A Method for Assessing the Impact of Altitude on Aerial Imagery for Crop and Weed Detection

Crop and weed management is essential for maintaining agricultural productivity and sustainability. Unmanned Aerial Vehicles (UAVs) have become valuable remote sensing tools for crop monitoring due to their ability to capture high-resolution aerial imagery. The altitude at which a UAV operates plays a critical role in balancing the trade-offs between image spatial resolution, coverage, flight time, and monitoring accuracy. However, this key factor is often overlooked by previous work, resulting in suboptimal UAV-based approaches. In this paper, we propose a model to estimate the impact of altitude in the classification accuracy of crop and weed detection strategies. Specifically, we propose an efficient method for simulating UAV imaging at different altitudes leveraging nearest-neighbor interpolation and image augmentation techniques. Using this approach, we generate new datasets with simulated images across 11 altitudes, based on a limited dataset with only two altitudes. A YOLO11s model is then trained and tested on the augmented datasets for each altitude, revealing a linear relationship between weed detection performance and altitude and a quadratic relation between the number of required images, flight length, and flight altitude. The combination of these analyses provides valuable insights into the trade-offs for optimizing UAV-based aerial imagery collection.