摘要
精准、动态监测经济作物种植信息是农业精细化管理面临的迫切需求。为实现不同果树品种的精细分类,以桂林市六塘默科特柑橘试验基地为研究区,获取机载高光谱影像数据,深度挖掘不同柑橘果树品种的光谱信息,构建多维数据集,提出一种利用SULOV结合极端梯度提升(eXtreme gradient boosting,XGBoost)算法进行优选特征,并采用XGBoost分类算法进行柑橘果树品种精细分类的方法,最后,与随机森林(random forest,RF)和支持向量机(support vector machine,SVM)的分类结果的精度进行对比分析。结果发现:(1)所提的SULOV结合XGBoost算法(SULOV_XGBoost)柑橘果树精细分类算法能够有效进行特征差距较小场景的果树作物不同品种间的精细分类,算法整体分类效果优于传统的常用机器学习方法(RF与SVM);(2)一阶微分拐点处值与原始波段值的融合特征对提升精细分类精度具有极大作用;另外加入不同波长范围波段组合也能够显著提高柑橘果树精细分类结果;(3)SVM在地物可辨性较高的条件下其分类性能更佳,且抗干扰能力强。研究成果可为同一物种不同品种作物的精细分类提供新的思路和方法,亦可为作物种植信息精准普查、精细化管理以及农业产业结构布局、调整和动态监测等提供参考。
Accurate and dynamic monitoring of economic crop planting information is an urgent need for agricultural fine management.In order to realize the fine classification of different fruit tree varieties,this paper proposes a fine classification method of citrus fruit trees based on UAV hyperspectral images and the SULOV_XGBoost algorithm in liutang Mocott citrus experimental base in Guilin City.Firstly,multidimensional data sets were constructed by deep mining spectral information from different citrus tree varieties.Then,the SULOV_XGBoost algorithm was used to optimize features,and the XGBoost algorithm was used for the fine classification of citrus fruit varieties.Finally,the accuracy of classification results was compared with that of RF and SVM.The results show that:(1)The proposed SULOV_XGBoost algorithm can effectively classify the different varieties of fruit trees and crops in scenes with small feature gaps,and the overall classification effect is better than the traditional machine learning methods(RF and SVM).(2)The fusion characteristics of the first-order differential inflection point value and the original band value play a great role in improving the precision of fine classification;the combination of different wavelength bands can also significantly improve the fine classification results of citrus fruit trees.(3)SVM has better classification performance and strong anti-interference ability under high ground object discrimination.The research results can provide new ideas and methods for fine classification of different varieties of crops in the same species,and also provide a reference for the precise survey of crop planting information,fine management and layout,adjustment and dynamic monitoring of agricultural industrial structure.
作者
肖斌
何宏昌
窦世卿
范冬林
付波霖
张洁
熊远康
史今科
XIAO Bin;HE Hong-chang;DOU Shi-qing;FAN Dong-lin;FU Bo-lin;ZHANG Jie;XIONG Yuan-kang;SHI Jin-ke(College of Surveying and Mapping Geographic Information,Guilin University of Technology,Guilin 541006,China;Pearl River Water Resources Research Institute,Pearl River Water Resources Commission,Guangzhou 510610,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2024年第2期548-557,共10页
Spectroscopy and Spectral Analysis
基金
广西八桂学者专项项目,国家自然科学基金项目(42061059,32371966)
桂林市科技局开发项目(2020010701)
广西科技计划项目(20159037)资助。
