基于案例推理的土壤光谱识别和分类研究

项目名称： 基于案例推理的土壤光谱识别和分类研究

项目编号： No.41471175

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 地质学

项目作者： 解宪丽

作者单位： 中国科学院南京土壤研究所

项目金额： 80万元

中文摘要： 土壤类型的自动识别具有重要的学术价值和应用意义。土壤的反射光谱特性是土壤理化性状信息的综合反映，在土壤类型的快速识别上有很大的应用潜力。我国土系调查项目积累了丰富的土系样本和土壤系统分类知识及经验。本项目拟利用我国土系调查样本，建设覆盖我国最多土壤类型的光谱数据库；分析不同层级土壤分类类型的发生发育-物质组成-光谱反射率特性之间的关系，获得土壤类型的代表性光谱识别案例，建立不同层级土壤类型的光谱识别案例库；研究基于光谱匹配的土壤类型识别和基于光谱特征指标的土壤类型识别方法，运用案例推理，实现基于光谱的土壤自动识别和分类；并选择宁镇丘陵区和赣东北低丘岗地区作为试验区，验证基于案例推理的土壤光谱识别和分类的有效性。本项目将案例推理运用到土壤类型的光谱识别中，使土壤分类知识和经验隐含在土壤光谱识别案例中，从而在缺少土壤分类专家参与的情况下, 实现快速、较为客观的土壤分类结果。

中文关键词： 数字土壤制图；土壤计量学；机器学习；土壤光谱库；土壤分类

英文摘要： The recognition of soil types automatically has important academic value and application significance in soil science, land use planning, agricultural industry and environment assessment. The spectral reflectance characteristic of soil is the comprehensive representation of soil properties, and has great potential in the soil type recognition quickly. The soil series survey in China has implemented for several years, and detailed knowledge and experience of soil identification and classification according to the Chinese Soil Taxonomic Classification has obtained in this survey. The purpose of this project is to identify soil types using soil spectral reflectance based on case-based reasoning (CBR). The first objective of this project is the development of soil spectral database which will cover most soil types in China by using the archived soil samples from soil series survey. The second is based on the relation between soil development, composition and spectral reflectance characteristics of soil types at different classification levels, to find representative cases of the spectral identification of soil types, and build the case library of the spectral identification of soil types. The third is to find the identification methods of soil type using similarity computing based on spectral matching and spectral index, and then identify soil types using cased-based reasoning with the case library. Finally, the method is applied on the soils sampled from two test area, Nan jing-Zhen jiang Hilly regions in Jiangsu province and Low hilly regions in Jiangxi province , in order to assess if can acceptably be used. This project develop a new method for soil type identification, based on representative cases of the spectral identification of soil types which implicit soil classification knowledge and experience, to identify soil types quickly and correctly without the participation of soil classification experts.

英文关键词： Digital soil mapping;Petrometrics;Machine learning;Soil spectral library;Soil classification