Quantitative Analysis of the Relationship Between Maize Seedling Phenotypes and Soil Moisture Monitored by UAV RGB Imagery

Traditional plant phenotyping methods are typically constrained by several key limitations: insufficient segmentation accuracy, high computational costs, lack of multi-scale analysis capabilities, and ineffective modeling of complex nonlinear relationships. This study utilizes RGB imagery captured by Unmanned Aerial Vehicles (UAV), integrating plant phenotypic traits with soil moisture data, to quantitatively analyze the relationship between maize seedling phenotypes and soil moisture. Key morphological features of maize leaves—including connectivity, length, area, and shape characteristics—were transformed into matrix data using image segmentation, edge detection, and the Whale Optimization Algorithm (WOA). A quantitative relationship model between maize seedling phenotypes and soil moisture was established via regression analysis. To enhance segmentation accuracy further, we introduced the LSKNet module into YOLOv8, which optimizes leaf segmentation performance significantly under varying moisture stress conditions through multi-scale feature fusion and selective kernel convolution. The regression analysis results indicate high predictive accuracy under optimal moisture conditions (moisture content between 0.647 and 0.813), with an R2 of 0.869 and an RMSE of 0.0256. However, under drought conditions (moisture content between 0.246 and 0.354), prediction accuracy declines, with R2 decreasing to 0.851 and RMSE increasing to 0.0272. This study generates drought severity heatmaps using imagery from UAV, providing an efficient visualization tool for assessing maize plant health status. These heatmaps offer valuable references for early drought detection and precision irrigation decision-making, significantly advancing crop health management and the development of precision agriculture.