CN104036239A - Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering - Google Patents

Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering Download PDF

Info

Publication number
CN104036239A
CN104036239A CN201410234205.6A CN201410234205A CN104036239A CN 104036239 A CN104036239 A CN 104036239A CN 201410234205 A CN201410234205 A CN 201410234205A CN 104036239 A CN104036239 A CN 104036239A
Authority
CN
China
Prior art keywords
target
sar image
detection
region
area
Prior art date
Application number
CN201410234205.6A
Other languages
Chinese (zh)
Other versions
CN104036239B (en
Inventor
杨淑媛
焦李成
王士刚
侯彪
刘芳
刘红英
马晶晶
马文萍
熊涛
刘赵强
Original Assignee
西安电子科技大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 西安电子科技大学 filed Critical 西安电子科技大学
Priority to CN201410234205.6A priority Critical patent/CN104036239B/en
Publication of CN104036239A publication Critical patent/CN104036239A/en
Application granted granted Critical
Publication of CN104036239B publication Critical patent/CN104036239B/en

Links

Abstract

The invention discloses a fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering. The fast high-resolution SAR image ship detection method comprises the following steps: on the basis of the back scattering characteristics of each ground object and the prior information of a ship target in an SAR image, positioning a target potential position index map by an Otsu algorithm and range constraint; on the index map, pre-screening to obtain a detection binary segmentation map by a CFAR (constant false alarm rate) algorithm based on a local contrast; carrying out morphological processing to a detection result, and extracting a potential target slice from the SAR image and a detected binary segmentation map according to a processing result; and carrying out K-means clustering to the extracted slice by a designed identification feature to obtain a final identification result. According to the fast high-resolution SAR image ship detection method based on feature fusion and clustering, the data volume of a detection stage is effectively reduced by pre-processing, and point-to-point detection is not needed/the time of point-to-point detection is saved. Meanwhile, a target identification problem under the condition of insufficient training samples at present can be solved by the designed characteristic and a non-supervision clustering method, the target can be effectively positioned, and the size of the target can be estimated.

Description

The quick Ship Detection of High Resolution SAR image based on Fusion Features and cluster

Technical field

The invention belongs to technical field of image processing, relate to a kind of quick Ship Target Detection method of SAR image based on Fusion Features and cluster, for the synthesis of understanding and the decipher of aperture radar image.

Background technology

Synthetic-aperture radar (Synthetic Aperture Radar, SAR) be a kind of active sensor that uses microwave to carry out imaging, its imaging is not subject to the restriction of the condition such as weather, illumination, can carry out round-the-clock, round-the-clock observation to interesting target.The appearance of the SAR system of multiband, complete polarization, different working modes makes comprehensive earth observation become possibility.At present, the decipher ability of SAR image cannot meet the demand of the great amount of images processing of collecting, and this has directly caused the development of SAR image interpretation technology.Utilizing SAR image to carry out Ship Target Detection is the important means that realizes marine surveillance, fishery management and control, and China territorial waters is wide, ocean resources abundant, carries out SAR image naval vessel and detects significant.

The eighties in 20th century, Lincoln laboratory has proposed synthetic-aperture radar automatic target identification (SAR ATR) the tertiary treatment flow process based on layering attention mechanism, becomes the common recognition of scientific circles.This model adopts layering processing mode, and first view picture SAR image being carried out preview or detected is not obviously order target area to remove, and obtains potential target region; Then potential target region is differentiated, removal natural clutter false-alarm is wherein to obtain target area-of-interest; Finally area-of-interest is carried out to more complicated feature extraction and classification, with the object of realize target identification.Along with the in-depth of processing, data volume to be dealt with can be fewer and feweri, and computation complexity can be increasing, thereby can realize efficient understanding and decipher.SAR ATR has higher requirement for real-time and the accuracy rate of target identification, need to develop quick and efficient algorithm and carry out practical requirement.

The algorithm that current SAR image naval vessel detects utilizes Ship Target and marine site shows on SAR image feature difference around more, detects by setting an adaptive threshold.Classical constant false alarm rate (CFAR) detecting device is according to Ship Target and the difference of marine site gray feature around, with carrying out self-adaptation searching detection threshold according to the false alarm rate of setting after normal distribution modeling, can reach theoretic constant false alarm rate to background.It is a kind of simple and effective detection method that CFAR detects, and the existence that judges impact point according to following formula whether.

μ ROI - μ C σ C > λ

Wherein, μ rOIthe average in object support region, μ cthe average of clutter supporting zone, σ cbe the standard deviation of clutter supporting zone, λ is a constant relevant with presetting false alarm rate.If some place's above formula is satisfied at certain, this pixel to be measured is judged as target pixel points, otherwise is clutter pixel.The mode detection speed that this method adopts pointwise to detect is restricted, exist the fitting problems of owing of model to cause accuracy of detection to be affected simultaneously, estimate that with being positioned at the extremely limited observation pixel value in moving window region the parameter of normal distribution exists statistical deviation, this performance that tends to cause model mismatch and make detecting device is had a strong impact on.In SAR image, have bulk redundancy information, the partitioning algorithm based on Otsu and region operation can reduce data volume for detection-phase, improve decipher speed and efficiency.Above pre-service and two-parameter CFAR are combined to the overall performance that is expected to promote detecting device under the prerequisite that keeps constant false alarm rate.

The design of Discr. is two classification problems.Current method, based on there being supervision message, needs a large amount of typical target and false-alarm to cut into slices to obtain the parameter of sorter, as support vector machine (SVM) and QD.And obtaining of training data takes time and effort very much under actual conditions, and training data itself is to obtain under different sensors, imaging circumstances and target type, is difficult to characterize data to be tested.The Fusion Features of former figure section and binary map section upper texture, region and the contrast extracted is got up to provide more strong distinguishing ability.Based target and clutter feature priori without supervision K-means clustering method, the aggregation features of cutting into slices in feature space according to target and clutter, by the mode of cluster, target and clutter section are distinguished, can be broken away from the dependence to training sample, realize efficiently and differentiating.

Summary of the invention

For above-mentioned the deficiencies in the prior art, the present invention proposes a kind of quick Ship Target Detection method of SAR image based on Fusion Features and cluster, to reduce data volume, to reduce computing cost, adopt the unsupervised clustering of feature priori to break away from the restriction problem of training sample simultaneously, finally realize efficient target detection and location.

Realizing technical thought of the present invention is: at pretreatment stage, in conjunction with the difference of different atural object scattering propertiess and the prior imformation of target size, removal can not exist order target area; Detection-phase utilizes target and the difference of clutter scattering properties around, realizes prescreen obtain potential target region in conjunction with local contrast feature; After morphology is processed, area-of-interest section is proposed in the binary map from former figure and detecting; Design effective feature and according to feature priori, realize the discriminating of Ship Target by the mode of cluster.

The quick Ship Detection of High Resolution SAR image based on Fusion Features and cluster of the present invention, comprises the steps:

(1) pre-service based on Terrain Scattering characteristic and priori

1a) find according to the normalization histogram of image the Otsu optimum segmentation threshold value that makes inter-class variance maximum, original SAR image is divided into bright area and dark areas two parts, obtain binary segmentation figure.

1b) to step 1a) the binary segmentation figure that obtains carries out removing area in binary segmentation figure after hole filling and, much larger than the connected region of Ship Target area, obtains the key map in potential target region.

(2) CFAR based on local contrast detects

2a) according to the size of Ship Target, choose CFAR and detect required object support area size and the size of clutter supporting zone.

2b) original SAR image and key map are carried out to mirror reflection expansion around border, preset the false alarm rate of detection to determine adaptively the detection threshold at each some place according to background information in testing process.

2c) on the indicated potential target position of key map, calculate average energy and the interior average energy of clutter supporting zone of hollow moving window and the standard deviation of energy thereof in object support region, judge according to the magnitude relationship of the detection threshold at the pixel average of pixel supporting zone to be measured and this some place whether this point is target pixel points.

2d) hollow moving window moves after the result that obtains detecting on original SAR image, and the testing result obtaining is corroded to expansive working to remove isolated check point and to fill and supplement concealed impact point with hole.

(3) region of interesting extraction and feature construction thereof and fusion

3a) in the binary segmentation figure from detecting, extract all connected regions and remove the connected region of area much smaller than target, according to the size of the barycenter of remaining area and default section, in the binary segmentation figure from original SAR image and detecting, extract the section of area-of-interest.

3b) in the each original SAR image slice of extracting, calculate the feature of logarithm standard deviation as this field strength undulatory property of tolerance; The number of finding eight connected regions in binary segmentation figure section after each detection of extracting is as the feature of the strong scattering space of points divergence in description region; Meanwhile, the tolerance as target area average energy according to the mean value of energy in original SAR image corresponding to the indicated target area position calculation of the binary segmentation figure after detecting.

3c) logarithm standard deviation, eight connected region numbers and target area average energy are normalized respectively and merge the rear proper vector with stronger resolving ability that forms, as the comprehensive description for each area-of-interest; The feature in target area and clutter region presents aggregation properties on feature space, and is all the guidance that has priori for each feature in target area and clutter region.

(4) target based on feature priori and K-means cluster is differentiated

The initial cluster center of 4a) setting cluster classification number, maximum iteration time and being determined by feature priori, measure the similarity of each sample to be tested and cluster centre according to Euclidean distance, and with this, each sample is sorted out, until K-means cluster reaches convergence.

4b), according to the sample class that finally obtains, the corresponding area-of-interest of the sample position of gathering for target is found, and with the bounding box of suitable size on original SAR image by the region labeling that is finally defined as Ship Target out.

4c) from be demarcated as the binary segmentation figure the corresponding detection in region of Ship Target, extract the minimum boundary rectangle of target, thereby draw length and the width of this Ship Target.

Further, described step 1a) in, suppose that original SAR image is I, its size is m × n, with the optimum segmentation threshold value T that makes inter-class variance maximum optimage is divided into territory, area pellucida and dark areas two parts, optimum segmentation threshold value T optobtain according to following formula:

T opt = max k [ m G P 1 ( k ) - m ( k ) ] 2 P 1 ( k ) [ 1 - P 1 ( k ) ]

Wherein, m g(k) be the average gray value of image, m (k) is the pixel average that gray-scale value is less than k, P 1(k) be the shared ratio of pixel that pixel value is less than k.

Further, described step 1b) in, the binary segmentation figure after Otsu is cut apart carries out hole filling with completion connected region, adds up after the size of all connected regions, and area is removed much larger than the region of 3 times of Ship Target sizes.

Further, described step 3a) in, the method of extracting all connected regions in binary segmentation figure from detecting is: the binary segmentation figure after detecting is carried out to morphological operation, described morphological operation comprises fills completion connected region with hole, remove isolated point and be connected abutment points with corroding with expanding, thereby obtaining all connected regions.

Further, described step 2c) in, the criterion of target pixel points is: if the pixel average of pixel supporting zone to be measured is greater than the detection threshold at this some place, this point is judged as target pixel points; If the pixel average of pixel supporting zone to be measured is less than or equal to the detection threshold at this some place, this point is judged as clutter pixel.

Compared with prior art, advantage of the present invention is:

1, the present invention is according to the difference of each atural object Electromagnetic Scattering Characteristics and the prior imformation of naval vessel size, cut apart and the constraint of connected region area by Otsu optimal threshold, effectively reduce the hunting zone of target, promote detection speed, overcome pointwise and detected the time and the computing cost that bring.

2, in the section that the present invention extracts in original SAR image and binary segmentation image, designed respectively the new feature of differentiating for target, and feature is combined and forms the feature group with sign ability, have better differentiation performance than classical diagnostic characteristics.Under the guidance of feature prior imformation, rely on target and clutter at the aggregation of feature space simultaneously, use unsupervised clustering method to realize automatic target discriminating, effectively overcome the problem of lack of training samples, obtained comparatively desirable decipher effect.

Below with reference to drawings and Examples, the present invention is described in further details.

Brief description of the drawings

Fig. 1 is process flow diagram of the present invention.

Fig. 2 is that the CFAR that the present invention adopts detects hollow moving window schematic diagram.

Fig. 3 (a) is original SAR image.

Fig. 3 (b) is the binary segmentation figure after Otsu is cut apart, and the optimal threshold wherein obtaining is 60.

Fig. 3 (c) fills and the key map of region after removing through hole.

Fig. 3 (d) is two-parameter CFAR testing result figure, wherein P fa=0.03.

Fig. 4 is that the present invention is at the ROC curve without pre-service and the detection of the CFAR under pre-service.

Fig. 5 is the area-of-interest that the present invention extracts from original SAR image and binary segmentation image.

Fig. 6 (a) is the logarithm standard deviation feature of extracting section.

Fig. 6 (b) is that eight connected regions of extracting section are counted feature.

Fig. 6 (c) is the target area average energy feature of extracting section.

Fig. 7 (a) is original SAR image.

Fig. 7 (b) is the potential target areal map of differentiating prelocalization.

Fig. 7 (c) is the results of three features that adopt of the present invention after K-means differentiates.

Fig. 7 (d) is the results of Lincoln laboratory three features after K-means differentiates.

Fig. 8 is the target size estimated result that the present invention obtains.

Embodiment

Embodiment 1:

The prior art of mentioning in background technology exists obtaining with computing cost of data volume too large, and be subject to the problems such as the restriction of training sample, make up these technological deficiencies for reaching, realize the object of efficient target detection and location, the invention provides a kind of quick Ship Target Detection method of SAR image based on Fusion Features and cluster.

Realizing technical thought of the present invention is: at pretreatment stage, in conjunction with the difference of different atural object scattering propertiess and the prior imformation of target size, removal can not exist order target area; Detection-phase utilizes target and the difference of clutter scattering properties around, realizes prescreen obtain potential target region in conjunction with local contrast feature; After morphology is processed, area-of-interest section is proposed in the binary map from former figure and detecting; Design effective feature and according to feature priori, realize the discriminating of Ship Target by the mode of cluster.

The quick Ship Target Detection method of the described SAR image based on Fusion Features and cluster, as shown in Figure 1, comprises the steps:

Step 1: based on the pre-service of Terrain Scattering characteristic and priori

1.1) suppose that original SAR image is I, its size is m × n, uses the optimum segmentation threshold value T of inter-class variance maximum optoriginal SAR image is divided into territory, area pellucida and dark areas two parts, optimum segmentation threshold value T optobtain according to following formula:

T opt = max k [ m G P 1 ( k ) - m ( k ) ] 2 P 1 ( k ) [ 1 - P 1 ( k ) ]

Wherein, m gbe the average gray value of image, m (k) is the pixel average that gray-scale value is less than k, P 1(k) be the shared ratio of pixel that pixel value is less than k.

1.2) the binary segmentation figure after Otsu is cut apart carries out hole filling with completion connected region, adds up after the size of all connected regions, and area is removed much larger than the region of 3 times of Ship Target sizes, obtains the key map I in potential target region index.

Step 2: the CFAR based on local contrast detects

2.1) preset CFAR according to the distance priori between the size on naval vessel and naval vessel and detect the required each parameter of hollow moving window, as shown in Figure 2.The present invention is made as 3 × 3 so that isolated strong clutter point is suppressed by the size in object support region; the radius of protection zone is made as 15 and avoids other target to be leaked in hollow moving window; the number that clutter supports pixel is made as 120, to represent surrounding's clutter information of pixel to be measured.

2.2) due to the existence of hollow moving window, need to be to original SAR image I and key map I indexcarry out mirror reflection and be extended to (m+30) × (n+30) size.In CFAR algorithm, false alarm rate P fachoose and should be as far as possible between true verification and measurement ratio and false alarm rate, reach compromise, if P fathat selects is excessive, can comprise a large amount of clutter points in testing result, and P fathe too small of choosing can cause undetected by lose objects point again.False alarm rate P in the present invention faelect 0.03 as.At every bit place, if calculate the average μ of clutter supporting zone cand standard deviation sigma cafter, corresponding detection threshold T can be tried to achieve by following formula:

∫ 0 T 1 2 π σ C e - ( x - μ C ) 2 2 σ C 2 dx = 1 - P fa

2.3) calculate the pixel average μ of pixel supporting zone to be measured rOIand compare with the detection threshold T obtaining, if μ rOI> T this point is judged as target pixel points, otherwise is clutter pixel.On the original SAR image of view picture, complete the detection to all potential target points according to the instruction of key map successively mobile hollow moving window, thereby obtain the binary segmentation figure I that CFAR detects det.

2.4) in order to verify the effect of pre-service to detection-phase and the performance of DP-CFAR CFAR detecting device to be analyzed, by false alarm rate P default in CFAR fabe set as respectively, since 0 taking 0.01 discrete value as interval to 0.25, under each false alarm rate, obtaining the testing result of CFAR, and the Groundtruth figure of hand labeled mates in advance, calculates real verification and measurement ratio P dwith real false alarm rate P f, false alarm rate and verification and measurement ratio are drawn out respectively to experimenter's operation curve (ROC) of CFAR detecting device as horizontal ordinate and ordinate.With above identical, also can obtain its corresponding ROC curve for the process of directly carrying out CFAR detection without pre-service.

Step 3: region of interesting extraction and feature construction thereof and fusion

3.1) for the binary segmentation figure I after detecting detcarry out morphological operation, comprise with hole and fill completion connected region, remove isolated point and be connected abutment points with corroding with expanding, thereby obtaining more regular region.

3.2) extract barycenter and the area information of all connected regions, if the area of certain connected region is too small, illustrate that this region is the nontarget area that comprises clutter, can get rid of the possibility that it is target area.According to the prior imformation of the spatial resolution of this experiment SAR image and naval vessel size, the region that is less than 50 (minimum target half) for area is rejected.Centered by residue connected region barycenter, in the binary segmentation image from original SAR image and detecting, extract respectively the section of 64 × 64 sizes, thereby obtain the section at potential target place.Suppose that the potential target number extracting is n, the section of extracting from original SAR image is expressed as C 1, C 2, C n, the section of extracting in the binary segmentation figure from detecting is respectively B 1, B 2, B n, the size of each section is 64 × 64.

3.3) from C 1, C 2, C nthe logarithm standard deviation feature of the each potential target section of middle calculating characterizes the fluctuation information of the reflection strength in this region.Computing formula is as follows:

σ i = S 2 i - S 1 i 2 / N N - 1 , i ∈ [ 1 , n ]

Wherein, N be section in pixel sum and

S 1 i = Σ r , a ∈ [ 1,64 ] 10 l og 10 C i ( r , a ) , i ∈ [ 1 , n ]

S 2 i = Σ r , a ∈ [ 1,64 ] [ 10 log 10 C i ( r , a ) ] 2 , i ∈ [ 1 , n ]

3.4) from B 1, B 2, B neight connected region numbers of the each potential target section of middle calculating characterize the distribution situation of the strong scattering space of points in this region.Under normal circumstances, target area strong scattering point is more assembled with respect to clutter, so the number of eight connected regions is fewer than clutter.

3.5) in conjunction with B 1, B 2, B nwith C 1, C 2, C ncalculate the target area average energy of each potential target section in order to characterize the average scattering intensity of this region internal object.Owing to detecting the existence of false-alarm, have compared with the clutter region of strong scattering characteristic can present with target class like structure and target is formed and disturbed.This feature can be distinguished target and clutter according to the scattering strength average of the some institute respective pixel that is detected as object pixel.Its computing formula is as follows:

p i = 1 n i Σ B i ( r , a ) ∈ t arg et C i ( r , a ) , i ∈ [ 1 , n ]

Wherein, n iit is the number of pixels that is judged to impact point in i section.

3.6) above Feature Combination is become to matrix form, every a line represents respectively three features of section, and total n is capable, and result is as follows:

F = f 11 f 12 f 13 . . . . . . . . . f n 1 f n 2 f n 3

The significance level of supposing each feature is identical, and due to each characteristic dimension difference, need to be normalized to operate to form to each row of F has the feature of resolving ability group.In the present invention, we normalize to each row in F between 0 and 1.

Step 4: the target based on feature priori and K-means cluster is differentiated

4.1) because the result of cluster only has target and clutter, so the cluster number in K-means is made as to 2, initial cluster center is made as to [1,0,1] and [0,1,0] according to the feature priori of target and clutter, maximum iteration time is made as 100.Ask for the Euclidean distance of each sample and cluster centre at every turn, sample is classified as to a nearer with it class, upgrade such center by the sample average that is classified as each class, until cluster centre no longer changes or reaches maximum iteration time simultaneously.

4.2) aggregation being had from target and clutter feature space, target will finally be classified as the first kind and clutter is classified as Equations of The Second Kind in theory.To differentiate and in original SAR image, mark out to determine its position with red square frame for order target area in conjunction with final cluster result.Simultaneously extract according to the binary map detecting the extraneous rectangle frame of minimum that surrounds target, can be used for according to this size of estimating target.

Embodiment 2:

Effect of the present invention further illustrates by following emulation experiment.

(1) experiment simulation condition:

The data that this experiment adopts are TerraSAR high-resolution SAR images.These data are diameter radar images that the resolution in the one regional shore line, the Straits of Gibraltar that obtained under HH polarization mode, X-band by TerraSAR satellite is 1m, comprise the types of ground objects such as ocean, mountain range, buildings, river, harbour and naval vessel, wherein Ship Target to be detected have 21 and target type different.The scene size that experimental data covers, for 2987m × 4134m, is the gray-scale map of each pixel 8bits.This experiment CPU be Intel (R) Core (TM) i5-3470, dominant frequency be 3.2GHz, in save as in the WINDOWS7 operating system of 4G and adopt the MATLAB2012a of 32 to carry out emulation.

(2) target detection and discriminating Performance evaluation criterion:

(2a) verification and measurement ratio and false alarm rate

Suppose that the big or small total number for M × N and object pixel of SAR image is N target, in SAR image clutter pixel add up to N clutter=M × N-N target.If the number of the object pixel detecting is N dt, the false-alarm number of generation is N dc, actual false-alarm probability and actual detection probability are:

P f = N dc N clutter

P d = N dt N t arg et

With the artificial reference diagram of demarcating, the testing result obtaining under different false alarm rates and reference diagram are compared with two numerical value above calculating in advance.The result obtaining is depicted as to ROC curve, and it is better that the area of curve and transverse axis institute enclosing region more shows to detect performance.

(2b) total wrong number, fail to report number, total accuracy and target accuracy

The discriminating stage with total wrong number, fail to report number, total accuracy and target accuracy and evaluate the quality of diagnostic characteristics and Discr., its definition is as follows respectively:

Total wrong number be target be judged to the number of clutter false-alarm and clutter false-alarm be judged to target number and, have

n e=n tc+n ct

Wherein, n etotal wrong number, n tcfor target is judged to the number of clutter false-alarm, also referred to as failing to report number, n ctfor clutter false-alarm is judged to the number of target.

If the number of slices that the discriminating stage extracts is n, real goal number of slices is wherein m, and total accuracy and target accuracy are defined as respectively:

R a = n - n e n

R t = m - n tc m

Generally, SARATR requires to detect and the discriminating stage should ensure under the prerequisite that verification and measurement ratio and target accuracy are 1, reduces as far as possible false alarm rate by adjusting parameter preset, thereby realizes complete, efficient location to target.

(2c) the decipher time

The decipher time refers to always consuming time from input SAR image to output detections result.Because SARATR has very high requirement to real-time, generally need to complete in a short period of time the location to target, so working time should be short in as much as possible to reach application request.

(3) experiment content

Experiment one

Utilize the designed quick Ship Detection of High Resolution SAR image based on Fusion Features and cluster of the present invention to carry out target detection, experimental result as shown in Figure 3.Wherein:

Fig. 3 (a) is original SAR image;

Fig. 3 (b) is the binary segmentation figure after Otsu is cut apart, and the optimal threshold wherein obtaining is 60;

Fig. 3 (c) fills and the key map of region after removing through hole;

Fig. 3 (d) is two-parameter CFAR testing result figure, wherein P fa=0.03;

From Fig. 3 (a), Fig. 3 (b), Fig. 3 (c), Fig. 3 (d), the present invention has reduced data processing amount effectively by pre-service, the false alarm rate detecting is greatly reduced, thereby accelerated the speed of detection-phase.Fig. 4 has drawn out without pre-service with through in two kinds of situations of pre-service, experimenter's operation curve that the two-parameter CFAR obtaining detects.As can be seen from the figure, pre-service makes experimenter's operation curve to left, has increased the area that Curves surrounds, thereby has improved the performance of detection-phase.Table 1 has been listed the calculating used time in two kinds of situations.

Table 1 detection-phase calculates the used time

As can be seen from Table 1, the combination of pre-service and two-parameter CFAR greatly reduces the data volume detecting, thereby has reduced the time complexity of algorithm.

Experiment two

The testing result obtaining is carried out after the morphology processing such as burn into expansion and hole filling, in binary segmentation image from original SAR image and detecting, extract size is 64 × 64 the slice map that comprises potential target simultaneously, and the partial results of experiment as shown in Figure 5.Between the section of extracting as seen from Figure 5, there is very strong similarity, need effective method to distinguish it.

Calculate respectively the value of the corresponding logarithm standard deviation of each section, eight connected region numbers and three features of target area average energy, form the description to section by being combined into proper vector after feature normalization, experimental result as shown in Figure 6.Wherein:

Fig. 6 (a) is the logarithm standard deviation feature of extracting section;

Fig. 6 (b) is that eight connected regions of extracting section are counted feature;

Fig. 6 (c) is the target area average energy feature of extracting section;

Similarity based on each section in feature, need to combine multiple features the fine differentiation that could form target and clutter false-alarm.

The prior imformation of cutting into slices in feature according to target slice and false-alarm, uses K-means cluster to realize the discriminating for target, and result as shown in Figure 7.Wherein:

Fig. 7 (a) is original SAR image;

Fig. 7 (b) is the potential target areal map of differentiating prelocalization;

Fig. 7 (c) is the results of three features that adopt of the present invention after K-means differentiates;

Fig. 7 (d) is the results of Lincoln laboratory three features after K-means differentiates.

From Fig. 7 (a), Fig. 7 (b), Fig. 7 (c), Fig. 7 (d), the present invention can overcome the SAR image rare problem of training sample in actual applications, by Fusion Features is combined with unsupervised clustering and can effectively complete the task of Ship Target Detection.Designed feature has better differentiation performance compared with classical Lincoln laboratory three features, has obtained better identification result.

High Resolution SAR image provides abundanter scene information, can carry out the size of estimating target by the result after differentiating, thereby provides useful help for marine surveillance and fishery management and control.The minimum boundary rectangle figure of gained part target as shown in Figure 8, can estimate the size of each target in conjunction with the resolution of SAR image.

Table 2 has been listed the present invention and the comparing result of Lincoln laboratory method in discriminating performance, and wherein Lincoln tests three real features and is respectively logarithm standard deviation, fractal dimension and arrangement energy Ratios.

The total wrong number of table 2 liang stack features under K-means, fail to report number, total accuracy and target accuracy

As seen from Table 2, designed traditional Lincoln laboratory three features of aspect ratio of the present invention have more sign, and the discrimination method based on K-means cluster has been obtained comparatively satisfied result simultaneously, and every identification beacon all approaches ideal value.

To sum up, the present invention can significantly reduce the data complexity of SARATR, improve the detection performance of detecting device, overcome the problem of lack of training samples simultaneously, more efficiently target and clutter false-alarm are differentiated, reduce the required time of decipher, had certain application prospect at SAR image understanding and decipher field.

By reference to the accompanying drawings embodiments of the present invention are described above, but the present invention is not limited to above-mentioned embodiment, in the ken that one skilled in the relevant art possesses, can also under the prerequisite that does not depart from aim of the present invention, makes a variety of changes.

Claims (5)

1. the quick Ship Detection of High Resolution SAR image based on Fusion Features and cluster, is characterized in that, comprises the steps:
(1) pre-service based on Terrain Scattering characteristic and priori
1a) find according to the normalization histogram of image the Otsu optimum segmentation threshold value that makes inter-class variance maximum, original SAR image is divided into bright area and dark areas two parts, obtain binary segmentation figure;
1b) to step 1a) the binary segmentation figure that obtains carries out removing area in binary segmentation figure after hole filling and, much larger than the connected region of Ship Target area, obtains the key map in potential target region;
(2) CFAR based on local contrast detects
2a) according to the size of Ship Target, choose CFAR and detect required object support area size and the size of clutter supporting zone;
2b) original SAR image and key map are carried out to mirror reflection expansion around border, preset the false alarm rate of detection to determine adaptively the detection threshold at each some place according to background information in testing process;
2c) on the indicated potential target position of key map, calculate average energy and the interior average energy of clutter supporting zone of hollow moving window and the standard deviation of energy thereof in object support region, judge according to the magnitude relationship of the detection threshold at the pixel average of pixel supporting zone to be measured and this some place whether this point is target pixel points;
2d) hollow moving window moves after the result that obtains detecting on original SAR image, and the testing result obtaining is corroded to expansive working to remove isolated check point and to fill and supplement concealed impact point with hole;
(3) region of interesting extraction and feature construction thereof and fusion
3a) in the binary segmentation figure from detecting, extract all connected regions and remove the connected region of area much smaller than target, according to the size of the barycenter of remaining area and default section, in the binary segmentation figure from original SAR image and detecting, extract the section of area-of-interest;
3b) in the each original SAR image slice of extracting, calculate the feature of logarithm standard deviation as this field strength undulatory property of tolerance; The number of finding eight connected regions in binary segmentation figure section after each detection of extracting is as the feature of the strong scattering space of points divergence in description region; Meanwhile, the tolerance as target area average energy according to the mean value of energy in original SAR image corresponding to the indicated target area position calculation of the binary segmentation figure after detecting;
3c) logarithm standard deviation, eight connected region numbers and target area average energy are normalized respectively and merge the rear proper vector with stronger resolving ability that forms, as the comprehensive description for each area-of-interest; The feature in target area and clutter region presents aggregation properties on feature space, and is all the guidance that has priori for each feature in target area and clutter region;
(4) target based on feature priori and K-means cluster is differentiated
The initial cluster center of 4a) setting cluster classification number, maximum iteration time and being determined by feature priori, measure the similarity of each sample to be tested and cluster centre according to Euclidean distance, and with this, each sample is sorted out, until K-means cluster reaches convergence;
4b), according to the sample class that finally obtains, the corresponding area-of-interest of the sample position of gathering for target is found, and with the bounding box of suitable size on original SAR image by the region labeling that is finally defined as Ship Target out;
4c) from be demarcated as the binary segmentation figure the corresponding detection in region of Ship Target, extract the minimum boundary rectangle of target, thereby draw length and the width of this Ship Target.
2. the quick Ship Detection of High Resolution SAR image based on Fusion Features and cluster according to claim 1, it is characterized in that described step 1a) in, suppose that original SAR image is I, its size is m × n, with the optimum segmentation threshold value T that makes inter-class variance maximum optimage is divided into territory, area pellucida and dark areas two parts, optimum segmentation threshold value T optobtain according to following formula:
T opt = max k [ m G P 1 ( k ) - m ( k ) ] 2 P 1 ( k ) [ 1 - P 1 ( k ) ]
Wherein, m g(k) be the average gray value of image, m (k) is the pixel average that gray-scale value is less than k, P 1(k) be the shared ratio of pixel that pixel value is less than k.
3. the quick Ship Detection of High Resolution SAR image based on Fusion Features and cluster according to claim 1, it is characterized in that: described step 1b) in, binary segmentation figure after Otsu is cut apart carries out hole filling with completion connected region, add up after the size of all connected regions, area is removed much larger than the region of 3 times of Ship Target sizes.
4. the quick Ship Detection of High Resolution SAR image based on Fusion Features and cluster according to claim 1, it is characterized in that, described step 3a) in, the method of extracting all connected regions in binary segmentation figure from detecting is: the binary segmentation figure after detecting is carried out to morphological operation, described morphological operation comprises fills completion connected region with hole, remove isolated point and be connected abutment points with corroding with expanding, thereby obtaining all connected regions.
5. the quick Ship Detection of High Resolution SAR image based on Fusion Features and cluster according to claim 1, it is characterized in that, described step 2c) in, the criterion of target pixel points is: if the pixel average of pixel supporting zone to be measured is greater than the detection threshold at this some place, this point is judged as target pixel points; If the pixel average of pixel supporting zone to be measured is less than or equal to the detection threshold at this some place, this point is judged as clutter pixel.
CN201410234205.6A 2014-05-29 2014-05-29 Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering CN104036239B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410234205.6A CN104036239B (en) 2014-05-29 2014-05-29 Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410234205.6A CN104036239B (en) 2014-05-29 2014-05-29 Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering

Publications (2)

Publication Number Publication Date
CN104036239A true CN104036239A (en) 2014-09-10
CN104036239B CN104036239B (en) 2017-05-10

Family

ID=51467005

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410234205.6A CN104036239B (en) 2014-05-29 2014-05-29 Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering

Country Status (1)

Country Link
CN (1) CN104036239B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104408482A (en) * 2014-12-08 2015-03-11 电子科技大学 Detecting method for high-resolution SAR (Synthetic Aperture Radar) image object
CN105205450A (en) * 2015-08-24 2015-12-30 辽宁工程技术大学 SAR image target extraction method based on irregular marked point process
CN105223561A (en) * 2015-10-23 2016-01-06 西安电子科技大学 Based on the radar terrain object Discr. method for designing of space distribution
WO2016101279A1 (en) * 2014-12-26 2016-06-30 中国海洋大学 Quick detecting method for synthetic aperture radar image of ship target
CN106803078A (en) * 2017-01-18 2017-06-06 哈尔滨工业大学 A kind of SAR image Ship Target dividing method
CN107977667A (en) * 2016-10-21 2018-05-01 西安电子科技大学 SAR target discrimination methods based on semi-supervised coorinated training
CN107330465B (en) * 2017-06-30 2019-07-30 清华大学深圳研究生院 A kind of images steganalysis method and device

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102208017A (en) * 2011-05-30 2011-10-05 西安电子科技大学 Ship detection method based on high-resolution synthetic aperture radar images
CN102622607A (en) * 2012-02-24 2012-08-01 河海大学 Remote sensing image classification method based on multi-feature fusion
US8422738B1 (en) * 2008-08-25 2013-04-16 The United States Of America As Represented By The Secretary Of The Navy Adaptive automated synthetic aperture radar vessel detection method with false alarm mitigation
CN103400156A (en) * 2013-07-04 2013-11-20 西安电子科技大学 CFAR (Constant False Alarm Rate) and sparse representation-based high-resolution SAR (Synthetic Aperture Radar) image ship detection method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8422738B1 (en) * 2008-08-25 2013-04-16 The United States Of America As Represented By The Secretary Of The Navy Adaptive automated synthetic aperture radar vessel detection method with false alarm mitigation
CN102208017A (en) * 2011-05-30 2011-10-05 西安电子科技大学 Ship detection method based on high-resolution synthetic aperture radar images
CN102622607A (en) * 2012-02-24 2012-08-01 河海大学 Remote sensing image classification method based on multi-feature fusion
CN103400156A (en) * 2013-07-04 2013-11-20 西安电子科技大学 CFAR (Constant False Alarm Rate) and sparse representation-based high-resolution SAR (Synthetic Aperture Radar) image ship detection method

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104408482A (en) * 2014-12-08 2015-03-11 电子科技大学 Detecting method for high-resolution SAR (Synthetic Aperture Radar) image object
CN104408482B (en) * 2014-12-08 2019-02-12 电子科技大学 A kind of High Resolution SAR Images object detection method
WO2016101279A1 (en) * 2014-12-26 2016-06-30 中国海洋大学 Quick detecting method for synthetic aperture radar image of ship target
CN106170819A (en) * 2014-12-26 2016-11-30 中国海洋大学 A kind of diameter radar image Ship Target method for quick
CN106170819B (en) * 2014-12-26 2019-04-05 中国海洋大学 A kind of diameter radar image Ship Target rapid detection method
CN105205450A (en) * 2015-08-24 2015-12-30 辽宁工程技术大学 SAR image target extraction method based on irregular marked point process
CN105205450B (en) * 2015-08-24 2018-08-07 辽宁工程技术大学 A kind of SAR image target extraction method based on irregular mark point process
CN105223561A (en) * 2015-10-23 2016-01-06 西安电子科技大学 Based on the radar terrain object Discr. method for designing of space distribution
CN107977667A (en) * 2016-10-21 2018-05-01 西安电子科技大学 SAR target discrimination methods based on semi-supervised coorinated training
CN106803078A (en) * 2017-01-18 2017-06-06 哈尔滨工业大学 A kind of SAR image Ship Target dividing method
CN107330465B (en) * 2017-06-30 2019-07-30 清华大学深圳研究生院 A kind of images steganalysis method and device

Also Published As

Publication number Publication date
CN104036239B (en) 2017-05-10

Similar Documents

Publication Publication Date Title
Parizzi et al. Adaptive InSAR stack multilooking exploiting amplitude statistics: A comparison between different techniques and practical results
Prieto et al. Using self-organising maps in the detection and recognition of road signs
CN102622607B (en) Remote sensing image classification method based on multi-feature fusion
JP2014232533A (en) System and method for ocr output verification
Huang et al. A comparative study of spatial approaches for urban mapping using hyperspectral ROSIS images over Pavia City, northern Italy
Solberg et al. Automatic detection of oil spills in ERS SAR images
US8611604B2 (en) Object detection device
Mishra et al. Land cover classification of PALSAR images by knowledge based decision tree classifier and supervised classifiers based on SAR observables
Hu et al. Automatic segmentation of high-resolution satellite imagery by integrating texture, intensity, and color features
Yu et al. Learning hierarchical features for automated extraction of road markings from 3-D mobile LiDAR point clouds
Cord et al. Automatic road defect detection by textural pattern recognition based on AdaBoost
Overett et al. Large scale sign detection using HOG feature variants
Xu et al. Multiple-entity based classification of airborne laser scanning data in urban areas
Carlberg et al. Classifying urban landscape in aerial LiDAR using 3D shape analysis
Wang et al. Ship detection for high-resolution SAR images based on feature analysis
JP4997178B2 (en) Object detection device
Leigh et al. Automated ice–water classification using dual polarization SAR satellite imagery
Lu et al. Two-class weather classification
Solberg et al. Automatic detection of oil spills in Envisat, Radarsat and ERS SAR images
CN102521565B (en) Garment identification method and system for low-resolution video
Chen et al. Vehicle detection in high-resolution aerial images via sparse representation and superpixels
Zhang et al. Character energy and link energy-based text extraction in scene images
CN104077599A (en) Polarization SAR image classification method based on deep neural network
Topouzelis et al. Oil spill feature selection and classification using decision tree forest on SAR image data
CN103400147B (en) Target fish recognition method and system based on image procossing

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
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20180702

Address after: 710077 Shaanxi provincial science and technology resources co ordination center, room 10, 10 zhang85 Road, Xi'an high tech Zone, A405, China

Patentee after: Xi'an Wien think Union Data Technology Co., Ltd.

Address before: No. 2 Taibai Road, Xi'an, Shaanxi Province, Shaanxi

Patentee before: Xidian University