Integrating Vegetation Indices and Texture Features from UAV multispectral image for Non-destructive Peanut Aboveground Biomass Estimation

Abstract. High-throughput phenotyping monitoring has become increasingly important in modern agriculture, as it can collect plant images to extract and analyse phenotype data related to growth and yield, thereby reducing crop monitoring costs. Aboveground biomass (AGB) is a key indicator for evaluating plant health, growth, and productivity, and reflects the impact of environmental factors (such as water, soil nutrients, and temperature) on plants. However, traditional methods for measuring AGB are often labor-intensive, costly, and limited in spatial coverage. Unmanned aerial vehicles (UAVs)-based remote sensing offer new solutions, enabling large-scale, high-resolution data collection in agricultural fields. Therefore, this study evaluates the use of Vegetation indices (VIs) and Texture features (TFs), as well as their combinations, derived from UAV multispectral imagery to estimate peanut AGB across different growth stages. Specifically, nine VIs and eight TFs with different parameter settings were first derived from RGB and four single-band UAV images. Based on random forest (RF) regression, the study explored the impact of different parameter combinations on the performance of AGB models and analysed the potential of combining VIs and TFs to improve AGB estimation. The results show that TFs effectively complement VIs, significantly enhancing peanut AGB estimation performance. The optimal window size was 7×7, with a direction of 90° and a grey level of 16. The combined VIs and TFs yield a regression with R² and RMSE of 0.929 and 0.032, respectively. These findings suggest that the strategy of extracting image textures and combining features significantly improves the accuracy of AGB estimation, providing a more precise method for monitoring AGB.