Comparison of different machine learning algorithms for predicting maize grain yield using UAV-based hyperspectral images

• DOI: 10.1016/j.jag.2023.103528
• Publication Date: 2023-10-21T10:24:09Z
• Authors: Yahui Guo; Yi Xiao; Fanghua Hao; Xuan Zhang; Jiahao Chen; Kirsten de Beurs; Yuhong He; Yongshuo H. Fu
• Keywords: Precision Agriculture; Spectral bands

Timely and accurately predicting maize grain yields will contribute to making adaptive measures improve management practice adjust consumption patterns for ensuring food security. Unmanned aerial vehicles (UAV) are widely used obtain high-temporal high-spatial resolution remote sensing images of crops, enabling a possible sensor performance comparison. To date, few studies have compared the potential abilities multispectral-based hyperspectral-based images, only sensitive spectral wavelength full hyperspectral spectra, various machine learning approaches in estimating physiological characteristics such as chlorophyll meter values, leaf area index (LAI), agricultural high vegetation coverage. In this study, multispectral with ground measurement crop traits were collected on 13 22 September 2021 Nanpi experimental station, CangZhou, China. The ability retrieving LAI, explored using formed two-band (2D) indices (VIs) 2D textural (TIs). wavelengths confirmed correlation analyses, then VIs spectra also yield five commonly applied deep convolutional neural network (CNN). results indicated narrow bands remained than adoption significantly improved accuracy predictions adopting built sensitivity wavelength. Based selected VIs, random forest regression (RF) LightGBM achieved highest accuracy, R2 (RMSE) 0.90 (0.55 t/ha), 0.85 (0.59 respectively. While based RF CNN150 performed best, being 0.92 (0.53 0.91 This research concluded integration combination highly recommended yields, especially crops