基于Sentinel-2遥感影像的自然资源分类提取研究  

Research on natural resource classification and extraction based on Sentinel-2 remote sensing images

作　　者：刘知 刘小松 杨波 王鑫 张航 LIU Zhi;LIU Xiaosong;YANG Bo;WANG Xin;ZHANG Hang(Coal Geological Geophysical Exploration Surveying&Mapping Institute of Shanxi Province,Jinzhong 030600,Shanxi,China;Key Laboratory of Investigation,Monitoring and Protection of Natural Resources in Mining Cities,Ministry of Natural Resources,Jinzhong 030600,Shanxi,China;Shanxi Key Laboratory of Geological Disaster Monitoring,Early Warning and Prevention,Jinzhong 030600,Shanxi,China)

机构地区：[1]山西省煤炭地质物探测绘院有限公司,山西晋中030600 [2]自然资源部矿业城市自然资源调查监测与保护重点实验室,山西晋中030600 [3]地质灾害监测预警与防治山西省重点实验室,山西晋中030600

出　　处：《城市地质》2024年第4期490-499,共10页Urban Geology

基　　金：山西省地质勘查建设与发展基金项目(2023-007);山西省地质灾害防治重大专项项目(晋分采〔2020-00162〕G153-C53)联合资助。

摘　　要：全国自然资源更新调查的主要手段之一就是通过遥感影像提取地表覆盖物,进而掌握各种地物类型的空间分布情况,而卫星数据源是否具备高重访周期是开展该项工作数据源选择的重要考量因素之一。文章以宜兴市张渚镇为研究区,利用Sentinel-2卫星的高重访优势,通过计算不同地物类型的时序NDVI值以及在Sentinel-2影像上不同波段的光谱差异特性构建决策树分类模型。该模型包含7个节点层,根据分类对象之间的光谱特性差异确定不同决策节点所选取的阈值,最终成功分类提取了乔木林、灌木林、草地、耕地等9种地物,总体分类精度达到了88.26%。通过与只依靠地物光谱特性方法的提取结果对比可知,引入时序数据构建决策树的方法准确性显著提高,证明这种基于Sentinel-2时序数据和决策树思想的地物信息提取分类方法具有很大的实用性,可以为以后的自然资源变更调查工作提供参考。One of the main means of the National Natural Resources Update Survey is to extract the surface cover through remote sensing images,and then to grasp the spatial distribution of various land types.Whether the satellite data source has a high revisit period is one of the important considerations for the selection of data sources of this work.This paper takes Zhangzhu Town of Yixing City as the study area,utilizes the high revisit advantage of Sentinel-2 satellite,and constructs a decision tree classification model by calculating the time-series NDVI values of different feature types and the spectral difference characteristics of different bands on the Sentinel-2 image.This model contains seven node layers,and determines the thresholds selected by different decision nodes according to the differences in spectral characteristics of the classified objects,and finally successfully classifies and extracts trees and shrubs.The thresholds selected by different decision nodes are determined according to the differences in spectral characteristics between the classified objects.The overall classification accuracy reaches 88.26%.By comparing with the extraction results relying only on the spectral characteristics of the features,it can be seen that the accuracy of the method of constructing a decision tree by introducing time series data is significantly improved,which proves that this method of extracting and classifying feature information based on the idea of time series data and decision tree of Sentinel-2 is of great practicability,and it can provide a methodological reference for the future investigation of the change of natural resources.

关 键 词：哨兵2号 时序数据 光谱特性 地物分类 决策树 

分 类 号：P237[天文地球—摄影测量与遥感] P962[天文地球—测绘科学与技术]