SAR image represents, extract contour feature and the direction character of different scale images, utilize the poor operator of central peripheral to obtain brightness vision characteristic pattern, profile visual signature figure and directional vision characteristic pattern.

3. a kind of SAR image based on vision noticing mechanism according to claim 2 is worked in coordination with target identification method, it is characterized in that, in the step 2, by normalization brightness vision characteristic pattern, profile visual signature figure and directional vision characteristic pattern, the visual task of semantic expressiveness is converted to the visual signature figure corresponding with the Low Level Vision feature, and then significantly schemes according to the weight generation visual task relevant with visual task of each visual signature figure.

4. the collaborative target identification method of a kind of SAR image based on vision noticing mechanism according to claim 3 is characterized in that, in the step 3, significantly schemes based on visual task, selects successively the well-marked target in the SAR image regional according to inhibition of return mechanism.

5. a kind of SAR image based on vision noticing mechanism according to claim 4 is worked in coordination with target identification method, it is characterized in that, in the step 4, the proper vector of known prototype target sample is calculated in utilization based on the Cooperative Study algorithm of independent component analysis, make up known prototype target sample feature space, generate known prototype object knowledge storehouse, and calculate the adjoint vector of each known prototype target sample proper vector.

6. a kind of SAR image based on vision noticing mechanism according to claim 5 is worked in coordination with target identification method, it is characterized in that, step 5 comprises: the SAR image well-marked target zone of selecting is mapped in the target signature space, generate well-marked target provincial characteristics vector, calculate the initial order parameter in well-marked target zone, if the maximal value of the initial order parameter in well-marked target zone greater than setting threshold, is carried out the identification of well-marked target zone and output recognition result by order parameter dynamic iteration equation; If the maximal value of the initial order parameter in well-marked target zone less than setting threshold, is learnt unknown object by the Cooperative Study algorithm based on independent component analysis again, and the result that will learn dynamically is increased in the known prototype object knowledge storehouse.

Images