Reproducibility of crop surface maps extracted from Unmanned Aerial Vehicle (UAV) derived digital surface maps
KAUST DepartmentWater Desalination and Reuse Research Center (WDRC)
MetadataShow full item record
AbstractCrop height measured from UAVs fitted with commercially available RGB cameras provide an affordable alternative to retrieve field scale high resolution estimates. The study presents an assessment of between flight reproducibility of Crop Surface Maps (CSM) extracted from Digital Surface Maps (DSM) generated by Structure from Motion (SfM) algorithms. Flights were conducted over a centre pivot irrigation system covered with an alfalfa crop. An important step in calculating the absolute crop height from the UAV derived DSM is determining the height of the underlying terrain. Here we use automatic thresholding techniques applied to RGB vegetation index maps to classify vegetated and soil pixels. From interpolation of classified soil pixels, a terrain map is calculated and subtracted from the DSM. The influence of three different thresholding techniques on CSMs are investigated. Median Alfalfa crop heights determined with the different thresholding methods varied from 18cm for K means thresholding to 13cm for Otsu thresholding methods. Otsu thresholding also gave the smallest range of crop heights and K means thresholding the largest. Reproducibility of median crop heights between flight surveys was 4-6cm for all thresholding techniques. For the flight conducted later in the afternoon shadowing caused soil pixels to be classified as vegetation in key locations around the domain, leading to lower crop height estimates. The range of crop heights was similar for both flights using K means thresholding (35-36cm), local minimum thresholding depended on whether raw or normalised RGB intensities were used to calculate vegetation indices (30-35cm), while Otsu thresholding had a smaller range of heights and varied most between flights (26-30cm). This study showed that crop heights from multiple survey flights are comparable, however, they were dependent on the thresholding method applied to classify soil pixels and the time of day the flight was conducted.
CitationParkes SD, McCabe MF, Al-Mashhawari SK, Rosas J (2016) Reproducibility of crop surface maps extracted from Unmanned Aerial Vehicle (UAV) derived digital surface maps . Remote Sensing for Agriculture, Ecosystems, and Hydrology XVIII. Available: http://dx.doi.org/10.1117/12.2241280.
PublisherSPIE-Intl Soc Optical Eng
Conference/Event nameRemote Sensing for Agriculture, Ecosystems, and Hydrology XVIII