Modelling of geometric correction and relative radiometric normalization for near equatorial earth observation satellite images

The near-equatorial earth observation satellite is a new-generation optical satellite that is expected to cater to the needs of many countries located in or near the equator. It does not move in a fixed row and path during capturing time. Therefore, in each visit,the images exhibit differences depending on the viewing angle, time, illumination, sun zenith and azimuth angles, sensor zenith and azimuth angles, altitude, and attitude (roll, yaw, and pitch) of the satellite. These factors result in near-equatorial imagery having:(a) mis-registration bands, (b) no geometric matching between features located in the sequences of two images even though they are captured from the same strip, and (c) no reflectance (radiometric) matching between features located in two images captured in the same area at different times. Conventional modeling does not able to process the near equatorial satellite images. The research aims is study the near equatorial satellite images and develop models that can overcome the difficulties that facing processing of near equatorial satellite imagery as follow: (1) To develop band to band registration model for near equatorial image, to overcome the highly nonlinear band shifting of the near equatorial image bands. (2) To develop geometric correction model for near equatorial images, to reduce the highly geometric correction between the near equatorial imagery. (3) To design and implement remote sensing goniometer to simulate NEqO system images, to simulate the near equatorial satellite image. (4) To develop relative radiometric normalization model for NEqO images, to normalize the near equatorial images that have been captured at different time. A new technique to overcome band-to-band registration through the automatic generation of control points from satellite images via scale-invariant feature transform (SIFT) is proposed in this study. The SIFT generated control points are utilized to perform registration with first- and second-order polynomials and spline transformations to correct the misregistration between near-equatorial orbit (NEqO) image bands. The image employed in this study is provided by Malaysian National Space Agency (ANGKASA) it was captured by the RazakSAT satellite. An accuracy assessment is performed by comparing the result of the proposed method with the result of both automatic and manual transformations using polynomial transformation. The root mean square errors of the first- and second-order polynomial transformations are 4 and 3 m, respectively. Moreover, the spline transformation produces RMSE 1.1 × 10-6 m. A new technique is developed to improve the automatic extraction of control points. The technique is then utilized to perform geometric correction of near-equatorial images. The method, which is called refine and, improves scale-invariant feature transform (RI-SIFT). RazakSAT and SPOT-5 images are used. The proposed approachbegins by selecting the reference and sensed images. Then, grayscale and image compression are performed. Automatic control point extraction is then performed to generate control points automatically. The generated control points are refined by using the sum of absolute difference algorithm (SAD), with the help of an empirical threshold and control point locations to avoid obtaining inaccurate control points. The refined control points are applied in spline transformation to overcome the geometric errors of near-equatorial orbit (NEqO) images. Validation is then performed by comparing the result of the proposed approach with that of direct geometric correction through the use of polynomial transformation. Results show that the accuracy values obtained from using the refine transformation (RI-SIFT with spline) and polynomial transformation are 7.08 × 10-9 m and 104 m, respectively. The proposed model exhibits accuracy and precision. A relative radiometric normalization model is implemented in three stages. First, a goniometer is designed and built to simulate near-equatorial images and perform relative radiometric normalization. Second, the goniometer is used to collect images with different illumination to conduct radiometric normalization. In the third stage, the refine and improve scale-invariant feature transform method is developed to extract automatically radiometric control points over the image bands. The method aims to acquire the intensity values of the control points to use as pseudo invariant features (PIFs) between the reference and sensed image bands. The next step is to perform statistical linear regression on the control points of the reference and sensed image bands to generate regression transformation functions for use in normalizing the sensed image bands to reference image bands. The proposed model is validated by determining the correlation between the normalized and reference image bands. The correlation range is 0.69–0.85 over the bands of the slave image.

Text FK 2015 86.pdf - Submitted Version Download (428kB)

Actions (login required)

View Item