项目名称: 非结构化数据模式分析中的多核融合理论与学习方法研究
项目编号: No.61202332
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 汪洪桥
作者单位: 中国人民解放军第二炮兵工程大学
项目金额: 25万元
中文摘要: 多核学习是当前机器学习领域一个新的研究热点。本项目基于指挥信息系统中大量的非结构化数据,面向SAR遥感情报信息的处理与自动目标识别、导弹武器系统实时状态监测两类典型模式分析问题,拟研究高效的多核融合理论及多核学习方法。在理论上,以具有不同尺度效应核函数的融合、多尺度核的融合、多分辨特征提取与多尺度核方法的融合及相应核机器的设计为重点方向,从高实时性模式分类,通用多核的自适应学习,合成核机器的快速在线学习等角度展开研究。并从实际需求入手,充分考查噪声、高维异构、局部缺失和奇异值、多元与非平坦等因素下的非结构化数据,研究其结构化表示方法,从适应性、稳定性及速度等方面对算法进行理论改进和性能提升。在应用方面,依托指挥自动化实验室软、硬件环境,利用真实数据进行综合实验,验证所提算法的有效性和先进性,为部队指挥信息系统提供有效的技术支撑和可靠的决策依据,具有重要的理论意义和实用价值。
中文关键词: 多核学习;模式分析;非结构化数据;自动目标识别;在线预测
英文摘要: Multiple kernel learning is a new research focus in current machine learning field. In this project, the efficient multiple kernel fusion theories and learning methods will be studied facing two typical pattern analysis problems based on a large amount of unstructured data in the command information system. The two problems are the processing and automatic target recognition of SAR remote sensing information and the real-time state monitoring of the missile weapon system. In theory, the fusion of kernels with different scale effects, the fusion of multi-scale kernels, the fusion between multi-resolution feature extraction and multi-scale kernel method, and design of the corresponding kernel machines are regard as the key directions. The basic theoretical research mainly expands from three perspectives, which are the pattern classification with high real-time requirement, the universal algorithms on multiple kernels' adaptive learning and fast online learning of composite kernel machines. To meet the requirements of practical applications, and fully considering the influence factors, such as noise, high dimension and heterogeneity, local deletion and the singular value, multivariate and non-flat, the structured representation method is studied for unstructured data, on this basis, the algorithms are promoted from
英文关键词: Multiple kernel learning;Pattern analysis;Unstructured data;Automatic target recognition;Online prediction