MANAGEMENT TECHNOLOGIES FOR ARID RANGELANDS
Location: Range Management Research
Title: Image processing and classification procedures for analysis of sub-decimeter imagery acquired with an unmanned aircraft over arid rangelands
Submitted to: GIScience and Remote Sensing
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: December 1, 2010
Publication Date: March 1, 2011
Citation: Laliberte, A.S., Rango, A. 2011. Image processing and classification procedures for analysis of sub-decimeter imagery acquired with an unmanned aircraft over arid rangelands. GIScience and Remote Sensing. 48(1):4-23.
Interpretive Summary: Unmanned aircraft systems (UAS) are highly suitable for use as remote sensing platforms for collection of sub-decimeter resolution imagery due to the relatively low operating costs, ability for fast and repeated deployment, and greater flexibility than piloted aircraft. However, because UAS are commonly equipped with lightweight, low-cost digital cameras resulting in images with very high spatial but low spectral and radiometric resolution, classification procedures and field sampling have to be adapted to this imagery. Using a small UAS, we acquired 5-cm resolution true color aerial photography over arid rangelands in southern New Mexico. We tested field sampling approaches appropriate for object-based, species-level classification of this very high resolution imagery. The samples served as input for a classification tree to determine the optimal spectral, spatial, and contextual features, which were used in a combination of rule-based and nearest neighbor classifications. Results indicated that a process tree developed on a 0.49 ha plot could be applied to the remaining study area with relatively few edits. This improved efficiency, reduced processing times, and can provide an approach for classification of large image files over areas of similar vegetation. Overall classification accuracy was 78% at the species level, and 81% at the structure group level. The highest accuracies were achieved for the larger shrubs, while small shrubs and small bunchgrasses had lower accuracies, with the largest confusion occurring between litter and spectrally similar grasses. The integration of 1) appropriate field sampling, 2) determination of optimal features and classification approaches, and 3) suitable processing approaches for potentially large image mosaic files provides a roadmap for deriving quality classification products from UAS imagery, which can be used for mapping and monitoring vast and remote rangelands.
Using five centimeter resolution images acquired with an unmanned aircraft system (UAS), we developed and evaluated an image processing workflow that included the integration of resolution-appropriate field sampling, feature selection, object-based image analysis, and processing approaches for UAS image mosaics. Algorithms developed on a test plot could be applied to the remaining study area with relatively few edits. Overall classification accuracy was 78% at the species level, and 81% at the structure group level. The image processing approach provides a roadmap for deriving quality vegetation classification products from UAS imagery with very high spatial, but low spectral resolution.