基于综合指标的冬小麦长势无人机遥感监测

作物长势监测可以及时获取作物的长势信息,该文尝试建立新型长势指标,监测小麦总体长势情况。将反映小麦长势的叶面积指数(leaf area index,LAI)、叶片叶绿素含量、植株氮含量、植株水分含量和生物量5个指标按照均等权重综合成一个指标,综合长势指标(comprehensive growth index,CGI)。利用450~882 nm范围内单波段和任意两个波段构建归一化光谱指数(normalized difference spectral index,NDSI),比值光谱指数(ratio spectral index,RSI)和简单光谱指数(simple spectral index,SSI),计算CGI与光谱指数的相关性,筛选出相关性好的光谱指数,结合偏最小二乘回归(partial least squares regression,PLSR)建立反演模型。以CGI为指标,运用无人机高光谱影像对2015年小麦多生育期的长势监测。结果表明:1)冬小麦各生育期,总体上CGI与光谱指数的决定系数R~2均好于各项单独指标与相应光谱指数的R~2。仅孕穗期CGI和RSI(754,694)的R~2比叶绿素和RSI(486,518)的R~2低,开花期的CGI和R570的R~2比生物量和R834的R~2低以及灌浆期CGI和SSI(582,498)的R~2比植株含水量和SSI(790,862)的R~2低。2)拔节期,孕穗期,开花期,灌浆期和全生育期PLSR模型的建模R~2分别为0.70,0.72,0.78,0.78和0.61。拔节期,孕穗期和开花期的无人机CGI影像验证模型的均方根误差RMSE(root mean square error)分别为0.050,0.032和0.047。CGI与相应光谱指数的R~2高于单独各项指标与相应光谱指数的R~2,光谱指数能够很好反映CGI包含的信息。无人机高光谱影像反演CGI精度较高,能够判断出小麦总体的长势差异,可为监测小麦长势提供参考。

Remote sensing monitoring of winter wheat growth with UAV based on comprehensive index

Crop growth monitoring is an important research field of agricultural remote sensing. Crop growth monitoring can have timely access to crop growth information, and it is significant for crop management. The use of unmanned aerial vehicles (UAV) to monitor crop growth has become a fast and effective way. Some indicators can be used to characterize the growth of wheat including leaf area index (LAI), leaf chlorophyll content (LCC), plant nitrogen content (PNC), plant water content (PWC) and biomass of wheat. LAI and biomass were used to characterize the population growth of wheat. LCC, PNC and PWC were used to characterize the individual growth situation of wheat. This paper integrated the LAI, LCC, PNC, PWC and biomass of wheat, 5 indicators of wheat growth, into an index named comprehensive growth index (CGI) according to the equal weight for each indicator. The normalized difference spectral index (NDSI), ratio spectral index (RSI) and simple spectral index (SSI) were constructed by using single band and any 2 bands in the range of 450-882 nm. Then, the correlations between the CGI and the spectral indices were analyzed, and the spectral indices were selected based on good correlation between CGI and spectral indices i.e. NDSI, RSI, and SSI. At the same time, the spectral indices were constructed and selected by the correlation between LAI, LCC, PNC, PWC and biomass and spectral band. The inversion model was established with spectral indices and the partial least squares regression (PLSR). Four inversion models were established during jointing, booting, flowering, and grain filling period, respectively. CGI and the hyperspectral image of UAV were used to reflect the growth of wheat in different growth stages, realizing the monitoring of wheat growth during jointing, booting and flowering period. And through the hyperspectral image of UAV, the growth of wheat and difference of different growth stages of wheat can be monitored. The hyperspectral image information after inversion is extracted to verify the accuracy of the model for different growth stages. The results were as follows: 1) Some spectral indices and sensitive single band were selected for CGI in different growth stages of winter wheat. The spectral indices NDSI(746,742), RSI(746,742), SSI(526,454)and single band 690 nm were selected in wheat jointing period. The spectral indices NDSI(758,694), RSI(754,694), SSI(684,566) and single band 662 nm were selected in wheat booting stage. The spectral indices NDSI(774,758), RSI(774,758), SSI(510,486)and single band 570 nm were selected in wheat flowering period. The spectral indices NDSI(754,746), RSI(754,746), SSI(582,498) and single band 702 nm were selected in wheat filling period. 2) Overall, the coefficient of determinationR2 between CGI and spectral indices was higher than that of the individual index and the corresponding spectral index in the winter wheat growth period. Only theR2 between CGI and RSI(754,694) was lower than that between RSI(486,518) and LCC in the booting period. TheR2 between CGI and R570 was lower than that between biomass of wheat and R834 in flowering period. TheR2 between CGI and SSI(582,498) was lower than that between PWC and SSI(790,862) in filling period. 3) TheR2 values of modeling PLSR at jointing stage, booting stage, flowering stage, filling stage and full growth stage were 0.70, 0.72, 0.78, 0.78, and 0.61 respectively. For model validation, the NRMSE (standard root mean square error) values were 0.066, 0.038 and 0.062, and the RMSE (root mean square error) values were 0.050, 0.032 and 0.047, respectively, at the jointing stage, booting stage and flowering period. Overall, the R2 of the CGI and the corresponding spectral index is higher than that of each index and the corresponding spectral index. The spectrum can well reflect the information contained in CGI. The PLSR modeling and image validation accuracy are high. It is feasible to use the integrated growth index to monitor the growth of wheat, which is of high precision.