An Evaluation of Reflectance Calibration Methods for UAV Spectral Imagery

John Anderson

John Anderson US Army Corps of Engineers

23F

An Evaluation of Reflectance Calibration Methods for UAV Spectral Imagery

Jarrod Edwards1, John Anderson1 and Jason Woolard2

USACE Geospatial Research Lab, Corbin Field Station, Woodford, Va 22580

NOAA National Ocean Service, Corbin Field Station, Woodford, Va 22580

The popularity of micro-unmanned aerial vehicles is driving the miniaturization and capabilities of small-format imaging systems.  As these systems gain popularity, attention needs to be paid to the non-traditional way images are acquired (e.g., off nadir viewing geometries and post-processing) as well as the geometric and radiometric quality of the data itself.  We present here a study that explores a comparison of spectral calibration methods for drone-based multispectral imagery data.  The system used was the Sequoia four-band Multispectral Imager (MSI) that also features a visible camera and a down-welling sensor as part of the total instrument payload.  The spectral sensors are centered at: 550 nm, 660 nm, 735 nm and 790 nm (green, red, red-edge, and near infrared, respectively).  For comparison, two multiband orthomosaics were produced and a different spectral calibration method applied to each.  In the first calibration method, the imagery data were corrected using the PIX4D reflectance calibration procedure by applying a uniformly gray AIRINOV calibration target (imager manufacturer suggested calibration method).  A second set of data were calibrated by Empirical Line Calibration (ELC) method using concurrent ground radiometric data on specific targets in-scene.  Both scenes were analyzed for target spectral agreement with ground radiometer values, and subsequent classification accuracy.  A regression analysis demonstrated greater correlation values for the Empirical Line-calibrated data than for the AIRINOV-calibrated data.  Root mean square error (RMSE) analyses supported the regression results.  Finally, class maps were developed and compared between the ELC- calibrated (truth) data and the AIRINOV-calibrated data which resulted in an overall classification accuracy of 24% for the AIRINOV map with a considerable number of pixels associated with brighter targets unclassified.  The results have implications for the use of these types of systems and the need for standardized procedures in correcting small-format remote sensor data, particularly with the increasing ubiquity of these systems.

 

 

11:45 An Evaluation of Reflectance Calibration Methods for UAV Spectral Imagery, John Anderson

January 30 @ 11:45
11:45 — 12:00 (15′)

Granite ABC

John Anderson

Add to Google Calendar