项目名称: 普适计算对象感知多模态不精确性数据融合算法研究
项目编号: No.61462042
项目类型: 地区科学基金项目
立项/批准年度: 2015
项目学科: 其他
项目作者: 叶继华
作者单位: 江西师范大学
项目金额: 42万元
中文摘要: 普适计算是信息空间与物理空间的融合,普适计算使用多个或多类传感器获取感知对象数据,各传感器数据通常具有不同的特征,感知对象中大量的数据是以多模态的形式呈现的,同时,系统所处的环境也受到各种随机因素的影响而不断地变化。对于多模态数据带来的复杂信息处理难题,传统的解决方案主要面临的问题有:单一模态不同层次的特征表示差异很大;扩展到不同模态上时原有的计算和学习的理论与方法如何适应和更新。同时上述方案为了计算和分析上的方便对问题做非常严格的约束,这样的假设并不能有效地满足处理复杂非结构化数据的需求。本课题以多图像信号的有效融合进行对象感知为研究对象,使用多尺度分析进行特征提取;以不确定度理论直接计算训练样本和测试样本的权重,避免了对数据分布的假设;通过研究多子空间直和计算在多模态数据处理上的应用,避免了对数据同构的假设,研究并实现一种解决普适计算环境下多模态数据融合的新方法。
中文关键词: 普适计算;对象感知;多模态数据;多子空间;不确定度
英文摘要: Pervasive computing integrated cyberspace into physical space, pervasive computing using multiple sensors to obtains data from perception object, each sensor data often have different characteristics ,perception object have mass of multi-modal data,the environment in which the system is also affected by a variety of random factors and constantly changing.For Complex multi-modal data processing problems, traditional solutions existing main problem: the single modal characteristics of the different levels expressed differences; extended to different modes, It's difficult for the theories and methods of the original calculation and learning to adapt and update. Above scenario exist very strict constraints in order to calculate and analysis. This assumption can not effectively adapt to deal with the complexity of unstructured data. The topics research the effective integration of multi-image signals for object perception , using multi-scale analysis for feature extraction ;used uncertainty to direct calculation the weight of the training sample and test sample, avoid the assumption of data distribution; Applied multi-subspace direct-sum to multi-modal data processing to avoid data isomorphic assumptions, research and implemention a solution for multi-modal data fusion.
英文关键词: Pervasive Computing;Object Perception;Multi-modal Data;Multi-Subspace;Uncertainty