Evaluation of eCognition Developer and Orfeo ToolBox Performances for Segmenting Agrophotovoltaic Systems from Sentinel-2 Images

A deep energy transformation will be tried out across the globe in the next decades in order to detect potential green and renewable sources able to replace fossil fuels. Among the various alternatives, photovoltaic technology, recognized as sustainable, clean, and environmentally friendly essence, is considered one of the most relevant solutions. To date, the Remotely Piloted Aircraft System has been largely used to inspect solar parks albeit the treatment of very high-resolution satellite images through object-based models may be a valid option. In this work, the potentialities of two segmentation approaches (multi-resolution and mean-shift algorithms, implemented in eCognition Developer and Orfeo Toolbox software, respectively) in extracting photovoltaic panels from Sentinel-2 time series were explored and compared. Such techniques were tested in Montalto di Castro in Viterbo (Italy). Multi-resolution algorithm was applied by varying scale and shape parameters between 20 and 100 and 0.1 and 0.5, respectively. Conversely, the mean-shift approach was used by considering the default values of spatial radius and range radius. Their segmentation outcomes were compared on the base of i) minimum Euclidean Distance 2 (ED2), calculated in AssesSeg environment, ii) segmentation polygons statistics and areas value, and, lastly, iii) their performance in terms of processing time, versatility, ability of handling heavy data, and cost. ECognition Developer demonstrated a better performance in segmenting Sentinel 2-images for extracting PV systems in terms of segmentation parameters management and outcomes interpretation ability.