The advancement of unmanned aerial vehicles (UAVs) offers precise and accurate spectral and spatial information about crops and plays a pivotal role in precision agriculture. This study used UAVs, geographic information systems (GIS), and deep learning technology to monitor corn growth performance across different management practices. Two experimental corn fields were divided into four plots to evaluate the effects of varying corn management practices (i.e., seeding schedule, planting depth, and fertilization method) on corn growth performance. RGB and MicaSense multispectral cameras were mounted on UAVs to collect corn field images. YOLOv5 was investigated for counting corn plants. Plant height, Normalized Difference Vegetation Index (NDVI), Normalized Difference Red-Edge Index (NDRE), plant density, and plant volume were mapped based on UAV images. Additionally, the Otsu thresholding method was evaluated as an automatic method for separating plant height, NDVI, and NDRE values from the background. YOLOv5 and Otsu thresholding were efficient and accurate for automatically counting corn plants and extracting corn plant heights as well as VIs, respectively. The emergence rates of corn seeds were 40%, 33%, 41%, and 62% in plots A, B, C, and D, respectively. Variations in corn field management practices significantly affected the emergence rate, with fertilizer application close to seeds emerging as the optimal practice for achieving higher emergence rates across experimental plots. This study used deep learning and UAV to provide precise information and valuable insights into corn field practices, which can help farmers optimize corn cultivation. The techniques applied in this study could be extrapolated to improve cultivation processes for other crops.
Read full abstract