Deep Learning for Individual Tree Detection in Urban Areas
Trees in urban environments provide lots of benefits to residents, such as improving air quality, regulating temperature, regulating urban water, etc. Models that can detect and/or segment individual trees have been developed for both forest and urban areas. Such models can be used to create maps that provide a detailed overview of all trees, their spatial distribution and for example their crown size, height or biomass. In urban contexts, these maps can support urban planners in deciding where trees need to be supported or planted.
The goal of this master thesis is to develop a model that can be used to detect and/or segment individual trees in Münster, and if possible, NRW. This model can then be used to explore how metrics such as crown size, height or biomass change over time, which can be achieved by applying the model to data from different years.
The tasks for this master thesis consist of:
-
Creating an overview of existing literature and state-of-the-art models for individual tree detection.
-
Implementing and potentially fine-tuning a model for individual tree detection
-
Generating a map of Münster (and if possible NRW) with individually labelled trees.
-
Providing a change map that shows the change in tree number or size over a period of time.