Hyperspectral Camera Characterization of System Spectral Radiance Error for Spectral Identification of Reflective Objects Using Aerial Imagery

We define and implement a spectral radiative transfer calibration image processing procedure for multispectral and hyperspectral imaging systems with a global shutter. Using laboratory measurements with camera capturing scene filled with checkerboard targets, we quantify and correct relative spectral channel alignment error as a function of lateral displacement from optical center and electromagnetic wavelength. This is synonymous with measuring the combined effects of chromatic aberration and optical alignment error of an imaging system. We demonstrate that the spectral error quantified with our proposed procedure is a stable characterization of the imaging system hardware, consistent over time and when the hyperspectral imaging system is positioned on an aerial moving platform. This is applicable to airborne earth observing missions where spectral signatures are integral for detecting properties of ground cover materials such as image segmentation, mineral detection, precision agriculture, spectral linear mixture modeling, and object detection to name a few. We present a case study of using our proposed spectral calibration image processing on an aerial hyperspectral dataset for vehicle detection task, with vehicles having notably complex surface geometry and non-Lambertian reflective materials. The data after processing demonstrates clear improvement for manual labelling vehicles, providing physically meaningful at-sensor spectral radiance for each pixel within the dataset. By converting pixels from raw digital counts to physical units through camera spectral characterization, physics informed machine learning from airborne imagery datasets can be viable, enabling generalized learning from other datasets calibrated to the same physical units.