Detection and Estimation of Satellite Attitude Jitter Using Remote Sensing Imagery

Most satellite remote-sensing images are obtained using pushbroom or whiskbroom scans by the coupling of line scans in one dimension and satellite movements in another dimension. Since satellite imagery should be consistent with maps when used as geospatial data, good geometric performance is a challenge for satellite data processing (Lee et al., 2004; Storey et al., 2004). Accurate attitude estimation is essential for this purpose because satellites suffer from mechanical vibration, originating from solar paddles, wheels and high-gain antennas, during line scans. In most satellites, attitude is estimated by interpolating measurement data obtained from star sensors and gyros, which are sampled at low frequency. When the sampling rate of attitude information is low compared with the frequency of attitude, images processed by geometric correction suffer from distortion. Therefore, additional information is necessary to estimate attitude accurately. One solution is to estimate the correct attitude of spacecraft using ground control points (GCPs), which requires the preparation of GCPs in the target scene and depends on their accuracy (Shin et al., 1997). Landsat adopts angular displacement sensor (ADS) to detect and compensate high-frequency jitter (Barker & Seiferth, 1996). Another possible solution is to add an image sensor on the focal plane, which tracks the displacement of ground targets (Janschek et al., 2005). A sensor system with parallax observation enables the detection of the effect of pointing fluctuation, which can be rapidly measured by attitude sensors, such as star sensors and gyros, without additional high-performance sensors. A software methodology based on image processing and optimization techniques to accurately estimate the attitude information of spacecraft under operation is important for obtaining geometrically accurate images (Iwasaki & Fujisada, 2003; Bayard, 2004; Liu & Morgan, 2006; Teshima & Iwasaki, 2008). When optical sensors on satellites suffer from pointing fluctuation, images obtained at two different times cannot be co-registered by a parallel shift. Figure 1 shows the displacement between two remote-sensing images before and after the large earthquake in Pakistan on October 8, 2005, with a magnitude of 7.6 clarified by normalized cross-correlation and the difference in digital elevation models (DEMs). Wavy patterns are observed in Figs. 1(b), (c) and (d), which correspond to horizontal displacements in the crossand along-track directions and vertical displacement, respectively. Owing to this artifact, the fault running from upper left to lower right is difficult to trace. Assuming that the displacement occurred only around the fault zone, the wavy patterns are destriped in Figs. 1 (e), (f) and (g), which correspond to horizontal displacements in the cross and along track directions and vertical displacement, respectively. The displacement