| Automated and semi-automated techniques have been researched as an alternative way to reduce human interaction and thus improve the information extraction process from imagery. This research developed an innovative methodology by integrating machine learning algorithms with image processing and remote sensing procedures to form the evolutionary framework. In this biologically-inspired methodology, non-linear solutions are developed by iteratively updating a set of candidate solutions through operations such as: reproduction, competition, and selection. Uncertainty analysis is conducted to quantitatively assess the system's variability due to the random generation of the initial set of candidate solutions, from which the algorithm begins. A new convergence approach is proposed and results indicate that it not only reduces the overall variability of the system but also the number of iterations needed to obtain the optimal solution. Additionally, the evolutionary framework is evaluated in solving different remote sensing problems, such as: non-linear inverse modeling, integration of image texture with spectral information, and multitemporal feature extraction. The investigations in this research revealed that the use of evolutionary computation to solve remote sensing problems is feasible. Results also indicate that, the evolutionary framework reduces the overall dimensionality of the data by removing redundant information while generating robust solutions regardless of the variations in the statistics and the distribution of the data. Thus, signifying that the proposed framework is capable of mathematically incorporating the non-linear relationship between features into the final solution. |
