Automatic Determination of the Optimum Generic Sensor Model Based on Genetic Algorithm Concepts

Generic sensor models (GSMs) are comprehensive mathematical models by which different geometric structures of satellite images could be modeled in order to establish the connection between image and object spaces. Nevertheless, as they are mathematical models, rather than physical models, it is difficult to determine which term and order of GSMs can provide the best result. Therefore, conventional solutions need an expert operator to try different terms and orders for the best solution of GSMs or to find the best trade-off, which is a complex and time consuming process. Moreover, conventional solutions for automatic determination of the optimum GSM parameters are not practically efficient and instead of going towards the global optimum, frequently get trapped in some local optima. In this paper we propose a novel methodology which automatically determines the optimum GSM’s terms and orders based on genetic algorithm concepts. Extensive evaluations carried out on a wide range of different optical satellite images demonstrate the high potentials of the proposed strategy. Introduction In recent years, there have been some major efforts in designing and launching high-resolution satellite sensors for mapping applications (e.g., Ikonos-2, EROS-A1, QuickBird 2 and SPOT 5). With the successful launch and deployment of these satellites, the era of commercial high-resolution earth observation satellites for digital mapping has been initiated. In addition, several other high resolution commercial and governments imaging satellites in different countries, such as the United States (OrbView-3), India (IRS-P5), and Russia (RESURS-DK) are expected to be launched in a near future. Therefore, the number of high-resolution satellite sensors for mapping applications is growing rapidly. With regard to the geometric characteristics of satellite imagery, the existing and announced forthcoming highresolution imaging systems can be classified into three main groups: frame type, whiskbroom, and pushbroom imaging systems (Figure 1). In Frame-type images, the image is formed over the whole area of the imaging frame simultaneously, using a projection lens to produce a perspective view of the object from a single exposure station (e.g., KFA-1000, TK-350). Whiskbroom images are acquired by mechanicalAutomatic Determination of the Optimum Generic Sensor Model Based on Genetic Algorithm Concepts Farhad Samadzadegan, Ali Azizi, and Ahmad Abootalebi optical sensors which produce a single continuous strip image by means of mirror oscillation or rotation, and forward motion of the platform. Pushbroom images are acquired by a consecutive digital collection of individual scan lines at a frequency corresponding to the scanning velocity. There are three types of pushbroom imaging systems: cross-track imaging devices acquire stereo images of the ground from two adjacent orbits using their off-nadir viewing capabilities (e.g., SPOT 1–4 and IRS 1C/1D), along-track stereo scanners record stereo images of the ground from a single orbital pass using backward and forward-looking arrays (e.g., MOMS-02, ASTER, SPOT 5) or mono images using nadir-looking arrays, and flexible pointing systems equipped with pointing devices providing the capability to acquire both cross-track, and along-track stereo images from single or multiple orbital passes (e.g. Ikonos). Successful exploitation of the high accuracy potential of these imageries depends on the ability of the mathematical formulations for the sensor modeling. In this direction, the requirement for the development of an efficient 2D or 3D comprehensive sensor model formulation for various satellite images is a real challenge. In the sections that follow, after a brief review of the main features of the existing models, our approach for automatic determination of a comprehensive geometric model of the satellite images will be presented. Rigorous Versus Generic Sensors Models Mathematical modeling approaches for orientation and restitution of different optical satellite images have been investigated by different research groups (Konecny et al., 1987; Deren and Jiayu, 1988; Gugan and Dowman, 1988; Kratky, 1988; El-Manadili and Novak, 1996; Tao and Hu, 2001 and 2002). Mathematical formulations presented in these research works, may be divided into two main groups of rigorous sensor models (RSMs) and generic sensor models (GSMs). RSMs reconstruct the spatial relations between remotely sensed imagery and the ground scene based on conventional collinearity equations. The method is highly suited to frame type sensors. Non-linear effects caused by lens distortion, film shrinkage, or atmospheric effects are dealt with either by additional parameters or by a priori refinement process. The RSM models have proved to be quite appropriate for the PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING March 2005 277 Farhad Samadzadegan and Ali Azizi are with the Department of Surveying and Geomatics Engineering, Faculty of Engineering, University of Tehran, Tehran, Iran ([email protected], [email protected]). Ahmad Abootalebi is with the Research Institute of the National Cartographic Center (NCC) of Iran. Photogrammetric Engineering & Remote Sensing Vol. 71, No. 3, March 2005, pp. 277–288. 0099-1112/05/7103–0277/$3.00/0 © 2005 American Society for Photogrammetry and Remote Sensing 02-150.qxd 1/14/04 3:41 PM Page 277