Wind-induced response of rapeseed seedling stage and lodging prediction based on UAV imagery and machine learning methods

Farmers commonly enhance rapeseed grain yield by increasing nitrogen fertilizer application and planting density, but this raises lodging susceptibility. Lodging in rapeseed not only substantially diminishes yield and quality but also hampers mechanized harvesting. Thus, timely and accurate prediction of rapeseed lodging resistance, along with targeted field management, is imperative for enhanced productivity. However, current research on timely and accurate prediction of rapeseed lodging resistance remains limited. This study employs unmanned aerial vehicle (UAV) imagery in conjunction with machine learning techniques. UAVs equipped with cameras and downward airflow stimulation are utilized to capture wind-induced responses in rapeseed leaves and extract relevant parameters. Wind-induced response characteristics of rapeseed under different cultivation conditions are analyzed, the relationship between rapeseed vegetation indices and intrinsic properties is explored, and the obtained parameters are subjected to principal component analysis. Using the maturity stage rapeseed lodging index as the output, a predictive model for early-stage lodging is established, comparing the Genetic Algorithm-optimized Backpropagation Neural Network (GA-BP), Particle Swarm Optimization-optimized Backpropagation Neural Network (PSO-BP), and Cuckoo Search-optimized Support Vector Machine (CS-SVM) models. The results reveal a significant correlation between Rapeseed seedling-stage wind-induced response characteristics, certain vegetation indices, and lodging index. Three lodging index prediction models are created using the first four principal components from the analysis, yielding promising outcomes for all three periods (5-leaf stage, 10-leaf stage, and 10 days after the 10-leaf stage) and overall predictions. Among these models, the PSO-BP model exhibits superior performance in predicting rapeseed lodging index (R2 = 0.67, RMSE = 0.464, MAPE = 12.15). Therefore, leveraging wind-induced response characteristics and vegetation indices during the early growth stage enables a certain level of prediction for rapeseed lodging resistance in the mature stage. This study's findings contribute theoretical and technical support to the intelligent and precise management of large-scale rapeseed production.