Improving Geomorphological Classification via Binary Image Processing

Abstract

Landform classification is the basis for understanding and describing the processes and evolution of landscape. This process usually requires elevation information from different sources, expertise and time. Automatic geomorphological classification, via the geomorphons algorithm, supports expert classification by using local ternary patterns for labeling landform elements, significantly reducing the computation time. Nevertheless, it presents issues such as a noisy output, valleys that are not classified as continuous forms, valleys that are classified as peaks at low altitude, flat zones inside the valley that are not classified as a part of it, and other similar issues. In this proposal, we tackle the mentioned issues for valley classification by binarizing the geomorphons output and applying it binary-image operators. The proposal's performance is measured by using binary classification metrics and expert-made groundtruth images. The results show that the accuracy, balanced accuracy, and F1 metrics are greater than those delivered by the geomorphons classifier for all the instances in the testing data.