MATHEMATICAL MODELS OF NATURE INSPIRED COMPUTATIONAL INTELLIGENCE FOR TERRAIN UNDERSTANDING

Abstract

Nature inspired intelligence has emerged as a crucial means of implementing machine intelligence with human-like reasoning capabilities and an efficient mechanism for handling diverse uncertainty characteristics. These techniques can form the basis of building optimization algorithms which can adapt itself to suit the purpose of natural terrain understanding, and prove to be better by giving more accurate results than the other existing optimization techniques. The purpose of this research work is to develop new optimization models inspired from emerging geo-sciences techniques, extend the original optimization models of recent nature inspired techniques or build hybrid models of different nature inspired techniques by analyzing their performance governing factors, and hence present an adaptive framework for the terrain understanding problem. The research focuses on two major categorizations in the terrain understanding application: geo-spatial feature extraction, path planning and location prediction on remote sensing inputs. The research work starts by establishing the concept of information sharing in swarm intelligence techniques which is the key factor governing the heuristic function definition and leading towards optimal solutions. It establishes entropy and similarity index as the performance governing factors influencing the classification efficiency of biogeography based classifier. Based on this analysis, two new hybrid classifiers are developed, the first is an integration of BBO with ACO2/PSO and second, is an integration of BBO- GS with ACO2/PSO. These classifiers achieve very high kappa coefficients and prove to be great advancements over the existing classifiers. The research work moves ahead and focuses on the design of a new optimization algorithm – BBO/EE and establishment of the factors of extinction and evolution for the development of extended model of species abundance in biogeography. The above factors contribute towards the sophistication of the existing model of biogeography and thereby the concept of different HSI functions for different habitats according to the habitat characteristics is proposed. The developed extended model is then adapted for the design of a military application of enemy base station prediction and best feasible path finding targeted for two natural terrain scenarios- serving a dual purpose. The above application is a milestone in the history of automated recommender system design for battlefield planning. The research also implements feature extraction using BBO/EE with dynamic HSI function. This application when integrated with the existing hybrid bio-inspired classifier is able to achieve the highest kappa coefficient amongst all the classifiers developed till date. VI The research work finally aims to design a new algorithm based on plate tectonics called as the plate tectonics based optimization. This algorithm introduces the concept of plate mobility index – weights with decision variables having tolerance and the concept of dynamic adjustment of weights. The developed algorithm will utilize an optimal set of decision variables until all the problem constraints are satisfied. The research work concludes finally by extending the taxonomy of Nature Inspired Computational Intelligence to include the techniques inspired from geo-sciences successfully. The contents of this thesis are based on research papers accepted in refereed and mainstream international journals and conferences. The thesis is concluded with pointers for prospective future work. Additionally referred briefs are some of the prospective problems in the field that are unresolved as of now.