Monitoring Maize Growth Using a Model for Objective Weight Assignment Based on Multispectral Data From UAV

Abstract

Agricultural development and production management crucially depend on efficient and accurate acquisition of crop growth information. This study focuses on maize, employing drones to monitor its growth based on metrics such as plant height (PH), SPAD values and leaf area index (LAI). Using the entropy weighting method (EWM) and coefficient of variation method (CV), comprehensive growth indices, CGMI_EWM and CGMI_CV, were developed. These indices were correlated with 10 vegetation indices to select those with significant relevance. Subsequently, three machine learning methods—partial least squares (PLS), random forest (RF) and particle swarm optimisation-enhanced random forest (PSO-RF)—were utilised to construct models for inversely monitoring maize growth. The optimal model was determined through evaluative metrics, leading to the acquisition of spatial distribution information on maize growth within the study area. The results indicate that the CGMI_EWM derived from the entropy weight method shows a higher correlation than individual indices, significantly enhancing model precision over traditional single-index monitoring. Among the modelling techniques, the PSO-RF model achieved the best predictive accuracy for CGMI_EMW, with a coefficient of determination (R²) of 0.751, root mean square error (RMSE) of 0.102 and mean absolute error (MAE) of 0.074, indicating superior estimation precision over CGMI_CV. Based on the optimal model PSO-RF-CGMI_EMW, the spatial distribution and statistical results of maize inversion imagery demonstrate that the simulation results align well with the experimental data, indicating a good performance of the simulation inversion. This study investigates the development of a model for monitoring maize growth stages and evaluates the effectiveness of the monitoring. The findings verify the precision and reliability of this method, providing vital insights for maize growth monitoring and field management.