杏树遥感图像辨识的最佳时相与方法  

The Best Phase and The Optimal Identification Method for Identifying Apricot Tree from Remote Sensing Images

作　　者：邢东兴 王明军[2] 焦俏 车自力 封建民[1] 杨波[1] XING Dongxing;WANG Mingjun;JIAO Qiao;CHE Zili;FENG Jianmin;YANG Bo(Department of Resources&Environment,Xianyang Normal University,Xianyang 712000,China;Department of Physics&Electronic Engineering,Xianyang Normal University,Xianyang 712000,China)

机构地区：[1]咸阳师范学院资源环境学院,陕西咸阳712000 [2]咸阳师范学院物理与电子工程学院,陕西咸阳712000

出　　处：《河南农业科学》2020年第1期165-173,共9页Journal of Henan Agricultural Sciences

基　　金：国家自然科学基金项目(61771385);陕西省优势学科建设项目(060103)

摘　　要：采用2014、2017、2018年逐月GF1-WFV影像(共43景)探寻杏树遥感图像辨识的最佳时相与方法,以期为关中乃至全国其他果区开展杏树遥感监测提供理论依据。首先对各景影像分别进行预处理;随后利用在各期影像中采集到的各种果树样地的ROI(感兴趣区)数据,对6类辨识方法(即同期影像地物反射光谱比较、同期影像波段差值或比值分析、同期影像光谱指数求算与分析、同套邻期光谱指数变化追踪、影像复合与多指数联用分析)的辨识效能分别予以探试,以寻求最佳的辨识时相与方法;最后对探试结果用于全域影像中的辨识效能加以验证。结果显示:盛花期杏树相对其他果树树种具有较低的VI1值,用VI1阈值对该期杏树具有较佳的辨识效能;在杏树盛花期的影像中,利用NDVI(归一化植被指数)与VI1双重阈值,可明显提高总体分类精度;利用盛花期与花前影像(同序号)波段的比值也可较好地辨识杏树;联合应用NDVI、VI1、R b1/R b1花前、R b3/R b3花前4个光谱指数阈值,辨识花期杏树的精度更为理想,杏树类的正确识别率可达83.14%,总体分类精度可达80.93%;杏树盛花期是辨识杏树的最佳时相。The aim is to find out the best phase and the optimal method for identifying apricot trees by using the monthly GF1-WFV images collected in 2014,2017 and 2018(a total of 43 images),and to provide a theoretical basis for remote sensing monitoring of apricot trees in Guanzhong area and even all fruit regions of China.Firstly,the images of each period were preprocessed.Then,the identification efficiencies of six kinds of identification methods(comparison of reflectance spectra of ground objects in the same period of image,bands difference or ratio analysis in the same period of image,calculation and analysis of spectral index in the same period of image,spectral index change tracking among the adjacent periods of images divided into the same set,images compounding and analysis of the multiple indices combined using)were tested separately by using ROI(Region of interest)data collected from the sample plots of various fruit trees in different periods of images.Finally,the identification efficiency of exploration results was verified in global image.The conclusions were as follows:Compared with other fruit tree species,apricot trees had lower VI1 value at full flowering stage,so the thresholds of VI1 value had better identification efficiency for identifying apricot trees in this period;In the images of apricot trees blooming,the overall classification accuracy had been improved by using dual thresholds of NDVI(Normalized difference vegetation index)and VI1;Apricot trees could also be well identified by the band ratios between full flowering and pre-blooming period images(same serial number);The accuracy of identifying apricot trees was more ideal by using quadruple thresholds of NDVI,VI1,R b1/R b1 Pre-flowering and R b3/R b3 Pre-flowering,the classification accuracy of apricot trees could reach 83.14%,and overall classification accuracy could reach 80.93%;The flowering period was the best time to identify apricot trees.

关 键 词：关中地区 杏树 GF1-WFV 最佳时相 辨识方法 光谱指数 决策树分类 

分 类 号：S771.8[农业科学—森林工程]