基于无人机多光谱影像的完熟期玉米倒伏信息提取

以吉林省梨树县的玉米试验田为研究区,按受灾后完熟期玉米的状态将研究区分为倒伏、半倒伏和未倒伏3种类型。基于无人机采集的多光谱影像提取15种光谱指数和8种纹理特征,采用面向对象法、最大似然法和多元Logistic回归模型进行玉米倒伏信息的提取;而后通过目视方法选取400个样本点进行玉米倒伏信息提取结果的精度验证。结果表明：面向对象法精度最高,对玉米3种倒伏状态信息识别的总体精度为88.13%,Kappa系数为0.83。研究用于区分倒伏与未倒伏玉米的最佳光谱指数是归一化差异植被指数,对区分倒伏与半倒伏、半倒伏与未倒伏玉米贡献最大的特征均为对比度纹理特征。研究表明基于无人机多光谱影像的面向对象方法在对田块尺度玉米倒伏信息的精准识别中具有较大潜力。

Extraction of maize lodging information at mature stage based on UAV multispectral images

A maize field in Lishu County, Jilin Province was selected as the study area in this study.There were three types of maize status, including lodging, semi|lodging, and unlodging maize respectively, according to the state of maize at the mature stage after the typhoon disaster. Based on multi|spectral images collected by UAV, 15 spectral indices and 8 texture features were extracted. Object|oriented method, maximum likelihood method and multiple logistic regression model were used to extract lodging information of maize.Then, the 400 sample points were selected by visual method to verify the accuracy of maize lodging information extraction results.The results showed that the object|oriented method had the highest accuracy with the overall accuracy of 88.13% and the Kappa coefficient of 0.83. In this study, the best spectral index used to distinguish lodging and unlodging maize was normalized differential vegetation index, and the features that contributed most to distinguishing lodging, and semi|lodging and semi|lodging and unlodging maize were contrast texture features. This study showed that the object|oriented method based on UAV multispectral imagery has great potential in accurately identifying field|scale maize lodging information.