Premium
Counting of Pennisetum alopecuroides at heading stage in a grazed pasture using images from an unmanned aerial vehicle
Author(s) -
Yuba Norio,
Kawamura Kensuke,
Yasuda Taisuke,
Lim Jihyun,
Yoshitoshi Rena,
Kurokawa Yuzo,
Maeda Teruo
Publication year - 2020
Publication title -
grassland science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.388
H-Index - 19
eISSN - 1744-697X
pISSN - 1744-6961
DOI - 10.1111/grs.12277
Subject(s) - altitude (triangle) , remote sensing , heading (navigation) , ground sample distance , environmental science , rgb color model , pixel , image resolution , mathematics , geography , artificial intelligence , computer science , geodesy , geometry
This study focused on the fact that Pennisetum alopecuroides ears are black that represents a black dots in image from sky and developed a simple algorithm to count the number of P. alopecuroides plants from unmanned aerial vehicle (UAV) images. Ortho‐mosaicked red‐green‐blue (RGB) images from different flight altitudes (28, 56 and 82 m) were obtained in the heading growth stage. To remove small noisy pixels, various sized median filters (3 × 3–21 × 21) were tested. While the relationship between field and image counts was positive for most images obtained from 56 and 82 m altitudes, images in the 28‐m flight altitude (the highest spatial resolution) with small and moderate‐sized median filters (3 × 3–9 × 9) were overestimated. Moreover, higher resolution images from lower altitudes of UAV images decrease the operating efficiency (areas covered per 10 min). Overall, the best accuracy (80.3%) and lowest mean absolute error (MAE = 12.93) were obtained for the 56‐m flight altitude (ground sampling distance [GSD] = 1.82 cm) with a 7 × 7 filter size.