CN105759268B - The adaptive quick determination methods of SAR image CFAR based on multithreading - Google Patents

The adaptive quick determination methods of SAR image CFAR based on multithreading Download PDF

Info

Publication number
CN105759268B
CN105759268B CN201610173184.0A CN201610173184A CN105759268B CN 105759268 B CN105759268 B CN 105759268B CN 201610173184 A CN201610173184 A CN 201610173184A CN 105759268 B CN105759268 B CN 105759268B
Authority
CN
China
Prior art keywords
regions
sar image
window
pixel
detection
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201610173184.0A
Other languages
Chinese (zh)
Other versions
CN105759268A (en
Inventor
王瑞富
李洁
朱金山
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong University of Science and Technology
Original Assignee
Shandong University of Science and Technology
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 Shandong University of Science and Technology filed Critical Shandong University of Science and Technology
Priority to CN201610173184.0A priority Critical patent/CN105759268B/en
Publication of CN105759268A publication Critical patent/CN105759268A/en
Application granted granted Critical
Publication of CN105759268B publication Critical patent/CN105759268B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
    • G01S13/9021—SAR image post-processing techniques
    • G01S13/9027—Pattern recognition for feature extraction
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques

Abstract

The invention discloses a kind of adaptive quick determination methods of SAR image CFAR based on multithreading, belong to Ship targets detection system regions, first, the quick screening of possible target point in global threshold completion image is asked for using statistics with histogram CFAR algorithms, obtains index matrix；Then, the sliding window detection based on K distributions is carried out to the possibility target point filtered out by index matrix, which improves local detection speed with multithreading；Four connected region area statistics finally are made to the target point of gained in local detect, possible ship is filtered out according to minimum ship elemental area；By analyzing testing result of the algorithm to SAR image, show that the algorithm is more suitable for applying to the conclusion in Ship targets detection system.Quick determination method proposed by the present invention is compared with Salazar algorithms, and in the case of Detection accuracy is undiminished, detection speed is greatly improved, and is more suitable for applying in Ship targets detection system.

Description

The adaptive quick determination methods of SAR image CFAR based on multithreading
Technical field
The invention belongs to Ship targets detection system regions, and in particular to the SAR image CFAR based on multithreading is adaptively quick Detection method.
Background technology
With the development of global ocean technology, the real-time and accuracy requirement of Ship targets detection are also higher and higher.It is so far Only, much algorithms on target detection are proposed in document, in these algorithms, constant false alarm rate (Constant False Alarm Rate, CFAR) detection algorithm because calculate simple, threshold adaptive, complex background detection speed it is fast (Salazar, 1999, Bisceglie and Galdi, 2005) and it is considered as a kind of important algorithm of SAR image target detection, oneself is extensive It is applied in many Ship targets detection systems (Crisp, 2004).
Realize that target detection usually requires to select suitable clutter distributed model and detector using CFAR.In ocean clutter The statistical model applied in background has Gauss, Rayleigh, lognormal (Farina and Russo, 1986), Wei Buer (Riflcin, 1994), G0 (Frery, 1997), K distributions (Erfanian and Vakili, 2009) etc., common CFAR inspections Survey device (Novak and Hesse, 1991, Kuttikkad and Chellappa, 1994, Salazar, 1999, Hofele, 2001) there are cell-average (CA), ordered statistics (OS), take big (GO), take small (SO)-CFAR, and several inspections based on more than Survey combination CFAR of device etc..CA-CFAR detection results under uniform clutter environment are fine, but the performance under non-homogeneous background It is poor；GO, SO, OS scheduling algorithm for there are the non-homogeneous background of noise jamming all lack universality (Liu Jiayuan and Jiao are refined red, 2013).In order to improve the universality of algorithm, Smith and Varshney (Smith and Varshney, 1997, Smith and Varshney, 2000) from the angle that algorithm is adaptive selected, selected accordingly most in the case of different clutter backgrounds Good detection algorithm, it is proposed that VI-CFAR (Variability Index-CFAR), can be dynamically selected according to VI hypothesis testings CA, SO, GO scheduling algorithm；Farrouki and Barkat (Farrouki and Barkat, 2005) are from being adaptive selected reference image The angle of element set is set out, with reference to CA-CFAR detection performances good under uniform clutter background, it is proposed that based on ordered data Change automatic screening cell-average (the Automatic Censored of statistic (Ordered Data Variability, ODV) Cell Averaging, ACCA) algorithm；Bisceglie(Bisceglie and Galdi,2005,Bisceglie and Galdi, 2001) a kind of CFAR algorithms (being referred to as Bisceglie algorithms herein) for removing a part of numerical value et al. are proposed, very It is adapted to the target detection of regional area；Salazar (Salazar, 1999) et al. is proposed based on Beta-prime distributions CFAR algorithms (are referred to as Salazar algorithms) herein；Kuttikkad (Kuttikkad and Chellappa, 1994) et al. is carried Target CFAR algorithms being distributed based on Weir and based on K distributions etc. are gone out.As can be seen that being based on CFAR in these literature research The research of SAR algorithm of target detection the most extensively, it is practical.
Gui Gao (Gao, 2009) et al. improve Salazars algorithms, it is proposed that a kind of CFAR based on AC is adaptive Fast algorithm of detecting, the algorithm can accordingly improve detection speed, but still do not reach in the case where ensureing target accuracy rate Requirement of real-time of the military combat to Ship Target Detection.
The content of the invention
For above-mentioned technical problem existing in the prior art, the present invention proposes a kind of SAR image based on multithreading The adaptive quick determination methods of CFAR, design is reasonable, overcomes existing deficiency, has good promotion effect.
To achieve these goals, the present invention adopts the following technical scheme that：
A kind of adaptive quick detection sides of SAR (synthetic aperture radar) image CFAR (constant false alarm rate) based on multithreading Method, carries out in accordance with the following steps：
Step 1：Input SAR image；
Step 2：Global detection is carried out to view picture SAR image, CFAR values are set by statistics with histogram or make the right way of conduct by oneself Method asks for global threshold；
Step 3：Index matrix is obtained, is specifically carried out in accordance with the following steps：
Step 3.1：Compare the value of each pixel and the global threshold calculated by step 2 in SAR image；
If：The value of pixel is less than global threshold in SAR image, then the value of pixel in SAR image is arranged to 0；
Or the value of pixel is more than global threshold in SAR image, then the value of pixel in SAR image is arranged to 1, and will These pixels are marked, and are known as mark point；
Step 3.2：Judge whether the value of all pixels point and global threshold are completeer in SAR image；
If：Judging result is that the value of all pixels point and global threshold are completeer in SAR image, then obtains index square Battle array；
Or judging result is that the value of all pixels point and global threshold have not had also in SAR image, then performs step 3.1；
Step 4：SAR image is divided into four regions, is specifically carried out in accordance with the following steps：
Step 4.1：SAR image is equally divided into four regions；
Step 4.2：The mark point number averagely contained in all mark point numbers and four regions in statistics SAR image, N is denoted as respectivelyf、Nm；
Step 4.3：Repartition SAR image, SAR image be equally assigned into four regions, respectively marked as A1, A2, A3, A4, correspondingly the number of mark point is denoted as N respectively in four regions1、N2、N3、N4；
Step 4.4：Compare N1、N2、N3、N4, it is assumed that N4Maximum, then the most region of mark point is A4 regions；
Step 4.5：Compare N4With 2 × Nm, if N4More than 2 × Nm, then four regions are repartitioned；
Step 4.6：Compare the mark point number in the A2 regions and A3 regions adjacent with A4 regions；
If：N2>N3, then need to repartition A4 regions and A3 regions, i.e., the subregion in A4 regions removed and be fed to In A3 regions；
Otherwise, A4 regions and A2 regions are repartitioned, i.e., the subregion in A4 regions is removed and be fed to A2 regions In；
Step 5：Local detection is carried out at the same time by multithreading to the mark point in four regions after repartitioning, is calculated Local threshold, specifically carries out in accordance with the following steps：
Step 5.1：Centered on the mark point in index matrix, sliding window is established：Target window, protecting window, the back of the body Scape window；
Step 5.2：The pixel value for removing all pixels point in backdrop window is more than the pixel of global threshold；
Step 5.3：Residual pixel point in backdrop window is counted using K distribution probability density models, is calculated Threshold value, that is, local threshold of backdrop window；
Step 5.4：The pixel value of mark point in comparison object window and the threshold size of backdrop window, if target window The pixel value of mark point in mouthful is more than the threshold value of backdrop window, then the mark point in the target window is ship target point, such as The pixel value of mark point in fruit target window is less than the threshold value of backdrop window, then the mark point in the target window is miscellaneous for ocean Ripple；
Step 6：Judge whether view picture SAR image detects to finish；
If：Judging result is that the detection of view picture SAR image finishes, then performs step 7；
Or judging result is that view picture SAR image does not have detection to finish, then these steps 5；
Step 7：The each four connected regions area formed to the pixel locally detected according to step 5 counts, Possible ship is filtered out according to the elemental area of most canoe, is specifically carried out in accordance with the following steps：
Step 7.1：The minimum ship of deck product is defined as：M meters long, N meters of width, the pixel faces of resolution ratio R, then most canoe Product is
Step 7.2：Compare the size of each four connected regions area and the elemental area of most canoe, if four connected regions Area is less than the size of the elemental area of most canoe, then removes four connected region, remaining four connected region is exactly to detect The ship gone out.
Preferably, four regions are rectangular block region.
In the step 5.1, specifically include
Step 5.1.1：Establish target window：Target window is used as using the wherein mark point in index matrix；
Step 5.1.2：Establish protecting window：The edge of protecting window and the Edge Distance of target window are set to a captain 1.5 times of pixel length, be denoted as D1, then D1=1.5 × LP；WhereinL in formulaPFor the pixel length of captain, LmaxFor the maximum length of captain, R is the resolution ratio of captain；
Step 5.1.3：Establish backdrop window：Region area between the edge of backdrop window and the edge of protecting window exists When being set to be set to 1000 pixels under 20000 pixels, low resolution under high-resolution, backdrop window detection is more accurate；If The edge of backdrop window and the Edge Distance of protecting window are D2, then, at high resolutions, D1And D2Relation be Under low resolution, D1And D2Relation be
Advantageous effects caused by the present invention：
The present invention proposes a kind of adaptive quick determination methods of SAR image CFAR based on multithreading, with the prior art Compare, this algorithm is divided into global detection and local detection；The threshold value screening of global detection greatly reduces to be needed in local detection The ships target number accurately detected, while the method for employing multithreads computing is locally detected, significantly carry again The speed of detection algorithm is risen；Last testing result shows, fast algorithm of detecting proposed in this paper and Salazar algorithm phases Than in the case of Detection accuracy is undiminished, detection speed is greatly improved, and is more suitable for applying to Ship targets detection In system.
Brief description of the drawings
Fig. 1 is the FB(flow block) of the SAR image CFAR adaptive quick determination methods of the invention based on multithreading.
Fig. 2 is the schematic diagram of the CFAR detection sliding windows in the present invention.
Fig. 3 is by the schematic diagram of the SAR image quartering.
Fig. 4 (a) is the schematic diagram before SAR image is redistributed.
Fig. 4 (b) is the schematic diagram after SAR image is redistributed.
Fig. 5 (a) is to be schemed using the SAR of COSMO-SkyMed satellites shooting.
Fig. 5 (b) is the global detection result figure for using histogram distribution to draw to Fig. 5 (a).
Fig. 5 (c) is the testing result figure for using Salazar algorithms to draw to Fig. 5 (b).
Fig. 5 (d) is the testing result figure for using fast algorithm to draw to Fig. 5 (b).
Fig. 6 (a) is to be schemed using the SAR of TSX1_SAR satellites shooting.
Fig. 6 (b) is the global detection result figure for using histogram distribution to draw to Fig. 6 (a).
Fig. 6 (c) is the testing result figure for using Salazar algorithms to draw to Fig. 6 (b).
Fig. 6 (d) is the testing result figure for using fast algorithm to draw to Fig. 6 (b).
Embodiment
Below in conjunction with the accompanying drawings and embodiment is described in further detail the present invention：
1st, Fast Algorithm Design thinking source
The adaptive fast algorithm of Ship targets detection proposed in this paper adopts some thoughts in classical CFAR algorithms.(1)VI- The index value used in CFAR (Smith and Varshney, 1997, Smith and Varshney, 2000) can be selected Target point in background clutter region, but index matrix should obtain by adaptive threshold value rather than pass through empirical value Obtain.(2) the CFAR algorithms based on Beta-prime distributions that Salazar (Salazar, 1999) is proposed are in uniform clutter and more Detection result is fine under background environment, so the target point in more background environments single should be carried out local detection, to ensure The accuracy rate of target detection.(3)Bisceglie(Bisceglie and Galdi,2005,Bisceglie and Galdi, 2001) influence of the possible ship target point influential on Background statistic to avoid interference clutter is rejected in the algorithm proposed, This algorithm idea should be applied in local detection.(4), can be with by being detected to the sliding window of target point in index matrix It is to interfere clutter to judge the target point, or ships target point.
The algorithm idea based on more than devises following adaptive fast algorithm of detecting herein：View picture SAR is schemed first As carrying out global detection, a simple statistics with histogram is carried out to the pixel value of SAR image all the points, has set the first order CFAR values, ask for global threshold, and all possible target point is quickly filtered out with index value, is marked, herein by it Referred to as mark point；Then local detection is carried out, sliding window is set to the mark point filtered out in global detection, in background clutter Possible ship target point influential on Background statistic is rejected in window, it is then more accurate to remaining pixel value use K distributed models are counted, and are set second level CFAR values, are asked for local threshold, pass through local threshold and target point pixel value Compare, complete the part detection to target point, which improves the speed locally detected with multithreading；Finally to part The target point of gained makees four connected region area statistics in detection, and possible ship is filtered out according to minimum deck product.
2nd, the idiographic flow of fast algorithm, detailed step (as shown in Figure 1) are described as follows：
(1) SAR image is inputted, asks for global threshold
Ask for global threshold and (be denoted as Tlow) there is two ways, first, carrying out data statistics to view picture SAR image, and pass through Statistics with histogram setting CFAR values ask for global threshold, second, one global threshold of sets itself.Certainly, setting CFAR is asked for TlowEffect is more preferable, also more accurate.
(2) index matrix is obtained
Calculate global threshold TlowAfterwards, the value and global threshold of each pixel are compared.If its pixel value is less than Tlow, then its pixel value is arranged to 0, if its pixel value is more than Tlow, its pixel value is arranged to 1, when all pixels point all After completeer, it is possible to obtain index matrix.Pixel value is that 1 pixel may be exactly object pixel on ship in index matrix Point, so these pixels are marked herein, and referred to as mark point.
(3) by fast algorithm, SAR image is divided into 4 regions, parallel detection is carried out with multithreading
Since K distributed model parametric solutions are more complicated, when the target point of detection is more, speed is very slow, herein in part SAR image is divided into 4 regions during detection, has used multithreading to detect 4 regions at the same time, so as to improve the speed locally detected Degree.
(4) mark point is based on, carries out sliding window setting
As shown in Fig. 2, with the mark dot center in index matrix, three windows are established：Target window (T), protecting window (S), backdrop window (G).
We using one of target pixel points as target window, the edge of protecting window and the edge of target window away from From 1.5 times of the pixel length for being set to a captain, D is denoted as1.Maximum captain is set to 300 meters herein, resolution ratio is denoted as R, The pixel length of captain is denoted as LP, thenBy the experience of test of many times, backdrop window Region area between edge and the edge of protecting window is set to 20000 pixels, low resolution below 15 meters of high-resolution Under be set to 1000 pixels, sliding window detection is more accurate.If the edge of backdrop window and the Edge Distance of protecting window are D2, then, at high resolutions, D1And D2Relation beUnder low resolution, D1And D2 Relation be
(5) local threshold is calculated
In general, at least will be at a distance of distances more than 2 captains between ship and ship.There should not be ship in backdrop window Appearance, so backdrop window count when need remove more than global threshold TlowPixel, to prevent that pixel value is higher Sea clutter is taken as ship detection to come out.Then residual pixel point is counted using K distribution probability density models again, calculated The threshold value for going out backdrop window (is denoted as TG).Pixel value (gray value) (being denoted as T) and threshold value T in last comparison object windowGGreatly It is small, if T>TG, then it is the point on ship to illustrate the pixel, if T<TG, then it is sea clutter to illustrate the pixel.Judge view picture Whether image, which detects, finishes, if detection finishes, performs next step；Finished if do not detected, continue to detect.
(6) ship is filtered out according to four connected region areas
Each four connected regions area of the pixel composition locally detected is counted, is screened according to most canoe elemental area Go out ship.It is recognized herein that long 15 meters wide 4 meters of deck product is minimum, resolution ratio is denoted as R, and the elemental area of most canoe isIf four connected region areas are less than Smin, just the pixel region is removed, remaining pixel region is exactly The ship detected.
3rd, CFAR fast algorithm of detecting principle
3.1st, K distributed models
Compared by research and statistical model discovery, the K being derived from by SAR image product model are commonly used in sea clutter background Distribution can describe the hangover characteristic again of sea clutter distribution well, can be extra large miscellaneous with observation well in a wide range Wave number matches according to amplitude distribution, can meet local auto-adaptive testing requirements (Xing Xiangwei, 2009) in most cases, so The Ship Target Detection based on CFAR is carried out using K distributed models herein.
K distribution probability density function be
Wherein, Γ (*) is Gamma functions,Kγ-LFor the modified Bezier of γ-L the second classes of rank Function；L, γ, μ can be by the parameters of equations for three, and L+ γ -1 are form parameter, determine the sharp journey of curve Degree, and the hangover of the smaller K distributions of shape parameter values is more obvious, therefore describe sea clutter with K distribution curves and be more suitable for.
It is as follows that three parameters L, γ, μ solve expression formula:
Wherein, Φ0(L) it is double Gamma functions, is the logarithmic derivative of gamma function,It is Gamma function more than one；K1、K2、K3For known parameters, K1For average value, k2For variance, K3For variance three times.
3.2nd, CFAR detects basic principle
1) first order CFAR testing principles
Setting first order false alarm rate (is denoted as Pfa), statistics with histogram is made to view picture SAR image all clutters point, can obtain straight Square figure probability density function f (x), solution formula 3.5 can obtain global threshold Tlow.If the pixel value T of pixel in SAR image More than threshold value TlowThen it is considered abnormal in surrounding clutter point, is determined as ships target, is otherwise just judged as ocean clutter (Xing Xiang Wei, 2009), decision rule is as shown in formula 3.6.
Decision rule is：
2) second level CFAR testing principles
The second level CFAR detections of SAR image are mainly to be slided by sliding window in entire image to complete.With Sliding window is established centered on the target point that first order CFAR is detected, rejects and has an impact in backdrop window to Background statistic Possible ship target, then make K distribution statistics to residual spur pixel in backdrop window, obtain K distribution probability density letters Number f2(x), P (is finally denoted as according to second level false alarm ratefa2) and formula 3.5 solve adaptive local detection threshold value TG。
3.3rd, fast algorithm
Although two-stage CFAR algorithms than the adaptive algorithm detection efficiency that is distributed based on K have large increase (Xing Xiangwei, 2009) data calculation amount is bigger during, but due to K distributed model parametric solutions, in detection high-resolution and larger-size SAR During image, speed is still very slow.The main reason for speed is slow when two-stage CFAR algorithms locally detect is the mesh that global detection goes out Mark is relatively more, it is desirable to improves two-stage CFAR detection speeds, it is necessary to improve the speed locally detected.In this regard, herein in detection process In employ multithreading and two methods of mark point screening lift the speed locally detected.
1) multithreading
Multithreading refers to allow an application program at the same time there are two or more threads, for supporting affairs simultaneously Hair and multitasking.Multithreading refers to the technology for realizing that multiple threads are concurrently performed from software or hardware.If SAR image can be divided into several parts by we, this several part is detected at the same time using multithreading, then can be big The local detection speed of amplitude lifting, so multithreading is employed herein to lift detection speed.
Proved by many experiments, the large scale SAR image after global detection is divided into rectangular block is easier to count Calculate, it is also relatively more reasonable.However, it is less obvious that SAR image is divided into the speed-raising of two rectangular blocks, it is divided into three rectangular block number of targets point Cloth be easy to it is uneven, be divided into more than 4 rectangular block again it is very complicated.So as shown in figure 3, SAR image is equally divided into four A rectangular block, is then carried out at the same time local detection to the target point in this four pieces of regions with multithreading, both relatively uniform, Speed can be lifted again, more rationally.
But if the mark point distributed pole after global detection is uneven, institute in four pieces of regions being divided into The mark point quantity difference contained is very big, even if can not be greatly improved with multithreading, speed at this time, so we use The mode of SAR image is repartitioned to lift detection speed.
Concrete mode is as follows：
The mark point number that all mark point numbers and four pieces of zone levelings contain in SAR image is counted first, is remembered respectively For Nf、Nm；
Secondly SAR image is equally assigned into 4 regions, respectively marked as A1, A2, A3, A4, as shown in Fig. 4 (a), phase The number of mark point in the four pieces of regions in ground is answered to be denoted as N1、N2、N3、N4；
Compare N again1、N2、N3、N4, it is assumed that N4The most region of maximum, i.e. mark point is A4 regions；
Then N is compared4With 2 × NmIf N4More than 2 × Nm, then four pieces of regions are repartitioned；
Finally compare the mark point number in the A2 regions and A3 regions adjacent with A4 regions, it is assumed that N2>N3, then need pair A4 regions and A3 regions are repartitioned, i.e., remove in A3 regions 1/3 region and be fed in A4 regions, otherwise to A4 regions and A2 is repartitioned in region.Repartition shown in result such as Fig. 4 (b).
After region is repartitioned, mark tally just can be relatively uniform in 4 regions, then, to 4 areas after repartitioning Domain is carried out at the same time local detection by multithreading, so as to lift the speed locally detected.Finally the testing result in 4 regions is spelled It is connected into the complete SAR testing result figures of a width.
2) mark point is screened
In general, at least will be at a distance of distances more than 2 captains, so the one of a ship target point between ship and ship Be not in necessarily other ships in a captain's distance range.When second level CFAR is detected, mark point follows from left to right, from upper Order under and.When a mark point is confirmed to be ship target point by local detection, it is believed that in this ship target point In one captain's distance range, more than the mark point of its pixel value, it is possible to be directly screened as ship target point, it is not necessary to right It carries out sliding window detection, the calculation times of K distributions can be thus reduced, so as to lift detection rates.
4th, Analysis of test results
Herein its testing result is analyzed by taking two width SAR images as an example.
Fig. 5 (a) is that a width size is the Sea SAR image that 1517 × 1864, pixel resolution is 3 meters, which is For high-resolution radar satellite COSMO_SkyMed in 2007 using the shooting of VV polarization modes, used coordinate system is GCS_ WGS_1984, geographical coordinate are 121.96 ° of longitude, 38.96 ° of latitude, there is 19 ship targets.
Fig. 6 (a) is that a width size is the Sea SAR image that 2028 × 1876, pixel resolution is 4 meters, which is For high-resolution radar satellite TSX1_SAR in 2007 using the shooting of HH polarization modes, used coordinate system is GCS_WGS_ 1984, geographical coordinate is 119.55 ° of longitude, 35.06 ° of latitude, there is 17 ship targets.
Two images are detected using fast algorithm proposed in this paper and Salazar algorithms, two kinds of calculations of comparative study The Detection accuracy and speed of method, analyze testing result.Sliding window size is true by target size and image resolution ratio in experiment It is fixed.
For Fig. 5 (a), target window is 1 × 1, and protecting window is 301 × 301, and backdrop window is 333 × 333, is demanded perfection P during office's threshold valuefa3e-3 is set to, P when seeking local thresholdfa2It is set to 1e-3.
For Fig. 6 (a), target window is 1 × 1, and protecting window is 227 × 227, and backdrop window is 269 × 269.Demand perfection P during office's threshold valuefa1.5e-3 is set to, P when seeking local thresholdfa2It is set to 1e-4.
After obtaining global threshold using histogram distribution, the target point of SAR image can be quickly filtered out.Due to histogram The true distribution of all data of statistic record, and the false alarm rate set is higher, it can be ensured that included in the selection result All target points；As shown in Fig. 5 (b) and 6 (b), 8339 and 5640 target points are filtered out respectively；Then by Fig. 5 (b) and 6 (b) four rectangular areas are divided into, carry out the detection of sliding window to mark point in four regions at the same time with multithreading；Last root Ship is filtered out according to four connected region areas, draws testing result figure.
Result detected by two kinds of algorithms is respectively 3.6G in dominant frequency as shown in Fig. 5 (c), 5 (d) and Fig. 6 (c), 6 (d) It is respectively 192 seconds, 70 seconds and 140 seconds, 50 seconds inside to save as detection time on the computer of 8G.
By the comparison of two kinds of algorithms in Tables 1 and 2, two kinds of algorithms can detect all ships, but originally The detection speed for the fast algorithm that text proposes is substantially faster than Salazar algorithm.
So fast algorithm proposed in this paper is more suitable for applying in Ship targets detection system.
Table 1 COSMO-SkyMed satellite images, two kinds of algorithm testing results compare
Table 2 TSX1_SAR satellite images, two kinds of algorithm testing results compare
5th, conclusion and prospect
This paper presents a kind of SAR image naval vessel based on multithreading to detect adaptive fast algorithm, this algorithm is divided into entirely Office's detection and local detection.The threshold value screening of global detection greatly reduces the ships target for needing accurately to detect in local detection Number, while the method for employing multithreads computing is locally detected, the speed of detection algorithm is significantly improved again. Last testing result shows that fast algorithm of detecting proposed in this paper does not drop compared with Salazar algorithms in Detection accuracy In the case of low, detection speed is greatly improved, and is more suitable for applying in Ship targets detection system.
But since military combat is very high to the requirement of real-time of Ship Target Detection, so also needing to further visit Rope CFAR detection algorithms, until fully meeting the requirement that can be handled in real time oceanographic data.
Certainly, described above is not limitation of the present invention, and the present invention is also not limited to the example above, this technology neck The variations, modifications, additions or substitutions that the technical staff in domain is made in the essential scope of the present invention, should also belong to the present invention's Protection domain.

Claims (2)

  1. A kind of 1. adaptive quick determination methods of SAR image CFAR based on multithreading, it is characterised in that：In accordance with the following steps into OK：
    Step 1：Input SAR image；
    Step 2：Global detection is carried out to view picture SAR image, CFAR values are set by statistics with histogram or customizing method is asked Take global threshold；
    Step 3：Index matrix is obtained, is specifically carried out in accordance with the following steps：
    Step 3.1：Compare the value of each pixel and the global threshold calculated by step 2 in SAR image；
    If：The value of pixel is less than global threshold in SAR image, then the value of pixel in SAR image is arranged to 0；
    Or the value of pixel is more than global threshold in SAR image, then the value of pixel in SAR image is arranged to 1, and by these Pixel is marked, and is known as mark point；
    Step 3.2：Judge whether the value of all pixels point and global threshold are completeer in SAR image；
    If：Judging result is that the value of all pixels point and global threshold are completeer in SAR image, then obtains index matrix；
    Or judging result is that the value of all pixels point and global threshold have not had also in SAR image, then step 3.1 is performed；
    Step 4：SAR image is divided into four regions, is specifically carried out in accordance with the following steps：
    Step 4.1：SAR image is equally divided into four regions；
    Step 4.2：The mark point number averagely contained in all mark point numbers and four regions in statistics SAR image, respectively It is denoted as Nf、Nm；
    Step 4.3：Repartition SAR image, SAR image be equally assigned into four regions, respectively marked as A1, A2, A3, A4, correspondingly the number of mark point is denoted as N respectively in four regions1、N2、N3、N4；
    Step 4.4：Compare N1、N2、N3、N4, it is assumed that N4Maximum, then the most region of mark point is A4 regions；
    Step 4.5：Compare N4With 2 × Nm, if N4More than 2 × Nm, then four regions are repartitioned；
    Step 4.6：Compare the mark point number in the A2 regions and A3 regions adjacent with A4 regions；
    If：N2>N3, then need to repartition A4 regions and A3 regions, i.e., the subregion in A4 regions removed and be fed to A3 areas In domain；
    Otherwise, A4 regions and A2 regions are repartitioned, i.e., the subregion in A4 regions is removed and be fed in A2 regions；
    Step 5：Local detection is carried out at the same time by multithreading to the mark point in four regions after repartitioning, is calculated local Threshold value, specifically carries out in accordance with the following steps：
    Step 5.1：Centered on the mark point in index matrix, sliding window is established：Target window, protecting window, background window Mouthful；
    Step 5.2：The pixel value for removing all pixels point in backdrop window is more than the pixel of global threshold；
    Step 5.3：Residual pixel point in backdrop window is counted using K distribution probability density models, calculates background Threshold value, that is, local threshold of window；
    Step 5.4：The pixel value of mark point in comparison object window and the threshold size of backdrop window, if in target window Mark point pixel value be more than backdrop window threshold value, then the mark point in the target window is ship target point, if mesh The pixel value for marking the mark point in window is less than the threshold value of backdrop window, then the mark point in the target window is ocean clutter；
    Step 6：Judge whether view picture SAR image detects to finish；
    If：Judging result is that the detection of view picture SAR image finishes, then performs step 7；
    Or judging result is that view picture SAR image does not have detection to finish, then these steps 5；
    Step 7：The each four connected regions area formed to the pixel locally detected according to step 5 counts, according to The elemental area of most canoe filters out possible ship, specifically carries out in accordance with the following steps：
    Step 7.1：The minimum ship of deck product is defined as：M meters long, N meters of width, resolution ratio R, then the elemental area of most canoe be
    Step 7.2：Compare the size of each four connected regions area and the elemental area of most canoe, if four connected region areas Less than the size of the elemental area of most canoe, then four connected region is removed, remaining four connected region is exactly to detect Ship.
  2. 2. the SAR image CFAR adaptive quick determination methods according to claim 1 based on multithreading, its feature exist In：Four regions are rectangular block region.
CN201610173184.0A 2016-03-24 2016-03-24 The adaptive quick determination methods of SAR image CFAR based on multithreading Active CN105759268B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610173184.0A CN105759268B (en) 2016-03-24 2016-03-24 The adaptive quick determination methods of SAR image CFAR based on multithreading

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610173184.0A CN105759268B (en) 2016-03-24 2016-03-24 The adaptive quick determination methods of SAR image CFAR based on multithreading

Publications (2)

Publication Number Publication Date
CN105759268A CN105759268A (en) 2016-07-13
CN105759268B true CN105759268B (en) 2018-04-13

Family

ID=56346395

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610173184.0A Active CN105759268B (en) 2016-03-24 2016-03-24 The adaptive quick determination methods of SAR image CFAR based on multithreading

Country Status (1)

Country Link
CN (1) CN105759268B (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107045126A (en) * 2017-03-06 2017-08-15 中国人民解放军国防科学技术大学 A kind of synthetic aperture radar movement overseas Ship Target Detection method
CN107942329B (en) * 2017-11-17 2021-04-06 西安电子科技大学 Method for detecting sea surface ship target by maneuvering platform single-channel SAR
CN108734111A (en) * 2018-04-26 2018-11-02 西南电子技术研究所（中国电子科技集团公司第十研究所） SAR image surface vessel recognition methods
CN108985288B (en) * 2018-07-17 2022-06-14 电子科技大学 TGMSERs-based SAR image oil spill detection method
CN110895329B (en) * 2019-07-12 2022-12-27 西安电子科技大学 Hybrid distribution model clutter map target detection method and device
CN111340677B (en) * 2020-02-27 2023-10-27 北京百度网讯科技有限公司 Video watermark detection method, apparatus, electronic device, and computer readable medium
KR20220105102A (en) * 2021-01-19 2022-07-26 주식회사 에스아이에이 Method and apparatus for processing radar image
CN113379695B (en) * 2021-06-01 2024-03-29 大连海事大学 SAR image offshore ship detection method based on local feature differential coupling
CN114355346B (en) * 2021-12-29 2023-01-20 北京卫星信息工程研究所 Multi-satellite transceiving networking SAR system and ultra-large-width sea surface ship target imaging method
CN114578384B (en) * 2022-05-07 2022-09-09 成都凯天电子股份有限公司 Self-adaptive constant false alarm detection method for laser atmospheric system
CN117078501B (en) * 2023-10-17 2023-12-29 中科星图深海科技有限公司 Method for improving two-dimensional imaging efficiency of marine environment data

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102200575A (en) * 2010-12-02 2011-09-28 南京大学 Image ship detection method based on constant false alarm rate
CN102968799A (en) * 2012-12-12 2013-03-13 北京航空航天大学 Integral image-based quick ACCA-CFAR SAR (Automatic Censored Cell Averaging-Constant False Alarm Rate Synthetic Aperture Radar) image target detection method
CN105005983A (en) * 2015-04-13 2015-10-28 西南科技大学 SAR image background clutter modeling and target detection method
CN105160648A (en) * 2014-11-26 2015-12-16 中国人民解放军第二炮兵工程大学 Radar target and shadow segmentation method based on wavelet and constant false alarm rate

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102200575A (en) * 2010-12-02 2011-09-28 南京大学 Image ship detection method based on constant false alarm rate
CN102968799A (en) * 2012-12-12 2013-03-13 北京航空航天大学 Integral image-based quick ACCA-CFAR SAR (Automatic Censored Cell Averaging-Constant False Alarm Rate Synthetic Aperture Radar) image target detection method
CN105160648A (en) * 2014-11-26 2015-12-16 中国人民解放军第二炮兵工程大学 Radar target and shadow segmentation method based on wavelet and constant false alarm rate
CN105005983A (en) * 2015-04-13 2015-10-28 西南科技大学 SAR image background clutter modeling and target detection method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
SAR图像舰船目标双阈值快速CFAR检测算法;曲长文 等;《舰船科学技术》;20151115(第11期);第136-140页 *

Also Published As

Publication number Publication date
CN105759268A (en) 2016-07-13

Similar Documents

Publication Publication Date Title
CN105759268B (en) The adaptive quick determination methods of SAR image CFAR based on multithreading
Yang et al. Ship detection from optical satellite images based on sea surface analysis
CN106170819B (en) A kind of diameter radar image Ship Target rapid detection method
Tao et al. Robust CFAR detector based on truncated statistics in multiple-target situations
CN105354541B (en) The SAR image object detection method of view-based access control model attention model and constant false alarm rate
CN102968799B (en) Integral image-based quick ACCA-CFAR SAR (Automatic Censored Cell Averaging-Constant False Alarm Rate Synthetic Aperture Radar) image target detection method
CN104408482B (en) A kind of High Resolution SAR Images object detection method
CN109284786B (en) SAR image terrain classification method for generating countermeasure network based on distribution and structure matching
CN107808383B (en) Rapid detection method for SAR image target under strong sea clutter
CN105160648B (en) Radar target and Shadow segmentation method based on small echo and constant false alarm rate
CN109242870A (en) A kind of sea horizon detection method divided based on image with textural characteristics
CN101727662A (en) SAR image nonlocal mean value speckle filtering method
CN102096824A (en) Multi-spectral image ship detection method based on selective visual attention mechanism
US10497128B2 (en) Method and system for sea background modeling and suppression on high-resolution remote sensing sea images
CN107784655A (en) A kind of visual attention model SAR naval vessels detection algorithm of adaptive threshold
CN103778613A (en) Polarization SAR image filtering method for window self-adaptation
CN114821358A (en) Optical remote sensing image marine ship target extraction and identification method
CN112435249A (en) Dynamic small target detection method based on periodic scanning infrared search system
CN107729903A (en) SAR image object detection method based on area probability statistics and significance analysis
CN112329677A (en) Remote sensing image river target detection method and device based on feature fusion
Su et al. Ship detection based on RetinaNet-plus for high-resolution SAR imagery
CN108596844A (en) Background suppression method for playing big gun Remote Control Weapon Station
CN108985292A (en) A kind of SAR image CFAR object detection method and system based on multi-scale division
CN113963178A (en) Method, device, equipment and medium for detecting infrared dim and small target under ground-air background
US20220148288A1 (en) Method for detecting objects

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