基于无人机影像的采煤沉陷区玉米生物量反演与分析  

Inversion and Analysis of Maize Biomass in Coal Mining Subsidence Area Based on UAV Images

作　　者：肖武[1,2] 陈佳乐 笪宏志[1] 任河 张建勇 张雷[1] XIAO Wu;CHEN Jiale;DA Hongzhi;REN He;ZHANGJlanyong;ZHANG Lei(Institute of Land Reclamation and Ecological Rehabilitation,China University of Mining and Technology,Beijing 100083,China;School of Public Affairs,Zhejiaag University,Hangzhou 310058,China)

机构地区：[1]中国矿业大学(北京)土地复垦与生态重建研究所,北京100083 [2]浙江大学公共管理学院,杭州310058

出　　处：《农业机械学报》2018年第8期169-180,共12页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金项目(41401609);山东省重点研发计划项目(2016ZDJS11A02);中央高校基本科研业务费专项资金项目(ZJUGG201801)

摘　　要：为了探索运用无人机多光谱遥感技术监测高潜水位矿区采煤扰动下原有生态系统破坏及地表耕地损毁程度的方法,以高潜水位矿区开采沉陷导致地面积水所引起的农作物渍害影响为例,基于无人机多光谱影像,在传统植被指数的基础上引入红边波段进行扩展,优选了22种植被指数,结合田间同步实测生物量数据,采用经验模型法分别构建了一元回归、基于最小二乘法的多元逐步回归(Multivariable linear regression,MLR)、反向传播神经网络(Back propagation neural networks,BPNN)的生物量反演模型,通过决定系数(R2)、均方根误差(RMSE)和估测精度(EA)3个指标筛选出最佳模型。最后,基于最佳模型进行研究区玉米生物量的空间分布反演和分析,结果显示,所选的植被指数均与生物量显著相关,其中,BP神经网络模型的估算精度最高,其决定系数R2为0.83,比其他模型增加了0.10~0.17,预测均方根误差RMSE为178.72 g/m2,比其他模型减少了29.65~60.23 g/m2,估测精度EA可达到79.4%,比其他模型提高了3.3%~7.1%。这说明红边波段更适于采煤沉陷区作物生物量的估算,引入红边波段构建生物量反演模型,可以显著提高采煤沉陷影响下玉米生物量无人机遥感反演模型的精度。研究结果表明:采煤沉陷盆地内玉米生物量主要分布于592~1 050 g/m2,其面积占研究区的74.4%,地表生物量低于352 g/m2的作物面积达到14.1%,玉米整体长势受采煤扰动影响较为严重,玉米生物量呈现从沉陷盆地边缘往中心逐渐降低的趋势。本文研究为同类型其他高潜水位矿区土地损毁监测与评价、土地复垦与生态修复等提供基础数据与理论支撑。The surface arable land damage and destruction of the original ecosystem caused by the influence of coal mining disturbance are the major ecological disasters in the high underground water mining area. Identifying an arable-damaged area and obtaining its spatial distribution are important for ecological disaster monitoring. The influence of crop waterlogging caused by mining subsidence in high underground water mining areas was taken as an example,and based on the UAV multi-spectral images,the red band was introduced on the basis of traditional vegetation index to expand,which allowed to select the best 22 VI. Univariate regression,multivariable linear regression（MLR） based on the principle of least square method and back propagation neural networks（BPNN） were built accordingly by using the22 VI along with field measurements of biomass data under the empirical modeling method. There were three indices should be taken into account to determine the optimal model,which were coefficient of determination（R2）,root mean square error（RMSE） and estimation accuracy（EA）. The spatial distribution inversion and analysis of maize biomass were undertaken in the study area by using the selected optimal model. It was concluded that the selected vegetation index was significantly related tobiomass. And the highest estimation accuracy was obtained by using BP model. The value of R2 was 0. 83 accordingly,which was generally increased by 0. 10 ~ 0. 17. The value of predicted root mean square error（RMSE） was 178. 72 g/m2,which was generally reduced by 29. 65 ~ 60. 23 g/m2. The estimation accuracy（EA） could eventually reach 79. 4%,which was increased by 3. 3% ~ 7. 1%. It can be concluded that the red edge band was more suited to the estimation of crop biomass in the mining subsidence area. Furthermore,the accuracy rate of the inversion model under the influence of coal mining subsidence could be increased dramatically by introducing red edge band to the construction of biomass inversion model. The research s

关 键 词：无人机 采煤沉陷区 生物量反演 植被指数 土地复垦 高潜水位 

分 类 号：TP722.5[自动化与计算机技术—检测技术与自动化装置]