Performance Improvement of the RFM Estimation by Modifying the Initial Population in the Genetic based Optimization

Abstract

Rational Function Models (RFMs) are known as the most famous mathematical transformations used in geometric correction of satellite images. Considering the lack of enough and well-distributed Ground Control Points (GCPs), the structure optimization is a critical step in the terrain-dependent RFM estimation strategy. Heretofore, the binary encoding Genetic Algorithm (GA) optimization method has been used to find the optimal structure of RFMs. However, randomized generation of initial population can directly impact the convergence and also computational costs. In this paper, an approach has been proposed to modify the initial population of the GA algorithm based on the correlations of the column vectors of the least square design matrix. In this approach, probability of the presence of each RFM term in the GA initial population is linearly dependent on its correlation with other terms. Although this method slightly decreases the geometric accuracy, it can fall the processing time by 37.02% on average.