الفهرس 
EMPIRICAL SENSOR MODELS FOR HIGH RESOLUTION SATELLITE IMAGESOnly 14 pages are availabe for public view
 |  | from | 322 |  |  |
| | | from | 322 | | |
Abstract
High-resolution satellite imagery with a ground pixel size of better than 1 meter offers the potential to extract useful and accurate spatial information for mapping applications. The geometric correction of these images is a first and essential step before using them in mapping applications. Thus to remove or correct the different types of geometric distortions the images contain. Physical sensor models with the aid of satellite ephemeris information can present the relationship between the satellite image space and the ground space. Most commercial high-resolution satellites, which have been recently launched, do not provide satellite ephemeris information to construct the physical sensor models. Therefore, alternatively empirical sensor models being independent of satellite ephemeris should be used. The rational function model (RFM) is mathematically a generic form of empirical sensor models, which relates the two-dimensional (2D) image space to three-dimensional (3D) ground space or vice versa. In this study, a new software, called EMAN has been developed based on MATLAB programming language to allow the implementation of 36 empirical models. These 36 empirical models are categorized by model direction (forward models and inverse models), model type (rational function models, rational function models with equal denominators and polynomial models), model order (third order model, second order model and first order model), and model dimension (three dimensions and two dimensions). In addition, the effect of model type, model order and model dimension have been investigated using the results obtained from EMAN software. This investigation showed that iv generally, with respect to model type, RFM provide more accurate results than RFM with equal denominators and polynomial models. The geometric correction of satellite images in EMAN software passes through four steps, each step has its tools to implement this step and make it easier. These steps start with collecting control points that are necessary to compute and evaluate the used model, followed by selecting and solving the model by estimating the model coefficients using previously selected ground control points, then evaluating the selected model using the check points, and finally generating the rectified image. In the solving model step, because of the terrain dependent solution, RFM can be deteriorated by the over parameterization among its polynomial coefficients and consequently a numerical instability RFM solution can be resulted. In this study, thirteen methods have been tested for obtaining stable solution of each empirical sensor model and the most accurate method for each empirical sensor model has been used in EMAN software. The models solution methods used in EMAN software are: • Direct solution method for solving different orders of polynomial models, first order rational function model with equal denominators and first order rational function model. • The combination of unweighted L-curve and ridge trace solution for solving 2D second order rational function model with equal denominators and 2D second order rational function model. • The combination of weighted L-curve, ridge trace and iteration by correcting characteristic value (ICCV) solution for solving 2D third order rational function model with equal denominators and 2D third order rational function model. v • The combination of weighted generalized cross validation (GCV) + ridge trace + ICCV solution for solving 3D second and third orders rational function models with equal denominators and 3D second and third orders rational function models. EMAN software has been evaluated to check its validity to construct empirical sensor models for rectifying images by comparing the results obtained from it with those resulted using ERDAS Imaging and PCI Geomatica software. Two different data sets (Fredericton and Tanta) from two different satellites (IKONOS and GeoEye) were used for this evaluation process. The results obtained from EMAN software package proved its efficiency to rectify high-resolution images using all of the 36 empirical sensor models. By contrast, PCI software does not support group models of the RFM with equal denominators and ERDAS software does not support any of the inverse empirical models. 3D polynomial models are not supported by either PCI software or ERDAS software. Hence, EMAN software can be considered an effective standalone software, that can be used by several user to perform the geometric corrections by different empirical sensor models.