/$ IEEE

Transcription

1 222 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 46, NO. 1, JANUARY 2008 Correction of Attitude Fluctuation of Terra Spacecraft Using ASTER/SWIR Imagery With Parallax Observation Yu Teshima and Akira Iwasaki Abstract Accurate attitude estimation of spacecrafts is the main requirement to provide good geometric performance of remote-sensing imagery. The Advanced Spaceborne Thermal Emission and Reflection Radiometer/short-wave-infrared subsystem has six linear-array sensors arranged in parallel, and each line scans the same ground target with a time interval of ms between neighboring bands. The registration performance between bands becomes worse when attitude fluctuation occurs during a time lag between observations. Since the time resolution of the line scan is higher than that of the attitude information provided from the satellite, attitude data are estimated with a high frequency. We succeeded in correcting the image-registration error using the revised attitude information. As a result, the image distortion of 0.2 pixels caused by spacecraft-attitude jitter is reduced to less than 0.08 pixels, showing that band-to-band registration errors of a sensor with parallax observation are available to improve the image distortion caused by attitude fluctuation. Index Terms Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), attitude estimation, precise geometric correction, short-wave infrared (SWIR). TABLE I ASTER SPECTRAL PASSBAND I. INTRODUCTION THE GEOMETRIC performance of images acquired by Earth-observation sensors is important from the view of geolocation and band-to-band registration accuracy, but it is influenced by the accuracy of the spacecraft position and attitude; evaluation has been carried out for the imagery obtained by various sensors [1] [6]. The attitude is estimated by interpolating the measurement data of star sensors and gyros, which are sampled at a low frequency. When the sampling rate of attitude information is low, compared with the attitude jitter with a high frequency, the images processed by the geometric correction suffer from distortion. Therefore, additional information is necessary to estimate attitude accurately. One solution is to estimate the correct attitude of the spacecrafts using ground control points (GCPs), which needs preparation of the GCPs Manuscript received March 12, 2007; revised August 2, This work was supported in part by the Ministry of Education, Science, Sports and Culture Grant-in-Aid for Scientific Research (B), , Y. Teshima was with the Department of Aeronautics and Astronautics, University of Tokyo, Tokyo , Japan. He is now with Toshiba Electronics Engineering Corporation, Tokyo , Japan. A. Iwasaki is with the Department of Aeronautics and Astronautics, University of Tokyo, Tokyo , Japan ( a.iwasaki@aist.go.jp). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /TGRS in the scene and depends on the accuracy of the GCPs [7]. Another solution is to use high-performance attitude sensors, such as angular displacement sensor (ADS) [8], to obtain the attitude information with a high time resolution and a high angular resolution. The other possible solution is to add an image sensor on the focal plane [9]. However, these methods require additional high-performance sensors to obtain an accurate attitude information. Therefore, it is necessary to develop a software methodology with image processing and optimization techniques of estimating the correct attitude information of the spacecraft under operation in order to obtain geometrically accurate images. The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) is a multispectral imager onboard the Terra (EOS-AM1) spacecraft which was launched in December 1999 [10], [11]. The Terra spacecraft is in a sunsynchronous orbit at an altitude of 705 km with a local equatorial crossing time of 10:30. The repeat cycle is 16 days, and the distance between neighboring orbits is 172 km. The ASTER sensor is composed of the following three subsystems: visible-and-near-infrared (VNIR), short-wave-infrared (SWIR), and thermal-infrared radiometers. Table I shows the spectral passband of each subsystem. The VNIR subsystem has two telescopes: one at the nadir and the other with backward viewing to construct a digital elevation model (DEM). Parallax error does not exist in the VNIR because /$ IEEE

2 TESHIMA AND IWASAKI: CORRECTION OF ATTITUDE FLUCTUATION OF TERRA SPACECRAFT 223 Fig. 1. (a) Configuration of the ASTER/SWIR sensor. (b) Timetable of sampling. a dichroic filter divides incident light into each VNIR band. Since the ASTER sensor is a complex system of three subsystems with four telescopes, a level-1 data-processing algorithm based on the configuration of the ASTER sensor has been constructed [4]. The SWIR radiometer is a sensor with six bands in the SWIR region ( µm) for discriminating rock and minerals. The SWIR images are obtained by means of a pushbroom system of 2048 linear-array detectors of a PtSi charge-coupled device, which operates at liquid nitrogen temperature and must be cooled mechanically. Therefore, the linear-array detectors are arranged in parallel on the focal plane, as shown in Fig. 1(a). Since the distance between the neighboring bands is 1.33 mm, each line scans the same ground target with a time difference of about ms (nominal), as shown in Fig. 1(b). The registration performance between the bands becomes worse when attitude fluctuation occurs during a time lag between observations. The Terra spacecraft provides the attitude data with a resolution of 1 arcsec and the attitude rate data with a resolution of 0.5 arcsec/s every s, which is the major clock of the spacecraft information. Images are corrected geometrically using the attitude data, which are interpolated in third-order polynomials. Therefore, we cannot accommodate the attitude data with a high accuracy when fluctuation occurs at a higher frequency. The angular error of one arcsec corresponds to about 3.4 m on the ground for the Terra spacecraft. Since the groundsampling distance (GSD) of SWIR is 30 m, the angular error of 1 arcsec leads to a registration error between bands of 0.1 pixels or less. The sampling interval of each line of the SWIR sensor is ms to meet the GSD of 30 m, and the time resolution is considerably higher than that of the spacecraft-attitude data. When we utilize the registration error between the bands of the SWIR images, the attitude information will be revised with a higher sampling rate. In this paper, we investigated the satellite-attitude fluctuations using SWIR images. We developed a software methodology to reduce the image distortion based on the revised attitude data that are estimated using the band-to-band registration of the SWIR sensor. II. DETECTION OF THE IMAGE-REGISTRATION ERROR USING IMAGE CORRELATION Normalized cross correlation is used as the criterion of the registration error between images [3], [5], [6]. The covariance, Fig. 2. (a) SWIR scene on June 7, 2002, at Tokyo. Relative registration-error map of band 5 relative to band 6 (b) in the cross-track direction and (c) in the along-track direction. (d) Relative registration-error map of band 5 relative to band 6 in the cross-track direction after correction (correlation window size: 7 7). which is calculated using (1), has a maximum value when two images match each other C(m, n) Ly Lx{ } { } S(i, j) S D(i m, j n) D j i =. Ly Lx{ } Ly 2 Lx{ } 2 S(i, j) S D(i m, j n) D j i Here, S(i, j) and D(i, j) denote the pixel values of the source and destination images for the pixel coordinates of (i, j), respectively. Lx and Ly denote the size of the correlation window in the cross-track and along-track directions. S and D denote the average pixel value within the correlation window. The registration error is obtained by the following steps. The integer set with the maximum value of correlation is first selected as the candidates for misregistration between two images. Next, the highest correlation point is obtained in a subpixel level by parabola fitting to the neighboring correlation values and by determining the vertex value. Fig. 2(a) shows the nadir-looking SWIR image of level-1b data products analyzed in this paper, which was acquired on June 7, 2002, in Tokyo. The level-1b data products are pathoriented images, which are projected to the map using the radiometric calibration and geometric correction coefficients for resampling. Thus, the line of the level-1b data products in the horizontal direction is almost equal to the scan line. It should be noted that the level-1b data products are not orthorectified images. The acquisition duration of one scene is about 9 s, which corresponds to 2100 lines of data. We used bands 5 and 6 images because they are placed at neighboring positions on the focal plane, and the spectral characteristics are sufficiently j i (1)

3 224 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 46, NO. 1, JANUARY 2008 Fig. 3. Averaged relative registration error of band 5 relative to band 6 (a) before and (b) after correction. similar to be suitable for correlation. The registration error is calculated separately in the cross- and along-track directions. Fig. 2(b) shows the registration-error map of band 5 relative to band 6 in the cross-track direction. The relative registration error is plotted on the center pixel of the correlation window, the size of which is set to 7 (cross-track) 7 (along-track). The window size is one-third of that used in the parallax correction of the level-1 data processing and is used only to evaluate the registration error in this scene. The bright pixel denotes that the band 5 pixel deviates to the right of the image. The black pixel denotes a point at which a good correlation is not obtained, such as a water area. It can be seen that a periodic misregistration pattern occurs in the error map. The relative registration error between the bands is due to time-dependent jitter on the line of sight, which is attributable to attitude variation. This means that the absolute registration error of the acquired image relative to the accurate geographical position exists. It should be noted that the fluctuation plotted in Fig. 2(b) corresponds to the relative registration error between bands. In the ASTER system, band 6 represents the SWIR subsystem, and the geometric calibration relative to the ASTER/VNIR is performed intensively. The smaller time interval between the bands is preferable, and the spectral of bands is so close that band 5 is in good correlation with band 6. Therefore, we use bands 5 and 6 in this paper. Fig. 3(a) shows the time course of the relative deviation between the bands in the cross-track direction. The relative registration errors are calculated using the window size of 501 (cross-track) 1 (along-track) and are averaged in the cross-track direction, which is almost equivalent to one scan line because a path-oriented image is used. The frequency and the amplitude of the jitter are about 1.5 Hz and pixels, which corresponds to 6 9 m on the ground, respectively. This jitter corresponds to the roll fluctuation of the Terra satellite. Since the attitude sensors of the Terra spacecraft cannot follow the vibration of this frequency, the relative deviation between the bands remains. The registration error in the along-track direction also exhibits a periodic pattern, as shown in Fig. 2(c). This phenomenon corresponds to the pitch component of the attitude jitter, which is more stable than the roll component because a high stability is required on the satellite for DEM production. Furthermore, the parallax error in the along-track direction is corrected during level-1 data processing; the parallax error in the along-track direction is corrected by image matching between bands 6 and 7 at every lattice point of the image blocks [4], [5]. For these reasons, the amplitude of the relative registration error in the along-track direction is smaller than that in the cross-track direction. Henceforth, we consider only the jitter in the cross-track direction in this paper. However, the uncorrected relative deviation in the along-track direction can be detected with the frequency of about 1.5 Hz, showing that the jitter in the along-track direction is related to that in the cross-track direction. III. ANALYSIS OF DEVIATION FROM CORRECT SPACECRAFT ATTITUDE To reduce the registration error and the internal distortion in images, the true attitude information should be estimated from the relative registration error. The correlation window size is set to 501 (cross-track) 1 (along-track) in order to estimate the relative registration error in the cross-track direction with a high time resolution. The oblong shape of the correlation window makes analysis robust to scene features. The water area is eliminated due to the low standard deviation of digital numbers in the window. The relative registration error is averaged in the cross-track direction with three sigma limits. Thus, the relative registration error is calculated using image correlation. The relative registration error g(t) at time t is expressed using the deviation from the correct attitude f(t) and the observation time lag τ between the bands as g(t) =f(t) f(t τ). (2) The relative registration error is obtained from an analysis of one scene in the range of 0 t 9 s. When the deviation from the correct attitude during the observation time lag, i.e., f(t)( τ t 0), is estimated, the satellite-attitude data are revised and are filled in the gaps using the relative registration error using the relationship f(t)= g(τ)+f(t τ)(0 t 9). Because the relative registration error is smooth, as shown in Fig. 3(a), and the attitude is interpolated in the third-order polynomials in the level-1 data processing, we can assume that the deviation from the correct attitude is also smooth. Based on this idea, we estimate the deviation from the correct attitude during the observation time lag using the evaluation function J(τ,f(t : τ t 0)), which is expressed as J (τ,f(t : τ t 0)) = {f(t) f(t t)} 2 dt. (3) Here, t denotes the line-sampling time of SWIR. The evaluation function expresses the smoothness of the deviation from the true attitude. Once f(t : τ t 0) is obtained, f(t : 0 t 9) is calculated based on (2). The smoothest solution under the constraint of (2) minimizes the evaluation function, meaning that the solution approaches the true value. We used the downhill simplex method in finding f(t : τ t 0) that minimizes the evaluation function.

4 TESHIMA AND IWASAKI: CORRECTION OF ATTITUDE FLUCTUATION OF TERRA SPACECRAFT 225 Fig. 4. Deviation from the correct attitude (a) before and (b) after eliminating the dc offset. IV. RESULTS AND DISCUSSION A. Estimation of Deviation From Correct Attitude Fig. 4(a) shows the estimated attitude fluctuation obtained to minimize the evaluation function J, provided that τ =80.9 lines. Since τ varies with changes in the spacecraft position and attitude, a representative value of τ in a scene is determined so as to minimize the relative registration error between the corrected images. The relative registration error has not only the alternating-current component but also the direct-current (dc) component, which means that the moving average of the relative registration error in the cross-track direction is not zero, but has the offset of about pixels. This offset originates from the relative error in the line-of-sight vector between the bands. Thus, we should subtract the offset from the relative registration error. Fig. 4(b) shows the estimated attitude fluctuation after eliminating the offset. Since the dc component of the relative registration error is eliminated, the monotonically increasing behavior is corrected. B. Comparison Between Bands SWIR images are corrected as follows. The relative registration error of band 5 relative to band 6 in Fig. 3(a) is calculated using image correlation. The deviation from the correct attitude in Fig. 4(b) is estimated using the relative registration error of band 5 relative to band 6. The roll component of the satellite attitude is determined using the deviation from the correct attitude in Fig. 4(b). Images are rearranged by bilinear interpolation in the cross-track direction based on the corrected attitude data. Fig. 2(d) shows the registration-error map of band 5 relative to band 6 in the cross-track direction after the correction. The primary periodic deviation observed in Fig. 2(b) disappears. Fig. 3(b) shows the averaged relative registration error at each line in the cross-track direction after the correction. The rms value and the amplitude of the relative registration error between uncorrected images are and pixels, respectively, whereas that between corrected images are and pixels, respectively. As a result, the averaged registration error between bands is reduced to less than 0.1 pixels, which is a good band-to-band registration performance and will be effective for discriminating rock and minerals, using SWIR data. The remaining jitter in the corrected image is explained as follows. The relative registration error is averaged in the correlation window, and the accuracy in the correlation measurement is limited. In addition, the peak value of the averaged relative Fig. 5. Relative registration-error map of the ASTER/SWIR band 6 of the June 7, 2002, image relative to that of the June 4, 2001, image (a) before and (b) after correction. Relative-registration-error map of the ASTER/SWIR band 6 of the June 7, 2002, image relative to the ETM+ band 7 of the March 29, 2000, image (c) before and (d) after correction (correlation window size: 7 7). registration error decreases due to the undulation of the lineof-sight vector and the parallax correction. Thus, the relative registration error is underestimated, and the jitter cannot be fully corrected. C. Comparison Between Multitemporal Data Since band-to-band registration is improved by correcting attitude data consistently, absolute distortion can also be compensated. We use the multitemporal data to validate our correction method. Since the rms value of the band-to-band registration error of band 5 relative to band 6 in the scene taken on June 4, 2001, in Tokyo is sufficiently small (0.018), the spacecraft attitude is considered to be steady during the period of scene acquisition. This means that the internal distortion in the June 4, 2001, scene is sufficiently small to be used as a reference image. Fig. 5(a) shows the registration-error map of the band-6 image taken on June 7, 2002, in Tokyo relative to that on June 4, A periodic misregistration pattern is observed in the registration-error map. Fig. 6(a) shows the averaged relative registration error at each line in Fig. 5(a) in the cross-track direction. The frequency of the pattern is almost the same as that observed in the relative registration error between the bands in Fig. 3(a). Fig. 5(b) shows the registration-error map of the corrected band-6 image on June 7, 2002, relative to that on June 4, The primary periodic deviation observed in Fig. 5(a) disappears. Fig. 6(b) shows the averaged relative registration error at each line in Fig. 5(b) in the cross-track direction. The rms value and the amplitude of the averaged relative registration error before the correction are and pixels, respectively, whereas that after correction are and pixels, respectively, showing that the internal distortion in the scene is reduced. This result shows that the method corrects the absolute distortion and is effective in the subpixel detection of landslides [12].

5 226 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 46, NO. 1, JANUARY 2008 Fig. 6. Averaged relative registration error of the ASTER/SWIR band 6 of the June 7, 2002, image relative to that of the June 4, 2001, image (a) before and (b) after correction. Averaged relative registration error of the ASTER/SWIR band 6 of the June 7, 2002, image relative to the ETM+ band 7 of the March 29, 2000, image (c) before and (d) after correction. D. Comparison Between ASTER/SWIR and Landsat-7/ETM+ Data To validate our correction method further, Landsat-7/ETM+ images are used as reference images. The orbital parameter of the Terra spacecraft is the same as that of the Landsat-7 spacecraft, except for a local equatorial crossing time. The time and angular resolutions of the Landsat-7 satellite-attitude information are markedly high because an ADS is equipped [8]. The ADS supplies attitude information every 2 ms with an angular resolution of arcsec, which means that the ADS can accommodate an attitude jitter of 250 Hz. Therefore, the Landsat-7/ETM+ images have a sufficiently high geometric performance to be a reference image [1]. We use map-oriented images of level-1g products, which are not orthorectified images. Thus, we rotated the image orientation angle and rearranged them by bilinear interpolation to enable comparison with the ASTER/SWIR images. Fig. 5(c) shows the registration-error map of the ASTER/SWIR band-6 image taken on June 7, 2002, in Tokyo relative to the ETM+ band-7 ( µm) image taken on March 29, A good correlation is not obtained for the vegetation area because the season is different. Fig. 6(c) shows the averaged relative registration error between the ASTER/SWIR and Landsat-7/ETM+ images at each line in Fig. 5(c) in the cross-track direction. The periodic misregistration pattern and the frequency are almost the same as that in the relative registration error between the bands in Fig. 3(a). Fig. 5(d) shows the registration-error map of the corrected ASTER/SWIR image relative to that of the ETM+ image. The primary periodic deviation observed in Fig. 5(c) disappears. Fig. 6(d) shows the averaged relative registration error at each line in Fig. 5(d) in the cross-track direction. The rms value and the amplitude of the averaged relative registration error before the correction are and pixels, respectively, whereas that after the correction are and pixels, respectively. Since the performance of the correlation is worse, the relative registration error after the correction deviates at several lines. However, the geometric performance of the ASTER/SWIR Fig. 7. Relative registration-error map of band 9 relative to the corrected band 6 (a) before and (b) after correction using the revised attitude information. (c) Relative registration-error map of band 9 using the line-by-line coregistration method relative to the corrected band 6. Relative registration-error map of band 2 (d) before and (e) after correction using the line-by-line coregistration method relative to the corrected band 6 (correlation window size: 7 7). image after the correction is proved to be almost equivalent to that of the ETM+ image. E. Correction of Other ASTER Bands Once the distortion of the band-6 image is corrected, improvement in the registration of the other ASTER bands to the reference bands is possible. There are two ways to establish this. One is to apply the attitude correction data to the other ASTER bands, and the other is to coregister the other ASTER bands using the image-matching technique. Fig. 7(a) and (b) shows the registration error between bands 9 and 6 before and after the correction using the revised attitude information. The first 243 lines are not corrected because the attitude estimation is obtained from t =0 to t<9 using bands 5 and 6. Fig. 7(c) shows the registration error of band 9 using the lineby-line coregistration method relative to the corrected image of band 6, where the same size of correlation window as that in the attitude error estimation is used. The rms values are 0.04 and for Fig. 7(b) and (c), respectively. The former one is ideal in principle; however, the latter one provides a better result concerning to the band-to-band registration. In both cases, it should be noted that the correction of the discrepancy in the band-to-band registration due to the attitude error decreases the image distortion. The other SWIR bands can be corrected in the same manner. Fig. 7(d) shows the registration-error map of the band-2 image relative to the corrected image of band 6. Fig. 7(e) shows the registration error of band 2 using the line-by-line coregistration method relative to band 6, where the registration error is expressed in units of SWIR pixels. The registration error is reduced from to pixel (rms). Since other VNIR bands are optically coregistered, the same replacement as that of band 2 is sufficient for correction. This means that the correction of VNIR is also possible in the proposed scheme. Since the VNIR and SWIR are composed of different telescopes, application of the correct attitude to the VNIR telescope is more complicated than that in the SWIR telescope. These results

6 TESHIMA AND IWASAKI: CORRECTION OF ATTITUDE FLUCTUATION OF TERRA SPACECRAFT 227 show that band-to-band registration errors of a pushbroom sensor with parallax bands are used to detect and correct the influence of the attitude fluctuation of a high frequency. V. S UMMARY An accurate estimation of the spacecraft position and attitude is desired in order to realize a good geometric performance of the images obtained by the Earth-observation sensor. We proposed an image correction methodology using sensors arranged in parallel on the focal plane. An observation time lag arises between bands owing to this configuration. The relative deviation between two bands during the observation time lag is calculated using the image correlation. The deviation from the correct spacecraft attitude is estimated using the band-to-band misregistration. The attitude data are revised and interpolated using the estimated deviation from the correct spacecraft attitude. Then, the image is corrected using the revised satelliteattitude information. By this method, not only the registration accuracy but also the satellite attitude is improved. As a result, we succeeded in reducing the internal distortion in the scene. This technique is applicable to an observation sensor with the same parallel configuration on the focal plane, such as EO-1/ ALI [2], FORMOSAT-2 [3], and SELENE/LISM [13]. ACKNOWLEDGMENT The authors would like to thank the Earth Remote Sensing Data Analysis Center for supplying the ASTER data. The authors would also like to thank the anonymous reviewers for the constructive comments. [7] D. Shin, J. K. Pollard, and J.-P. Muller, Accurate geometric correction of ATSR images, IEEE Trans. Geosci. Remote Sens.,vol.35,no.4,pp , Jul [8] J. L. Barker and J. C. Seiferth, Landsat Thematic Mapper band-to-band registration, in Proc. Int. Geosci. Remote Sens. Symp., Lincoln, NE, May 27 31, 1996, pp [9] K. Janschek, V. Tchernykh, and S. Dyblenko, Integrated camera motion compensation by real-time image motion tracking and image deconvolution, in Proc. IEEE/ASME Int. Conf. Adv. Intell. Mechatronics, Monterey, CA, Jul , 2005, pp [10] S. P. Neeck, T. J. Venator, and J. T. Bolek, Jitter and stability calculation for the ASTER instrument, Proc. SPIE, vol. 2317, pp , [11] P. Kudva and A. Throckmorton, Attitude determination studies for the Earth Observation System AM1 (EOS-AM-1) mission, J. Guid. Control Dyn., vol. 19, no. 6, pp , [12] J. G. Liu and G. L. K. Morgan, FFT selective and adaptive filtering for removal of systematic noise in ETM+ imageodesy images, IEEE Trans. Geosci. Remote Sens., vol. 44, no. 12, pp , Dec [13] M. Ohtake, J. Haruyama, and T. Matsunaga, Scientific goals and performance of the multi-band imager for the SELENE mission, in Proc. 23rd Int. Symp. Space Technol. Sci., Matsue, Japan, May 26 Jun. 2, 2002, p k-07. Yu Teshima received B.Sc. degree in aerospace engineering from the University of Tokyo, Tokyo, Japan, in 2001, where he specialized on space technology and remote-sensing systems, and the M.Sc. degree in aerospace engineering from the University of Tokyo in He is currently with the Toshiba Electronics Engineering Corporation, Tokyo Japan. REFERENCES [1] D. S. Lee, J. C. Storey, M. J. Choate, and R. W. Hayes, Four years of Landsat-7 on-orbit geometric calibration and performance, IEEE Trans. Geosci. Remote Sens., vol. 42, no. 12, pp , Dec [2] J. C. Storey, M. J. Choate, and D. J. Meyer, A geometric performance assessment of the EO-1 Advanced Land Imager, IEEE Trans. Geosci. Remote Sens., vol. 42, no. 3, pp , Mar [3] C.-C. Liu, Processing of FORMOSAT-2 daily revisit imagery for site surveillance, IEEE Trans. Geosci. Remote Sens., vol. 44, no. 11, pp , Nov [4] H. Fujisada, ASTER level-1 data processing algorithm, IEEE Trans. Geosci. Remote Sens., vol. 36, no. 4, pp , Jul [5] A. Iwasaki and H. Fujisada, ASTER geometric performance, IEEE Trans. Geosci. Remote Sens., vol. 43, no. 12, pp , Dec [6] A. Iwasaki and H. Fujisada, Image correlation tool for ASTER geometric validation, Proc. SPIE, vol. 4881, pp , Akira Iwasaki received the M.Sc. degree in aerospace engineering and the Ph.D. degree in engineering from the University of Tokyo, Tokyo, Japan, in 1987 and 1996, respectively. He was with the Electrotechnical Laboratory in 1987, where he engaged in the research on space technology and remote-sensing systems. He is currently an Associate Professor with the Department of Aeronautics and Astronautics, University of Tokyo.