Optimizing object-based image analysis for semi-automated geomorphological mapping

Object-Based Image Analysis (OBIA) is considered a useful tool for analyzing high-resolution digital terrain data. In the past, both segmentation and classification parameters were optimized manually by trial and error. We propose a method to automatically optimize classification parameters for increasing the accuracy and efficiency of OBIA for semi-automated geomorphological mapping. We test our method by semi-automatically extracting three geomorphological ‘feature types’ (river terrace, gypsum sink holes, and fluvial incision) from a 1m Digital Terrain Model (DTM) of an alpine area in Vorarlberg, Austria. Segmentation parameters were optimized for each specific geomorphological ‘feature type’, by comparing frequency distribution matrices of training samples and automatically generated image objects. Subsequently image objects are iteratively classified with varying classification settings. The best classification scores and corresponding segmentation and classification settings are summarized in a library of feature signatures for stratified feature extraction. Our results show that through optimization, a limited number of classifiers can be used to accurately classify geomorphological features in complex terrain. This allows classification schemes to be standardized for automated and effective analysis of high-resolution terrain data. In addition, by automating mapping procedures, this research increases the efficiency of geomorphological research. Further research will include the classification of the remaining geomorphological ‘feature types’ to create a full-covered geomorphological map, and the application of the feature signature library to other areas.