Based On Decision Tree Incremental Learning Imaging Target Classification Technology Research

It makes more and more stringent demands on ATR in the new military and civilian applications.Classical decision tree classifier is combined with the recent rise of the rough set theory,the incremental learning and the data mining,which is expected to improve analysis efficiency of magnanimous information and adaptability of complex applying conditions in order to raise the speed and accuracy of the recognition and classification.To solve the problems of the traditional decision tree model in imaging target and recognition,the paper focuses around building the low-complexity decision tree and extending the incremental learning model in processing tasks of magnanimous information.The main contributions of this paper are listed as follows:This thesis comprehensive studies and analyses the relative literatures including the rough set,the decision tree and the incremental learning.It details the rough set and dividing methods of the variable precision rough set,and the description mechanism of attributing measure.Comparing features and performance of several typical incremental decision trees,the basic flow of the decision tree algorithm is given in the paper.Besides,features and application abilities are introduced about big data analytics and the design of the classifying algorithm of the incremental learning.According to the decision tree model with low classification accuracy rate and bad robustness,one of key techniques for making the decision tree algorithm design clear is the selection problem of the split attributes' standards.The paper presents a strategy of attribute measure based on the variable precision rough set,which can adapt the situations of the incomplete data or noise interference and improve accuracy and robustness of the decision tree classification.The strategy can control the scale of the decision tree and reduce the complexity of model structures according to training the sample reaching the node.The paper puts an improved algorithm of random forests to prevent classifying errors based on uneven of training data which combines between variable precision rough sets decision tree and information theory methods to improve the accuracy of classification and adaptability.It exists continuous wholesale obtained large sample information of imaging target in the big data context.Sample sort and dynamics need to the adaptive ability of the classifier,Paper presents a method of some examples of the incremental learning method which can complete the incremental learning without changing the decision tree model.This method greatly reduces the consumption of time and space conforming to requirements of artificial intelligence.Incremental decision tree algorithm gives typical examples and warns for abnormal data,which providing analysis for human interaction and assistant decision.The paper adopts typical simulation examples,internet UCI universal databases(sizes are 556,601,554,132)and letters image data sets(size is 20000),and tests verify the validity and feasibility of the proposed algorithm.