BRDF effect in UAV remote sensing and empirical improvement methods: A case study of cotton root rot identification

Abstract

<b><sc>Abstract.</sc></b> <b>Unmanned aerial vehicle (UAV) remote sensing has been drastically increasing in agricultural studies in the past a few years. Image mosaicking is an essential procedure for UAV remote sensing. In the mosaicking processing, numerous geo-referenced images are stitched together to form an integrated orthomosaic image. Bi-directional reflectance distribution function (BRDF) is an effect that can lead to darker corners or edges in each image and cause major problems in the orthomosaicking process. The result can be dark stripes along the UAV flight path in the orthomosaic, making it hard to derive low-noise information from the mosaic. Correcting this effect can be done either before or after mosaicking. In this study, we conducted a case study to identify cotton-root-rot infected areas in a cotton field with UAV data that clearly suffered from the BRDF effect. Preliminary results indicate that the disease-infected area was often misclassified, in large part because of the BRDF effect. We proposed and have tested empirical image-processing methods based on a pre-constructed image mosaic. These methods are able to greatly reduce the negative impact of BRDF in UAV mosaics and eventually significantly improve the classification result. BRDF correction for each individual image prior to mosaicking, which is preferable from a theoretical standpoint, is under development.</b>