基于无人机可见光数据荒漠灌木覆盖度提取方法研究

(1.内蒙古农业大学, 呼和浩特 010018; 2.水利部 牧区水利科学研究所, 呼和浩特 010020)

无人机; 植被指数; 荒漠灌木; 可见光数据

Research on Extraction Method of Desert Shrub Coverage Based on UAV Visible Light Data
CUI Wanxin1, LI Jinrong2, SI Qiancheng1, WANG Ru1, LUO Xiangying1, YANG Feng1, LI Yingkun1

(1.Inner Mongolia Agricultural University, Hohhot 010018, China; 2.Institute of Pastoral Water Science, Ministry of Water Resources, Hohhot 010020, China)

UAV; vegetation index; desert shrubs; visible light data

备注

为了提高荒漠植被的提取精度,以黄河乌兰布和沙漠段东岸的灌木为研究对象,基于可见光波段荒漠灌木的光谱特性,从指数、阈值方法、影像分辨率角度,对可见光波段差异植被指数(VDVI)、过绿指数(EXG)、归一化绿蓝差异指数(NGBDI)、归一化绿红差异指数(NGRDI)、红绿比值指数(RGRI)、过绿减过红指数(EXG-EXR)6种植被指数特征进行了对比和影响因素分析。结果表明:在荒漠灌丛研究区内,EXG和EXG-EXR指数直方图为双峰特征,其余均为非双峰特征。针对峰值特征,分别采用直方图阈值法和阈值迭代法计算阈值可行。VDVI,EXG-EXR和NGBDI的平均提取精度为98.11%,87.25%,84.73%,Kappa系数均大于0.7(K>0.61),适合提取荒漠灌木植被,其余植被指数提取精度均低于80%; 在航拍影像出现明暗纹理不均一的情况下,除NGBDI指数灰度图外,其他指数灰度图均出现阴影纹理,但仅EXG指数提取精度受阴影纹理影响,精度由77.03%降低到64.88%; 综合地面像素分辨率、影像拼接效果(明暗度、完整性)对提取精度的影响,40 m和50 m的飞行高度是野外调查的优先选择高度。
In order to improve the accuracy of desert vegetation extraction, taking the shrubs on the east bank of the Ulanbu and desert sections of the Yellow River as the research object, based on the spectral characteristics of the desert shrubs in the visible light band, from the perspective of index, threshold method and image resolution, we compared the characteristics of difference vegetation index(VDVI)and green index(EXG), normalized green-blue difference index(NGBDI), normalized green-red difference index(NGRDI), red-green ratio index(RGRI), excessive green minus red index(EXG-EXR)of 6 types of plants and analyzed the influencing factors. The results showed that in the desert shrub study area, the histograms of EXG and EXG-EXR indices were bimodal, and the rest were non-bimodal; for the peak characteristics, it was feasible to use the histogram threshold method and the threshold iteration method to calculate the threshold; the average extraction accuracy of VDVI, EXG-EXR and NGBDI was 98.11%, 87.25% and 84.73%, respectively, and the Kappa coefficients were greater than 0.7(K>0.61), which was suitable for extracting desert shrub vegetation, and the extraction accuracy of other vegetation indices was less than 80%; when the aerial image had uneven light and dark textures, all the index grayscale images except the NGBDI index grayscale image showed shadow texture, but only the EXG index extraction accuracy was affected by the shadow texture, and the accuracy was reduced from 77.03% to 64.88%; flying heights of 40 m and 50 m were the preferred altitudes for field surveys through integrating the effect of ground pixel resolution and image stitching effect(brightness and completeness)on the extraction accuracy.