项目名称: 基于多特征融合和集成学习的多目标识别技术研究
项目编号: No.61273275
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 王晓丹
作者单位: 中国人民解放军空军工程大学
项目金额: 80万元
中文摘要: 本项目以弹道中段目标识别为研究背景,将在前期研究基础上,深入研究基于多特征融合和集成学习的多目标识别技术,以减少识别结果的不确定性。通过突破多特征多层次融合方法、基于数据感知的ECOC编码和解码方法、集成学习理论指导下的多类分类方法、满足动态融合需要的适合连续多目标识别特点的有效决策融合方法等关键问题,进一步发展雷达目标识别新理论。项目将从特征融合、分类器融合、决策融合相结合出发,研究有效的多特征多层次融合方法,构建有利于目标识别的特征向量;研究基于数据感知的编码和解码方法,在保证分类精度的同时最大限度减少编码长度;研究基于ECOC框架的多类分类方法,尝试利用集成学习理论提升多类分类效果;研究有效的决策融合方法,并通过连续融合识别提高多目标识别准确率。本项目是前期项目研究工作的深化和拓展,其预期研究成果将为促进中段目标识别技术的研究和系统的实现提供新思路,具有重要的理论意义和实际应用价值。
中文关键词: 集成学习;目标识别;纠错输出编码;多类分类;证据理论
英文摘要: Under the background of radar target recognition in ballistic midcourse, and to reduce the uncertainty of recognition result, the project will make in-depth research on multi-target recognition based on multi-feature fusion and ensemble learning. By breaking through the key problems in radar target recognition, such as, effective ways of multi-feature hierarchical fusion, encoding and decoding strategies of data-perception ECOC, effective multi-class classification algorithm based on ensemble learning, and the effective decision fusion method meeting the requirements of both dynamic fusion and successive multi-target recognition, the project will try to give some new points in the development of the theory of radar target recognition. The project will investigate the following four aspects from the combination of feature fusion, classifiers fusion and decision fusion. The first one is to investigate the effective ways of multi-feature hierarchical fusion and form the feature vector that is propitious to get higher performance of target recognition. The second one is to investigate the encoding and decoding strategies of data-perception ECOC that can minimize the coding length on the premise of higher classification performance. The third one is to investigate the multi-class classification algorithm based on ECO
英文关键词: Target recognition;Ensemble learning;Error correcting output codes;Multi-class classification;Evidence theory