| Ratoon rice is a kind of low-cost planting mode of rice,which can make full use of the labor,seeds,fertilizer invested in the main rice and the idle thermal,light and land resources after the main rice harvest.It is of great significance to reduce the cost and increase the efficiency of rice production.Leaf area index(LAI)is an important parameter to characterize crop growth state,but the traditional measurement methods are difficult to obtain large-area and high spatial-temporal resolution LAI change monitoring of ratoon rice.The development of near ground remote sensing technology provides a new tool for the monitoring of field crop LAI with high temporal and spatial resolution,which can provide data support for the precision management of ratoon rice.Therefore,this study uses the ground hyperspectral and low altitude UAV platforms with a multi-spectral sensor to obtain the hyperspectral data and multi-spectral images of ratoon rice canopy respectively,the response of hyperspectral and multispectral features to LAI of ratoon rice at different growth stages was analyzed.and the influence of spectral index and texture features fusion modeling on the accuracy of inversion model was explored.the LAI inversion models of single growth stage and multiple growth stages were established,and the field LAI distribution map was plotted based on the optimal model.The main research results include:(1)The ground hyperspectral features and LAI were modeled and analyzed,results showed that the ratio vegetation index(RSI)and the normalized vegetation index(NDSI)were the best indicators of LAI for ratoon rice,the R~2(coefficient of determination)of the models at booting stage(847 nm,749 nm),heading stage(999nm,718 nm),filling stage(716 nm,539 nm)and late filling stage(357 nm,351 nm)were 0.83,0.68,0.54 and 0.45,respectively.RMSE(root mean square error)were 0.23,0.41,0.33 and 0.27,respectively.(2)In the multispectral feature-based model,the spectral index features with the highest inversion accuracy were MSRRE,MSRRE,CIgreen and Ex G-Ex R,with R2of0.38,0.75,0.41 and 0.44,and RMSE were 0.51,0.27,0.43 and 0.40,respectively.The accuracy of multispectral feature-based model was lower than that of hyperspectral index model at booting stage,filling stage and late filling stage.However,the accuracy of multi-spectral image texture feature-based model at booting stage,filling stage and late filling stage is better than that of hyperspectral feature-based model,and the features were B＿MEA,RE＿HOM and RE＿SEC,respectively.The inversion accuracy were 0.68,0.75,0.50,0.50,respectively,and RMSE were 0.37,0.27,0.40,and 0.38,respectively.(3)The spectral index and texture features of UAV multispectral images were fused and modeled.The results showed that the fusion of spectral index features and texture features could improve LAI inversion accuracy of ratoon rice.The accuracy of fusion model at booting stage,heading stage,filling stage and late filling stage were0.77,0.68,0.84,0.75,and RMSE were 0.30,0.38,0.22,0.27,respectively.The fusion of spectral index features and texture features can bridge the gap in spectral resolution and number of bands between multispectral and ground hyperspectral to a certain extent.(4)Simple regression model,multiple linear stepwise regression model and support vector machine model were used to establish inversion models with different growth gap respectively.The results show that the inversion ability of support vector machine model is the best,and the inversion accuracy of the model based on the fusion of spectral index features and texture features of support vector machine is 0.91 and RMSE is 0.26 across four growth stages.In conclusion,this study revealed the response law of canopy spectral features of ratoon rice to LAI.Based on the fusion modeling of UAV multi-spectral image features,LAI inversion and LAI spatial and temporal distribution mapping of ratoon rice in each growth period were completed,which can provide data support for the precision management of ratoon rice. |
PDF Full Text Request