UAV multispectral images for accurate estimation of the maize LAI considering the effect of soil background
Shuaibing Liu, Xiuliang Jin, Yi Bai, Wenbin Wu, Ningbo Cui, Minghan Cheng, Yadong Liu, Lin Meng, Xiao Jia, Chenwei Nie, Dameng Yin
Abstract
The high proportion of soil background pixels in UAV remote sensing images is an important reason for the uncertainty of high-precision leaf area index (LAI) estimation at early growth stages of crops. Although the traditional method of removing soil pixels from images based on canopy coverage (CC) eliminates pure soil pixels, it can cause spectral saturation at the early stages and therefore affect the accuracy of LAI estimation. In this study, a new method called reduced soil contribution (CS) was constructed to improve the accuracy of LAI estimation. This method can be improved by introducing a quantitative method to account for the contribution of soil information, which can be used to correct the calculation of vegetation indices and eliminate soil interference in maize LAI estimation. A six-rotor UAV equipped with a multispectral camera was used to collect field image data. Experimental plots with different maize breeding varieties were laid out to carefully evaluate the accuracy of the estimation model using UAV multispectral images collected at different growth stages. The performance of four estimation models, a light gradient boosting machine, gradient-boosting decision tree, random forest regression model and extreme gradient boosting, for LAI estimation was evaluated. The CS-based approach significantly improved the accuracy of the LAI model estimation, reducing the rRMSE by 1.89% for a single growing season compared to the traditional method. On average, the rRMSE for all growth stages decreased by 3.5%, demonstrating its effectiveness in improving maize LAI estimation accuracy. Randomness in error measured by Moran’s I metrics showed that the GBDT (gradient-boosting decision trees) model based on the CS method showed less spatial aggregation. These results showed that the CS model can effectively reduce the influence of soil on the estimation of the maize LAI and improve the LAI estimation accuracy compared with the direct removal of soil background pixels from an image.