Plot-Level Maize Early Stage Stand Counting and Spacing Detection Using Advanced Deep Learning Algorithms Based on UAV Imagery

Abstract

Phenotyping is one of the most important processes in modern breeding, especially for maize, which is an important crop for food, feeds, and industrial uses. Breeders invest considerable time in identifying genotypes with high productivity and stress tolerance. Plant spacing plays a critical role in determining the yield of crops in production settings to provide useful management information. In this study, we propose an automated solution using unmanned aerial vehicle (UAV) imagery and deep learning algorithms to provide accurate stand counting and plant-level spacing variabilities (PSV) in order to facilitate the breeders’ decision making. A high-resolution UAV was used to train three deep learning models, namely, YOLOv5, YOLOX, and YOLOR, for both maize stand counting and PSV detection. The results indicate that after optimizing the non-maximum suppression (NMS) intersection of union (IoU) threshold, YOLOv5 obtained the best stand counting accuracy, with a coefficient of determination (R2) of 0.936 and mean absolute error (MAE) of 1.958. Furthermore, the YOLOX model subsequently achieved an F1-score value of 0.896 for PSV detection. This study shows the promising accuracy and reliability of processed UAV imagery for automating stand counting and spacing evaluation and its potential to be implemented further into real-time breeding decision making.

Keywords

• maize stand counting
• plant spacing variability detection
• unmanned aerial vehicle
• deep learning