First results for an image processing workflow for hyperspatial imagery acquired with a low-cost unmanned aerial vehicle (UAV).

Authors: LALIBERTE, ANDREA - NEW MEXICO STATE UNIV; Rango, Albert; JENKINS, VINCE - SECURAPLANE TECHNOLOGIES; ROANHORSE, ABIGAIL - UNIVERSITY OF ARIZONA

Submitted to: Journal of Applied Remote Sensing (JARS)
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/3/2009
Publication Date: N/A
Citation: N/A

Interpretive Summary: Remotely sensed imagery plays an important role in rangeland monitoring and assessment. However, in many cases, satellite imagery and aerial photography do not have the required resolution to map and analyze fine-scale patterns of vegetation and soil. Imagery acquired with unmanned aerial vehicles (UAVs) has the needed sub-decimeter resolution, and this imagery fills the gap between ground-based observation and remotely sensed imagery from aerial or satellite sensors. However, because UAV imagery is often acquired at low altitudes and with consumer grade cameras, the distortion and variation in the viewing geometry coupled with lack of image overlap can present problems with orthorectification and mosaicking, which is a required step before vegetation and soil maps can be created. Here, we present the results of a test project of acquiring, orthorectifying, and classifying imagery obtained with a small UAV. The aircraft flew at 150 m above ground and obtained overlapping images over rangeland vegetation in southern New Mexico. After orthorectification and mosaicking, we used an object-based image analysis approach, which was well suited for the very high-resolution imagery. Producer’s, user’s and overall classification accuracy was in the mid to high ninety percent range. The relatively low cost of the aircraft and its components, and the ability of producing very high-resolution maps allowed for mapping vegetation types and structure in ways not feasible with piloted aircraft, and offered a landscape-scale view that could be compared to ground-based measurements. Improvements for future missions include increasing geolocation accuracy, improving the UAV’s autonomous flight performance, and reducing image distortion effects. The study demonstrates the feasiblity of using low-cost UAVs for rangeland monitoring and the potential to be applied by land management agencies to map public lands rapidly at relatively low cost.

Technical Abstract: Very high-resolution images from unmanned aerial vehicles (UAVs) have great potential for use in rangeland monitoring and assessment, because the imagery fills the gap between ground-based observations and remotely sensed imagery from aerial or satellite sensors. However, because UAV imagery is often acquired at low altitudes and with consumer grade cameras, the distortion and variation in the viewing geometry coupled with lack of image overlap can present problems with orthorectification and mosaicking. In this paper, we present the results of a test project of acquiring, orthorectifying, and classifying imagery obtained with a small UAV. The aircraft flew at 150 m above ground and obtained overlapping images over rangeland vegetation in southern New Mexico. After orthorectification and mosaicking, we used an object-based classification approach, which was well suited for the very high-resolution imagery. Producer’s, user’s and overall classification accuracy was in the mid to high ninety percent range. The relatively low cost of the aircraft and its components, and the ability of producing very high-resolution maps allow for mapping vegetation types and structure in ways not feasible with piloted aircraft. Improvements for future missions include increasing geolocation accuracy, improving the UAV’s autonomous flight performance, and reducing image distortion effects.