机构地区: 中国矿业大学环境与测绘学院地理信息与遥感科学系
出 处: 《科技导报》 2006年第1期47-51,共5页
摘 要: 高光谱遥感信息处理自动化与智能化具有重要的理论意义和实用价值。作为有效的机器学习算法,支持向量机具有适用高维特征、小样本与不确定性问题的优越性,是一种极具潜力的高光谱遥感分类方法,但需要解决多类问题分类策略、核函数选择与优化、不确定性控制等问题。对高光谱遥感数据挖掘的若干基本问题进行了分析,在构建其框架体系与处理流程的基础上,探讨了可以发现的知识类型、典型的挖掘模式,并分析了一些主要挖掘算法和关键技术。 Hyperspectral remote sensing information intelligent processing has significant theoretical and practical values to its wide applications. Support Vector Machine, as one effective mean of machine learning, is a potential classification approach to hyperspectral RS because of its suitability for high-dimensional dataset, insufficient samples and uncertainties. But it is necessary to pay more attentions to the following issues based on the characteristics of hyperspectral RS information: multi-class classification strategy, optimization of support vector and feature space, selection and optimization of kernel function and so on. Oriented to the requirements of intelligent information processing, the framework and processing flow of hyperspectral RS Data Mining (HRSDM) is proposed, and the types of discovered knowledge and typical DM modes are discussed.