Estimate leaf area index and leaf chlorophyll content in winter-wheat using image texture and vegetation indices derived from multi-temporal RGB images

Abstract

Abstract Aims Investigating the potential of combining data dimensionality reduction methods with various linear regression models and machine learning algorithms to improve the accuracy of leaf area index (LAI) and chlorophyll content (LCC) estimation in winter wheat based on UAV RGB imagery. Methods Constructed and compared the performance of three linear regression techniques: multiple linear regression (MLR), ridge regression (RR), and partial least squares regression (PLSR) and three machine learning algorithms: back-propagation neural networks(BP), random forests (RF) and support vector regression (SVR) with spectral vegetation indices (VIs), texture features (TEs) and their combinations extracted from UAV RGB images. Moreover, different data dimensionality reduction methods include principal component analysis (PCA), and stepwise selection (ST) were used to improve the accuracy of LAI and LCC estimation. Results The highest correlation between texture features and LAI, LCC was obtained with window size 5 × 5, orientation 45° and displacement 2 pixels. Combining VIs and TEs improved the accuracy of LAI and LCC estimation for wheat compared to using VIs or TEs alone. The RF model combined with ST_PCA for fusing VIs and TEs achieved the best estimations, with R2 of 0.86 and 0.91, RMSE of 0.26 and 2.01, and MAE of 0.22 and 1.66 for LAI and LCC, respectively. Conclusions The fusing of multiple features improved the accuracy of LAI and LCC estimation. ST_PCA, combined with machine learning algorithms, holds promising potential for monitoring crop physiological and biochemical parameters.