| 摘要: |
| 【目的】以东北建三江黑土区为示范区,开展高光谱遥感反演与地球化学验证,突破高光谱遥感在黑土地资源快速准确评价中的关键技术,开展黑土养分和农作物生长状态的航空高光谱反演研究,建立空地一体化的黑土质量快速评价技术体系,为黑土资源管理提供依据。【方法】引入航空高光谱成像系统CASI-1500,获取380 ~1 050 nm数据。均匀采样60个样品,化验获得其有机质、全氮、全磷和全钾含量数据,将机器学习解决问题的策略引入黑土养分信息的智能提取中,将学习过程与黑土养分光谱特征程紧密相连,通过搭建特征向量、逻辑语句、规则、语义网络和框架等方法,实现黑土养分信息的智能提取。【结果】按照智能降维、智能聚类、智能分类、智能回归的步骤,在理解养分信息知识的基础上,分析比较,做出假设,检验并修改假设,实现对现有知识的扩展和改进。【结论】机器学习方法的引入,能够从特征抽取、推荐、关联规则、优化等角度进一步提升光谱数据利用效率,从更深的角度挖掘高光谱数据隐而未现的土壤信息,随着机器学习算法的进步和光谱数据的积累,未来这一方法将成为地学大数据的重要组成部分,带来更深更广的应用效果。 |
| 关键词: 机器学习 高光谱遥感 黑土地 智能化 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学“基于深度学习的机载高光谱矿物非线性解混和丰度反演研究”(41602333)；“十三五”装备预先研究专项技术项目“多传感器探测项目”(32101080302) |
|
| Machine learning method for extracting black soil nutrient information from hyperspectral data |
|
Zhang Donghui1, Guo Wei2, Jiao Zhenhua3
|
|
1.Beijing Research Institute of Uranium Geology;2.Intelligence.Ally of Technology Co., Ltd.;3.CNOOC Research Institute Ltd.
|
| Abstract: |
| [Purpose] Taking Jiansanjiang black soil area in Northeast China as a study area, hyperspectral remote sensing extraction and geochemical verification are carried out to break through the key technology of hyperspectral remote sensing in the rapid and accurate evaluation of black soil resources., Method of the inversion of black soil nutrients and crop growth state by aerial hyperspectral is studied in order to provide the basis for the management of black soil resources. [Method] The aviation hyperspectral imaging system casi-1500 is introduced to obtain 380 ~ 1050 nm data. 60 samples are collected evenly, and the data of organic matter, total nitrogen, total phosphorus and total potassium are obtained. The solving problems strategy by machine learning is introduced into the intelligent extraction of black soil nutrient information. The learning process is closely connected with the spectral feature process of black soil nutrient. The intelligent extraction of black soil nutrient information is realized by building feature vector, logical statement, rule, semantic network and framework. [Result] According to the steps of intelligent dimensionality reduction, intelligent clustering, intelligent classification and intelligent regression, based on the understanding of nutrient information knowledge, analyze and compare, make assumptions, test and modify assumptions, and realize the expansion and improvement of existing knowledge. [Conclusion] The introduction of machine learning method can further improve the efficiency of spectral data utilization from the perspective of feature extraction, recommendation, association rules, optimization, etc., and mining the hidden but not existing soil information of hyperspectral data from a deeper perspective. With the progress of machine learning algorithm and the accumulation of spectral data, this method will become an important part of geoscience big data in the future, bringing deeper and wider application effect. |
| Key words: machine learning hyperspectral remote sensing black land intelligence |