Automated snow extent mapping based on orthophoto images from unmanned aerial vehicles
PBN-AR
Instytucja
Wydział Nauk o Ziemi i Kształtowania Środowiska (Uniwersytet Wrocławski)
Informacje podstawowe
Główny język publikacji
angielski
Czasopismo
PURE AND APPLIED GEOPHYSICS (25pkt w roku publikacji)
ISSN
0033-4553
EISSN
1420-9136
Wydawca
BIRKHAUSER VERLAG AG
DOI
URL
Rok publikacji
2018
Numer zeszytu
9
Strony od-do
3285–3302
Numer tomu
175
Identyfikator DOI
Liczba arkuszy
Streszczenia
Język
angielski
Treść
The paper presents the application of the k-means clustering in the process of automated snow extent mapping using orthophoto images generated using the Structure-from-Motion (SfM) algorithm from oblique aerial photographs taken by unmanned aerial vehicle (UAV). A simple classification approach has been implemented to discriminate between snow-free and snow-covered terrain. The procedure uses the k-means clustering and classifies orthophoto images based on the three-dimensional space of red–green–blue (RGB) or near-infrared–red–green (NIRRG) or near-infrared–green–blue (NIRGB) bands. To test the method, several field experiments have been carried out, both in situations when snow cover was continuous and when it was patchy. The experiments have been conducted using three fixed-wing UAVs (swinglet CAM by senseFly, eBee by senseFly, and Birdie by FlyTech UAV) on 10/04/2015, 23/03/2016, and 16/03/2017 within three test sites in the Izerskie Mountains in southwestern Poland. The resulting snow extent maps, produced automatically using the classification method, have been validated against real snow extents delineated through a visual analysis and interpretation offered by human analysts. For the simplest classification setup, which assumes two classes in the k-means clustering, the extent of snow patches was estimated accurately, with areal underestimation of 4.6% (RGB) and overestimation of 5.5% (NIRGB). For continuous snow cover with sparse discontinuities at places where trees or bushes protruded from snow, the agreement between automatically produced snow extent maps and observations was better, i.e. 1.5% (underestimation with RGB) and 0.7–0.9% (overestimation, either with RGB or with NIRRG). Shadows on snow were found to be mainly responsible for the misclassification.
Inne
System-identifier
PX-5ba8e617d5de36c2ae2914d3
CrossrefMetadata from Crossref logo
Cytowania
Liczba prac cytujących tę pracę
Brak danych
Referencje
Liczba prac cytowanych przez tę pracę
Brak danych