Landscape typology analysis in Slovenia through image analytics

Classifying landscapes into types is challenging because they are continuous, but it is essential for effective protection, management, and planning, as similar processes of change occur within the same landscape types. This study aimed to classify landscape types through image analytics. Landscape photographs were embedded using the Inception v3 model within the visual programming toolbox Orange. The best classification performance was achieved with logistic regression, with an area under the curve (AUC) of 0,995 and a classification accuracy of 0,930. The second-best method was k-nearest neighbors (AUC = 0,947), while the classification tree performed the worst (AUC = 0,820). Hierarchical clustering and t-SNE results were unsatisfactory, despite the selected photographs being representative and landscape typology being simplified into five types.