Orchard Meadow Trees: Tree Detection Using Deep Learning in ArcGIS Pro

Abstract

‘Orchard meadows’ refers to the combination of extensively managed fruit trees in combination with fields and pastures. In many regions, among others in Germany, Austria and Switzerland, they are a landscape-defining element and of particular ecological, economic and social importance. However, the numbers of orchard meadows and fruit trees have been decreasing for quite some time. Current and detailed data that allow for the identification of suitable countermeasures to maintain this cultural landscape element are often missing. Such data can be obtained through deep learning. Various deep learning frameworks can now be used in the context of ArcGIS Pro. But what exactly does the use of deep learning involve, in the context of ArcGIS Pro, to get an insight into the stocks of orchard meadow trees? What are the challenges? Initial analyses were carried out using selected areas in Franconian Switzerland (Northern Bavaria) as an example. The results confirm the potential of the approach, but also that training data, model and output data must be refined.