基于叶片光谱特征的农业区域植物分类

Classification of agricultural plants based on leaf spectral features

基于农业区域8种植物的叶片光谱数据,提取63种光谱特征变量,并对全波段光谱(原始、一阶微分和包络线去除光谱)提取主成分,分别采用k最近邻(k-nearest-neighbors,kNN)、支持向量机(support vector machine,SVM)和随机森林(random forest,RF)3种机器学习方法对不同植物进行遥感分类.比较3种方法所得的总精度、训练精度、验证精度及8种植物的生产者精度.结果表明,SVM的分类性能优于kNN与RF;单一的光谱特征变量识别精度都较低(<50%);基于主成分分析的一阶微分光谱识别性能优于原始光谱和包络线去除光谱.研究指出,叶片一阶微分光谱与SVM相结合的方法能够准确识别不同植物物种.可为景观或区域尺度的植被遥感分类、精准农业和森林资源调查等提供借鉴.

With the leaf hyperspectral data of eight plant species in the agricultural region,this paper aims to extract63 spectral characteristic variables and principal components derived from three types of full-spectrum（the original,the first derivative and the continuum-removed reflectance spectrum）. We employ three machine learning methods（k-nearest-neighbors,kNN; support vector machine,SVM; random forest,RF） in the remote classification of different plants. Based on the comprehensive comparisons of the overall accuracy,training accuracy,test accuracy and species producer＇s accuracy,we find that SVM outperforms kNN and RF in vegetation classification,and single spectral characteristic variable has relatively weak classification accuracy（ 50%）. Moreover,the classification models with the PCA-based first derivative reflectance outperforms those with the PCA-based original and continuumremoved reflectance. This study demonstrates that the combination of leaf-level first derivative reflectance and SVM method can accurately identify different plant species and provide the method and theory basis for the remoteclassification of vegetation,precision agriculture and forest resource inventory at the landscape or region level.