Technical advances in the field of remote sensing have made it possible to create a large number of datasets with very high spectral, temporal or spatial resolution, however, in the field of...Show moreTechnical advances in the field of remote sensing have made it possible to create a large number of datasets with very high spectral, temporal or spatial resolution, however, in the field of archaeology, the evaluation of this data is still largely a manual undertaking. The issue with manual interpretation is that human interpreters are increasingly having difficulty coping with the sheer amount of data while in some cases, the human eye is not capable of processing the full range of information contained in these datasets. It is for this reason that (semi-)automatic classification workflows need to be developed in order to aid human interpreters in their image classification tasks. This thesis is concerned with the development of a Geographic Object-Based Image Analysis (GeOBIA) workflow for classifying LiDAR visualisations containing heterogeneous and linear objects. The study area that this workflow is applied to is the terraced landscape of the Lower Engadine, Switzerland, where the complex and steep terrain contains multiple agricultural terraces, irrigation/drainage ditches, roads and more. The workflow makes use of only FOSS (Free and Open Source Software) applications in order to ensure full transparency, accessibility and reproducibility of the classification results. For this purpose, a number of FOSS and proprietary software was tested in order to determine the user friendliness, suitability and effectiveness of each of the options. In order to develop the final workflow, a number of studies regarding the suitability of different LiDAR visualisations as well as training data input options and smoothing filters were carried out. The final workflow makes use of an unfiltered slope visualisation, consists of six steps with an optional seventh step, and is capable of producing classification results that hold up against manual mapping results of the terrace edges that were used as a benchmark. Finally, in order to assess whether the classification results generated by the workflow are useful to a human interpreter, a user study was carried out. 13 out of the 14 users stated that the classification results were helpful to them and because the workflow takes no longer than 5-10 minutes to carry out, it can be said that this workflow is capable of producing a useful classification of the study area with minimal time and effort.Show less
