Postdoctoral Research Associate University of California, Davis
Disclosure: Disclosure information not submitted.
Aerial multi- and hyper-spectral imaging has recently drawn substantial attention for high-throughput plant phenotyping. Radiometric calibration of aerial spectral images is a key step before post-processing to assure the repeatability of a proposed analysis pipeline over various flight missions, dates, weather conditions, and imagers. An appropriate radiometric calibration process has two main steps: (i) converting raw images to radiance (Wm-2sr-1nm-1) to account for sensor-dependent factors such as spatial and spectral inconsistency in detectors of the camera (e.g., gain, offset, and quantum efficiency), (ii) converting radiance images to reflectance to account for variation in intensity of incident light over time. In this study, we used a multispectral camera (Micasense RedEdge) and a hyperspectral camera (Resonon PIKA L) to examine the importance of radiometric calibration for phenotyping applications. For conversion of values in raw multispectral images to radiance, we developed a Python script to batch process the images using the information embedded in the header files of images. Alternatively, for similar conversion in hyperspectral images, we used a vendor-supplied calibration file. For reflectance conversion, we used downwelling sensor data that includes the incoming spectral irradiance on top of the UAV. To refine the downwelling sensor data, an image of a reference panel with known reflectivity was captured before and after the flight mission. The preliminary results showed that the vignetting and distortion issues can degrade pixel values up to 25%. In addition, the quantitative results demonstrated the significance of radiometric calibration in reduction of the uncertainty caused by inconsistence factors, and subsequently, assure the repeatability of a proposed analysis pipeline.