Research On Feature Extraction Methods Of Time Series Of Gray Level Of Target Based On XCS

Long range?fast speed and mass destruction are important characteristics of ballistic missile. In modern military conflicts, ballistic missile is a key weapon taking on tasks of strategic deterrence and tactical application. In response to the threat, many countries are actively developing the technology of ballistic missile defense. Aiming at the target recognition of space detection system, this thesis studies the learning?classification and feature extraction of time series of gray level of ballistic targets. The main contribution of this thesis are as follows:1. Based on the Indian “Agni-3” ballistic missiles, this thesis constructs the obtainment model of time series data of gray level of ballistic targets. The main types?geometric shapes?sizes and motion forms of ballistic targets are designed, as well as the detection scene on a fast moving platform. This thesis calculates the temperature distribution of targets, completes the modeling and simulation of IR radiation characteristics of ballistic targets, analyses the differences of IR radiation characteristics among different targets, limits the range of parameters related to targets, obtains the time series data of gray level of ballistic targets by computer modeling.2. This thesis proposes an improved method of learning system by considering the problems of slower convergence and easy trapped into local optimal solution of XCS in classification problems. This thesis introduces each module in learning system in detail, systematically discusses the working process and data exchange method in XCS(e Xtended learning Classifier System), and improves the system from classifier attribute?discrimination parameter and “bad condition” set, leads the creation of new classifiers, Several classification experiments demonstrate the superior performance of the improved XCS.3. This thesis designs an expression algorithm of series based on gray mode by aiming at the problems of large dimension and difficulty in learning for time series data of gray level of ballistic targets. Direct learning and classification of series data may result in curse of dimensionality, meanwhile, the learning process will become very complicated combined with parasitic noise. In order to express the time series, this thesis constructs the parameter fitting model for series, solves the fitting parameters based on gray system, which can simulate the change features and restrain the noise in time series of gray level. This design express the series data with parameter data and reduce the dimension by a large margin.4. Considering the classification problem between warhead and non-warhead in midcourse, this thesis accomplishes the learning, classification and feature extraction of time series data of gray level of ballistic targets, conducts the learning and classification experiments of series parameter data by applying the improved XCS. The learning performance demonstrates the proposed parameter fitting, the classification results demonstrate the validity of the population optimization. In order to mine the hidden information in the parameter data, this thesis analyzes the population which obtained from learning process, selects the parameter related to categories as the feature of series, decreases the number of features by combined with fitting function, achieves a certain extent of classification performance.