PhD - Gent | More than two weeks ago
Hyperspectral imaging into the wild
Nowadays, Unmanned Aerial Vehicles (UAVs) are often employed as a measuring or monitoring device in a wide range of remote sensing applications, for example, in precision agriculture (the need for observing the inter- and intra-variability in crops) or in infrastructure inspection (the need for detecting defects). Hyperspectral technology has been proven to be superior in that to visual (RGB) imaging for many agricultural, ecological and inspection applications.
The major challenge with UAV image data is to interpret the measured data correctly in relation to the physical environment and to other captured data such as continuous point measurement devices. As opposed to controlled lab environments, the obtained UAV hyperspectral data is susceptible to the varying incident solar radiation (even if there are no clouds) and sun-sensor viewing angles, which depends on the position of the drone and the sun (which is different during the day and for every day in the year). The latter is usually modeled by a bidirectional reflectance distribution function (BRDF). On top of that, in order to avoid saturation and thus loss of information, these varying light conditions require the sensor to have a very high dynamic range, which can be solved by capturing at different exposure times in a fixed interleaved pattern or via an auto-exposure function. All this has a profound impact on the further data analysis and interpretation.
The purpose of this PhD is to design a methodology to obtain highly trustworthy and interpretable hyperspectral data from UAV platforms regardless of the varying light conditions. This encompasses the design of sensor fusion with calibrated spectrometers measuring the incoming radiation as well as the design of novel data representations (extensions to BRDF models) and algorithms for correcting the radiance for different exposures and compensating the light variance. This research will not only have a large impact on UAV hyperspectral data quality, but also on the research in the different UAV application domains such as precision agriculture, where a reduction in inputs without compromising yield is a key ambition which on its turn leads to a more sustainable planet.
Required background: Computer Science / Data Science
Type of work: 40% algorithm design, 30% experimental, 20% modeling, 10% literature
Supervisor: Hiep Luong
Co-supervisor: Wilfried Philips
Daily advisor: Andy Lambrechts
The reference code for this position is 2022-112. Mention this reference code on your application form.