Valley Classification using Convolutional Neural Network and a Geomorphons Map

Geomorphological classification serves as a valuable tool for comprehending the origin and evolution of landscapes, as well as for making informed decisions regarding environmental hazard mitigation and sustainable development. However, the process of classifying landforms is typically time-consuming and necessitates specialized expertise. This research article presents a novel approach that utilizes a convolutional neural network (CNN) to classify valleys. The methodology involves employing an initial classification generated by an unsupervised geomorphons classifier as input data, which is subsequently refined using human-generated ground truth. In contrast with the original geomorphons method, this novel method enhances spatial coherence by effectively connecting pixels classified as valleys. The results show that the proposed CNN-based method significantly enhances the accuracy of the classification. We are confident our approach is competitive according to the Total Operating Characteristic (TOC) curve as well as classification metrics.