基于无人机可见光影像的玉米冠层SPAD反演模型研究

SPAD Inversion Model of Corn Canopy Based on UAV Visible Light Image

叶绿素是植物进行光合作用的重要色素,利用作物光谱、纹理信息对叶绿素进行反演,为作物的实时监测和健康状况诊断提供重要依据。以大田环境下5个不同品种四叶期、拔节期的玉米为研究对象,利用无人机获取试验区可见光影像,对土壤背景进行掩膜处理,提取25种可见光植被指数、24种纹理特征,综合分析植被指数、纹理特征与玉米冠层叶绿素相对含量(SPAD)的相关性,分别建立基于植被指数、纹理特征和植被指数+纹理特征的逐步回归(SR)、偏最小二乘回归(PLSR)和支持向量回归(SVR)模型,定量估算叶绿素相对含量。在SR模型中,植被指数+纹理特征模型与植被指数模型相同,R2为0.7316,RMSE为2.9580,RPD为1.926,优于纹理特征模型;在PLSR模型中,植被指数+纹理特征模型较优,R2为0.8025,RMSE为2.4952,RPD为2.284,纹理特征模型次之,植被指数模型最差;在SVR模型中,植被指数+纹理特征模型较优,R2为0.8055,RMSE为2.6408,RPD为2.158,植被指数模型次之,纹理特征模型最差。综合分析采用基于PLSR植被指数+纹理特征模型可以实现玉米冠层SPAD快速、准确提取,为叶绿素反演提供一种新的方法,可为无人机遥感作物长势监测提供参考。

Chlorophyll is an important pigment in photosynthesis of plants.It can provide important basis for real-time monitoring and health diagnosis of crops by using crop spectral and texture information to retrieve chlorophyll.In the field environment,five different varieties of corn at four leaves stage and jointing stage were selected as the research objects.The visible light images were obtained by UAV,and the soil background was dealt with mask treatment.Totally 25 kinds of visible light vegetation index and 24 kinds of texture features were extracted.The correlation between vegetation index,texture feature and relative chlorophyll content(SPAD)of corn was comprehensively analyzed,and stepwise regression(SR),partial least squares regression(PLSR)and support vector regression(SVR)models based on vegetation index,texture feature and vegetation index+texture feature were respectively established to quantitatively estimate relative chlorophyll content.In SR model,vegetation index+texture feature model was the same as vegetation index model,which was better than texture feature model,R2 was 0.7316,RMSE was 2.9580,RPD was 1.926;in PLSR model,vegetation index+texture feature model was better,texture feature model was the second,vegetation index model was the worst,R2 was 0.8025,RMSE was 2.4952,RPD was 2.284;in SVR model,vegetation index+texture feature model was better,vegetation index model was next,texture feature model was the worst,R2 was 0.8055,RMSE was 2.6408,RPD was 2.158.Comprehensive analysis using the PLSR-based vegetation index+texture feature model can achieve rapid and accurate extraction of corn SPAD,providing a new method and experience for chlorophyll inversion,and also providing a reference for UAV remote sensing growth monitoring.