Framework Of Artificial Immune System And Its Application In MODIS Data Classification

Artificial Immune Systems (AIS) are defined as computational systems inspired by theoretical immunology and observed immune metaphors, applied to solve engineering problems. Its development follows those of soft computing paradigms such as artificial neural networks and evolutionary algorithms. As the core of AIS, current immune algorithms are negative selection algorithm, clone selection algorithm, artificial immune network. However, these immune algorithms have no entirety characteristics because they were only inspired by the works on theoretical immunology or some special processes that occur within the immune systems. Some immune mechanisms have not been used. And in present researches, any system designed by inspiring from immune system is called AIS, but no AIS have unified architecture. So the main researching target of this thesis is to develop a unified architecture of the AIS through integrating and enhancing the current immune algorithms. To check up the efficiency of the immune algorithm developed in the thesis and enlarge application fields of AIS, the author apply AIS to regional land cover image classification using MODIS data. The thesis introduces a new way to analyse MODIS data.In order to research AIS definitely, some important immune metaphors are extracted after studying the nature immune system as a whole. These metaphors include immune pattern recognition, self/nonself recognition, the clonal selection theory, the immune network theory and the shape space formalism.A Framework of Immune Algorithm(FIA)is proposed in the thesis. FIA is an abstract of the four stages of immunological events in the rpocedure of immune response. In initialization phase, we list a representation of the components of the system. In other words, we should define antibody, antigen and other immune cells or molecules in the artificial system. In recognition phase, we make an affinity measure, which quantifies the interactions between components of the system. The outcome of evolutionary phase is to generate high-quality memory B cells with specificity to the exposed antigens for future use. The last phase is the phase of problem solving. FIA make the most of the characters dealing with information which is occurred in immune system. Furthermore, the four phases can be implemented in different ways according the different case of FIA's applications..The architecture of the AIS given in this thesis has three layers which related to the three main contents. The architectrue displays the general structure of AIS. The first layer of AIS is a representation for the components of the system. A representation creates abstract models of immune organs, cells and molecules. The second layer of AIS is a set of mechanisms to evaluate the interaction of individuals with the environment and each other. A set of functions, termed affinity functions, quantify the interactions of these"artificial elements". The last layer of AIS is a set of general purpose algorithms to govern the dynamics of the AIS. Acting as the core of AIS, the immune algorithms tell us how to use AIS for solving engineering problems.The thesis presents a data classification algorithm based on AIS. To implement the T cell supervision mechanism we employed a chi-squared analysis for estimating the local feature relevance, whereby better classification performance can be achieved. Using both simulated and real-world data, the performance of the algorithm proposed is discussed, and the validity of this algorithm solving data classification problem is showed. The proposed AIS is applied to remote sensing image classification. Based on the advantage of MODIS multi-spectrum data, this research explored a classification method of feature selection and extraction, which combines the multi-spectrum data with multi-temporary data in order to improve the classification accuracy. The results indicate that it has higher classification accuracy using EVI (Enhanced Vegetation Index) and NDWI (Normalized Difference Water Index) than the only spectrum data. But NDSI (Normalized Difference Soil Index) almost has no effect for improving the classification accuracy.