A rapid field crop data collection method for complexity cropping patterns using UAV and YOLOv3

Abstract

Crop type mapping using remote sensing is critical for global agricultural monitoring and food security. However, the complexity of crop planting patterns and spatial heterogeneity pose significant challenges to field data collection, thereby limiting the accuracy of remotely sensed crop mapping. This study proposed a new approach for rapidly collecting field crop data by integrating unmanned aerial vehicle (UAV) images with the YOLOv3 (You Only Look Once version 3) algorithm. The impacts of UAV flight altitude and the number of training samples on the accuracy of crop identification models were investigated using peanut, soybean, and maize as examples. The results showed that the average F1-score for crop type detection accuracy reached 0.91 when utilizing UAV images captured at an altitude of 20 m. In addition, a positive correlation was observed between identification accuracy and the number of training samples. The model developed in this study can rapidly and automatically identify crop types from UAV images, which significantly improves the survey efficiency and provides an innovative solution for acquiring field crop data in large areas.