基于多源遥感数据的复杂地形区农作物分类

Classification of Crops in Complicated Topography Area Based on Multisource Remote Sensing Data

为准确获取中尺度复杂地形区西宁市内的农作物分布信息,该文利用HJ CCD和Landsat 8OLI影像数据提取研究区内农作物4-11月的NDVI时间序列数据,同时利用HJ-1A HSI高光谱数据,通过光谱特征变量提取和单因素方差分析后,选取5种光谱特征变量与NDVI时间序列数据组成多源数据集,并最终采用分类回归决策树(CART)和支持向量机(SVM)两种方法进行农作物分类。结果表明:1)采用多源数据集进行作物分类,最高分类精度达88.2%,明显高于采用单一时序NDVI数据的分类精度,表明在时间序列数据中融入高光谱数据能够提高作物的识别精度;2)采用CART决策树和支持向量机进行农作物识别,最优总体分类精度分别为88.2%和84.5%,表明基于CART方法的作物总体分类效果较优;3)在NDVI时间序列数据中融入高光谱数据能够提高生育期较为接近作物的识别精度。

In order to accurately get the information of crops distribution in mesoscale complicated topography area,in this paper,Xining City was taken as the study area,HJC CD and Landsat 8 OLI imaging data were utilized to extract NDVI time series data about crops in the study area from April to November,and HJ-1A HSI hyperspectral data were utilized to extract spectral signature variables and to perform one-way analysis of variance,then 5 kinds of spectral characteristic variables and NDVI time series data were selected to constitute multisource data set.And ultimately the project adopted classification and regression tree(CART)and support vector machine(SVM)to conduct crops classification.The results showed that:1)the precision of multisource data set for crops classification was apparently higher than that of single time series NDVI data,with the maximum classification precision of 88.2%,presenting combination of time series data and hyperspectral data can improve identification precision of crops;2)the optimized general classification precision of CART decision tree and SVM used to identify crops were 88.2%and 84.5%respectively,indicating the crops general classification effect of CART method was superior;3)combination of NDVI time series data and hyperspectral data can improve the identification precision of crops whose growth periods are relatively close.