A Dual-Threshold ATI-SAR Approach for Detecting Slow Moving Targets

Transcription

1 A Dual-Threshold ATI-SAR Approah for Deteting Slow Moving Targets Yuhong Zhang, Ph. D., Stiefvater Consultants Abdelhak Hajjari, Ph. D. Researh Assoiates for Defense Conversion In. Kyungjung Kim, Ph. D., Stiefvater Consultants Braham Himed, Ph. D., Air Fore Researh Lab/SNRT Key Words: along-trak, interferometry, syntheti aperture radar (SAR), ground moving target indiation (GMTI) ABSTRACT The high false alarm assoiated with onventional alongtrak interferometri syntheti aperture radar (ATI-SAR) is a big onern for any valuable military radar. To redue the false alarm rate, this paper proposes a dual-threshold approah that ombines the onventional interferometri phase detetion with the SAR image amplitude detetion. This yields two results: () the interferometri phase map (inluding target veloity information) obtained by applying the interferometri phase detetion only to the pixels seleted by the amplitude detetion, and (2) the amplitude map (inluding target strength information) obtained by applying the amplitude detetion only to the pixels seleted by the interferometri phase detetion. The onept is illustrated by the results obtained using the Jet Propulsion Laboratory s (JPL) AirSAR ATI data, olleted in the Monterey Bay area, California. The ATI-SAR proessing requires also a preise alibration of the platform s rab angle. This paper presents a simple blind-alibration method that does not require any knowledge of the bakground and/or the atual rab angle.. INTRODUCTION The ATI-SAR tehnique has been proven valuable to sense the earth-surfae motion suh as oean surfae urrents. Reently, there has been inreasing interest in applying ATI- SAR tehniques to detet slow moving target detetion, espeially ground moving target indiation (GMTI) using spae-based assets []. ATI-SAR is based on the aquisition of two omplex SAR images taken under idential geometries separated by a short time interval. The phase differene between the two interferometri images is used as a test statisti to be ompared with a deision threshold. The question then is whether ATI-SAR ould be used for deteting slow moving targets. In this paper, we propose to examine the detetion performane of onventional ATI-SAR. It is shown that the high false alarm rate assoiated with this tehnique would be a big onern for any valuable military radar. To redue the false alarm rate, this paper proposes a dual-threshold approah that ombines the onventional interferometri phase detetion with the SAR image amplitude detetion. Strong pixels ould ontain moving targets, stationary objets, and other disretes. The amplitude detetion suppresses the weak pixels from large smooth surfaes suh as road and water surfaes. This yields two important results: () the interferometri phase map (inluding target veloity information) obtained by applying the interferometri phase detetion only to the pixels seleted by the amplitude detetion (2) the amplitude map (inluding target strength information) obtained by applying the amplitude detetion only to the pixels seleted by the interferometri phase detetion. The onept is illustrated by the results obtained using JPL s AirSAR ATI data [2], olleted in the Monterey Bay, CA area. The ATI-SAR proessing requires also a preise alibration of the platform s rab angle. Referene [2] uses a known stationary orner refletor (strong satterer) array as the referene for the alibration. This paper presents a blind alibration method that does not require any knowledge of the ground referene satterer and/or the atual rab angle. 2. DETECTION PERFORMANCE OF CONVENTIONAL ATI-SAR The ATI-SAR method is based on the aquisition of two omplex SAR images (A and B), taken under idential geometries separated by a short time interval, with the interferometri phase being used as a test statisti, as shown in Figure. Aft-Antenna B x Fore-Antenna Fore-Antenna, Image A, t = t Aft-Antenna, Image B, t = t + T, where T = B / V x The interferometri phase: 4π B x v = V, where V is the target radial speed. t t λ Vp v η, H Test Statisti: v < η, H Figure. Priniples of onventional ATI-SAR. V p p

2 Report Doumentation Page Form Approved OMB No Publi reporting burden for the olletion of information is estimated to average hour per response, inluding the time for reviewing instrutions, searhing existing data soures, gathering and maintaining the data needed, and ompleting and reviewing the olletion of information. Send omments regarding this burden estimate or any other aspet of this olletion of information, inluding suggestions for reduing this burden, to Washington Headquarters Servies, Diretorate for Information Operations and Reports, 25 Jefferson Davis Highway, Suite 24, Arlington VA Respondents should be aware that notwithstanding any other provision of law, no person shall be subjet to a penalty for failing to omply with a olletion of information if it does not display a urrently valid OMB ontrol number.. REPORT DATE MAY REPORT TYPE N/A 3. DATES COVERED - 4. TITLE AND SUBTITLE A Dual-Threshold ATI-SAR Approah for Deteting Slow Moving Targets 5a. CONTRACT NUMBER 5b. GRANT NUMBER 5. PROGRAM ELEMENT NUMBER 6. AUTHOR(S) 5d. PROJECT NUMBER 5e. TASK NUMBER 5f. WORK UNIT NUMBER 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) Stiefvater Consultants 8. PERFORMING ORGANIZATION REPORT NUMBER 9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES). SPONSOR/MONITOR S ACRONYM(S) 2. DISTRIBUTION/AVAILABILITY STATEMENT Approved for publi release, distribution unlimited. SPONSOR/MONITOR S REPORT NUMBER(S) 3. SUPPLEMENTARY NOTES See also ADM27. Proeedings of the 25 IEEE International Radar Conferene Reord Held in Arlington, Virginia on May 9-2, 25. U.S. Government or Federal Purpose Rights Liense., The original doument ontains olor images. 4. ABSTRACT 5. SUBJECT TERMS 6. SECURITY CLASSIFICATION OF: 7. LIMITATION OF ABSTRACT UU a. REPORT unlassified b. ABSTRACT unlassified. THIS PAGE unlassified 8. NUMBER OF PAGES 5 9a. NAME OF RESPONSIBLE PERSON Standard Form 298 (Rev. 8-98) Presribed by ANSI Std Z39-8

3 2. False Alarm Rate The probability of false alarm (or false alarm rate), P FA, an be alulated from the probability density funtion (pdf) of the interferometri phase in the absene of target. For simpliity, onsider a Gaussian lutter assoiated with a homogeneous bakground and additive white Gaussian thermal noise. In the absene of target, the interferometri phase is the produt of two omplex Gaussian orrelated signals. The resulting pdf an then be expressed analytially [3] and is stritly dependent on the equivalent orrelation oeffiient given by γ γ = () + CNR where γ is the lutter orrelation oeffiient and CNR is the lutter-to-noise power ratio. Figure 2 shows the pdf of the interferometri phase in absene of target, for different CNR values and γ = (whih represents the best ase for ATI-SAR). The orresponding false alarm rates versus (vs.) the phase threshold are shown in Figure 3. The numerial results are listed in Table. Table P FA vs. CNR and phase threshold η (rad.) for γ =. CNR\ η db db db db db Probability of Detetion The pdf in the presene of target is required for omputing the probability of detetion P D. Unfortunately; this pdf is not analytially available. A Monte Carlo simulation is used in this paper. Figure 4 shows the pdf plots with, trials for three different values of signal-to-lutter ratio (SCR), for a CNR value of 2dB and γ =, and the interferometri phase indued by the target equal to 2 radians. The orresponding P D plots vs. the phase threshold are shown in Figure 5. 9 CNR=dB CNR=dB CNR=2dB CNR=3dB.6.4 SCR = -db SCR = db SCR = db pdf(φ) γ = pdf(φ).2 γ = φ [rad] Figure 2. Phase noise pdf for different CNR values and γ = φ [rad] Figure 4. Phase pdf for three different SCR values in presene of a moving target P FA γ = CNR = db CNR = db CNR = 2 db CNR = 3 db Phase Threshold Figure 3. Probability of false alarm for different CNR values and γ = P D γ = SCR = -db SCR = db SCR = db Phase Threshold Figure 5. Detetion probability for three different SCR values in the presene of a moving target

4 For example, if η =.5 radians, then P D = 6 for a db SCR and P D =.36 for a db SCR. From Table, we see that the orresponding false alarm rate is P FA = 5.4x -3, whih is too high for almost any radar appliation. angle, whih leads to an offset range in ross-trak diretion: 3. DUAL-THRESHOLD DETECTION This paper proposes to set another amplitude threshold ηa to the SAR amplitude image. Strong pixels ould ontain moving targets, stationary objets, and other disretes. The amplitude detetion suppresses the weak pixels from large smooth surfaes suh as road and water surfaes. This detetion is similar to the onventional onstant false alarm rate (CFAR) proessing. There are many algorithms to determine the threshold [4]. The performane depends on the environments. As an example, the threshold is ounted relative to the mean amplitude for bakground in this paper, whih is estimated by taking the root mean square (rms) mean value of the median amplitudes (of the orresponding slant-range pixels) over ross-range pixels. For example, setting the amplitude threshold as η a = db means those pixels whose amplitudes are larger than the estimated mean value by db are hosen as andidate target pixels. Jointly using the two thresholds (phase and amplitude), we get two outputs: () The interferometri phase map (target veloity) is obtained by applying the interferometri phase detetion only to the pixels seleted by the amplitude detetion. Speifially, the phase at a pixel will be fored to zero if its image amplitude is below a pre-determined threshold. And, (2) The amplitude map (target strength) is obtained by applying the amplitude detetion only to the pixels seleted by the interferometri phase detetion. This onept will be illustrated using JPL s AirSAR ATI data. 4. AIRSAR ATI DATA The AirSAR ATI data used in this paper was reorded at L-band in Monterey Bay, California area. The AirSAR system onsists of two ATI antennas separated by a distane of m. The strip-map SAR images are obtained using the line-byline imaging method with N=24 pulses in a oherent proessing interval (CPI), whih orresponds to approximately a 3.5m ross-range resolution. Figure 6 shows the SAR image (4 slant range ells for.32km, inluding roads, highways and water) from the data reorded on the Fore-Antenna. 4. Blind Calibration for Group Phase Shift Indued by Crab Angle In the ideal ase, two ATI-antennas are aligned with the moving trak. However, the rab angle (yaw and pith) of platform makes two antennas offset from the moving trak. Figure 7 illustrates the ase where the platform has a yaw B Figure 6. Image A for seleted area. B x Crab Angle A B Figure 7. A ross-trak offset between two antennas indued by the rab (yaw) angle. y = B x sin, (2) y = B x sin where B x is the baseline distane. The ross-trak offset of two-antennas will indue a phase shift between Images A and B, alled group phase shift, whih must be alibrated before the ATI-SAR proessing. Figure 8 shows the interferometri phase image before the phase alibration. Obviously, phases of most pixels are far from zero. It is well known that the rab angle varies during the platform motion [2]. Referene [2] uses a known stationary orner refletor (strong satterer) array on the ground as a referene to alibrate the phase shift. In this paper, we propose a simple blind alibration method that does not require any knowledge of the ground referene satterers and the atual rab angle. The proposed method first estimates the group phase shift as a funtion of ross range, and then ompensates for this phase shift. Several methods an be used for onduting this estimation proedure. The median phase as a funtion of ross-range is first alulated and an autoregressive (AR) smoothing proessing is then applied to the median phases. The AR-smoothed phase shift is used as an estimate of the true one for alibration purposes. However, it is seen that one estimation and ompensation proedure is usually not enough, beause the estimation of a group phase shift is based on wrapped interferometri phases vp

5 (within [ π, π] ), and the ompensated phases are wrapped again, whih leads to a new group phase shift. An iterative proedure is used in this paper to bypass this problem. It is found that only a few (usually 3 to 5) iterations are needed in most ases. Figure 9 shows the median and AR-smoothed group phase shift as a funtion of ross-range before phase alibration while Figure shows the orresponding results after alibration. The interferometri phase image after phase ompensation is shown in Figure. Clearly, the phases of most pixels are now lose to zero when ompared to Figure 8. Figure. Interferometri phase image after phase alibration 5. RESULTS WITH AIRSAR DATA 5. Phase Detetion Only Figure 8. Interferometri phase image before phase alibration Estimated Group Phase Shift (rad.) Median Phase AR Smoothed Phase detetion only is the onventional ATI-SAR method. Figure 2 shows the orresponding results with η =.5 radians. The tidal waves are learly shown in this image. However, the false alarm rate is too high for surveillane radar appliations suh as Ground Moving Target Indiation (GMTI) Cross Range (km) Figure 9. Estimated group phase shift as a funtion of ross range before phase alibration Median Phase AR Smoothed Estimated Group Phase Shift (rad.) Cross Range (km) Figure. Estimated group phase shift as a funtion of ross range after phase alibration Figure 2. Interferometri phase image with η =. 5 radians 5.2 Amplitude Detetion Only Figure 3 shows the SAR amplitude image trimmed with an amplitude threshold ( η a = db ) only. This image shows stronger pixels that ould possibly ontain moving targets, stationary objets, and other disretes. The amplitude-only detetion suppresses the weak pixels suh as those orresponding to road and water surfaes.

6 possible moving targets. The loations shown in these Figures are shifted from their real ones due to the SAR proessing. It is possible to restore the real loations of those slow targets without any interferometri phase ambiguity. Figure 3. SAR amplitude image with η db 5.3 Dual-Threshold Detetion a = The dual threshold detetion approah proposed in this paper ombines the detetion results from the above phase and amplitude detetion methods. Hene, two important parameters are obtained: phase and amplitude, or veloity and strength. Figure 4 shows the results of the interferometri phase map obtained by applying the amplitude detetion results of Figure 3 onto those of Figure 2, i.e., foring the phase at a pixel to zero if its amplitude is less than the mean value by an amplitude threshold ( η a = db here). Clearly, the false alarm rate is dramatially redued using this approah. Figure 5. SAR amplitude map with η =. 5 radians and η a = db 6. CONCLUSIONS Conventional ATI-SAR approahes an detet targets with very low radial speeds, but their false alarm rate is too high if they are to be used in surveillane radars. The proposed dualthreshold approah, whih ombines the onventional interferometri phase detetion and the SAR image amplitude detetion, an effetively redue the false alarm rate. The onept of the dual-threshold approah is illustrated using JPL's AirSAR ATI data. This data is alibrated using a simple blind-alibration proedure. However, this is only a very first try. Future work would inlude the determination of thresholds and ways to ombine them. 7. ACKNOWLEDGEMENT The authors would like to thank Dr. Elaine Chapin of JPL for her help in providing and using the AirSAR ATI data. 8. REFERENCES Figure 4. Interferometri phase map with η =. 5 radians and η a = db. Similarly, we an obtain an amplitude map of potential moving targets by applying the phase detetion results of Figure 2 onto those of Figure 3, i.e., eliminating the pixels whose phases are below the phase threshold ( η =. 5 radians here), as shown in Figure 5, whih orresponds to the interferometri phase map of Figure 4. In other words, Figure 4 and Figure 5 show veloity and strength information of. C. W. Chen, Performane assessment of along-trak interferometry for deteting ground moving targets, Pro. 24 IEEE Radar Conf., Philadelphia, PA, April 26-29, D. A. Imel, "AIRSAR along-trak interferometry data," AIRSAR Earth Siene and Appliations Workshop, 4-6 Marh R. Bamler and P. Hartl, "Syntheti aperture radar interferometry," Inverse Problems, vol. 4, R-R54, R. Nitzberg, Radar Signal Proessing and Adaptive Systems, Arteh House, In., Norwood, MA, 9999.