Predicting Sorghum Lodging With UAVs: A Regression-Based Approach

Lodging significantly reduces crop yields. Stalk lodging (breaking or bending of stalks) is a critical trait in sorghum cultivation [1]. Developing crop varieties with resistance to lodging is a top priority for modern breeding programs. Currently, breeders rely on manual observations to gather phenotypic measurements related to lodging, which is labor-intensive and prone to human bias.In this study, we investigated stalk lodging induced by the BOREAS machine [2] using a UAV platform. We developed a regression approach using EfficientNet-B4 to predict the percentage of lodging caused by the machine. Our results demonstrate that the UAV platform, combined with Deep learning, can accurately predict lodging percentages, significantly reducing the need for labor-intensive manual observations, and saving both time and money.