基于细粒度校正的育种小区小麦株高无人机测量方法  

UAV Measurement of Plant Height of Breeding Wheat Based on Fine⁃scale Correction

作　　者：吴婷婷[1] 刘昕哲 聂睿琪 刘佳 武璐 李涛 WU Tingting;LIU Xinzhe;NIE Ruiqi;LIU Jia;WU Lu;LI Tao(College of Mechanical and Electronic Engineering,Northwest A&F University,Yangling,Shaanxi 712100,China;College of Agriculture,Northwest A&F University,Yangling,Shaanxi 712100,China)

机构地区：[1]西北农林科技大学机械与电子工程学院,陕西杨凌712100 [2]西北农林科技大学农学院,陕西杨凌712100

出　　处：《农业机械学报》2023年第6期158-167,共10页Transactions of the Chinese Society for Agricultural Machinery

基　　金：陕西省重点研发计划项目(2021NY-045);校级科技创新与成果转化项目(TGZX2021-30);校级学科建设专项经费项目。

摘　　要：在小麦育种田间试验中,小区群体株高是最受关注的重要农艺性状之一。针对当前无人机遥感在小麦育种小区粒度下获取株高表型精确度低的问题,提出了两种方法:基于人工测量真值的近邻校正法(Nearest neighbor correction method,NNCM)和基于多光谱+RGB数据融合的光谱指数校正法(Spectral indices correction method,SICM),近邻校正法通过获取小区群体高程信息、结合地埂进行高程校正、再依据近邻真值滑动校正得到小区精确株高;光谱指数校正法通过计算植被指数并进行指数优选,从而构建株高-植被指数精确反演模型。试验结果表明,在具有地面真值的6个时期,传统无人机作物株高测量方法的相对均方根误差(Relative root mean square error,RMSE100)分别为11.15%、59.44%、11.76%、12.31%、8.05%、59.76%;NNCM的RMSE100分别为7.17%、8.18%、5.70%、5.62%、5.65%、7.74%;SICM的RMSE100分别为7.33%、8.17%、6.05%、6.15%、6.45%、10.50%;NNCM与SICM核密度分布曲线与地面真值更加接近,中位数、四分位数、最大值、最小值偏差不超过0.5%,表明提出的2种方法均可以校正无人机测量育种小区粒度上的株高性状。本文所提2种方法具有较高的精确度和较强的鲁棒性,NNCM适用于具有地面随机采样真值的场景,SICM则适用于大范围的农田株高检测,可依据不同使用条件选择相应的校正方法。In the field experiment of wheat breeding,an important measurement index is the plant height of the plot population.To solve the problem of low accuracy of wheat plant height measurement based on UAV remote sensing,two methods were proposed,including a nearest neighbor correction method(NNCM)and a spectral index correction method(SICM).NNCM based on the true value of manual measurement,the height information of the community group was obtained,the elevation correction was carried out in combination with the ridge,and then the accurate plant height of the community was obtained by sliding correction according to the true value of the neighbor.SICM of multi⁃spectral+RGB data fusion,by calculating vegetation index and performing index optimization,an accurate inversion model of plant height⁃vegetation index was constructed.The test results showed that the relative root mean square error(RMSE100)of the traditional UAV crop height measurement method in the six periods with ground truth were 11.15%,59.44%,11.76%,12.31%,8.05%and 59.76%;the RMSE100 of NNCM were 7.17%,8.18%,5.70%,5.62%,5.65%and 7.74%;the RMSE100 of SICM were 7.33%,8.17%,6.05%,6.15%,6.45%and 10.50%;the NNCM and SICM kernel density distribution curves were closer to the ground truth,and the median,quartile,maximum,and minimum deviations did not exceed 0.5%.These indicated that both the proposed methods can correct the plant height traits on the grain size of breeding plots measured by UAV.The two models proposed had high accuracy and strong robustness.NNCM was suitable for the scene of random sampling of ground truth on the ground,while SICM was used for plant height detection of large⁃scale farmland,and different methods were selected according to the using conditions.

关 键 词：小麦 株高 无人机遥感 植被指数 多光谱成像 高通量表型 

分 类 号：S123[农业科学—农业基础科学]