Automated Characterization of Yardangs Using Deep Convolutional Neural Networks

The morphological characteristics of yardangs are the direct evidence that reveals wind and fluvial erosion for lacustrine sediments in arid areas. These features can be critical indicators reconstructing local directions environment conditions. Thus, fast accurate extraction is key to studying their regional distribution evolution process. However, existing automated methods characterize limited generalization may only feasible specific types certain Deep learning methods, which superior representation learning, provide potential solutions mapping with complex variable features. In this study, we apply Mask region-based convolutional neural networks (Mask R-CNN) automatically delineate classify using very high spatial resolution images from Google Earth. yardang field Qaidam Basin, northwestern China selected conduct experiments method yields mean average precisions 0.869 0.671 intersection union (IoU) thresholds 0.5 0.75, respectively. manual validation results on additional study sites show an overall detection accuracy 74%, while more than 90% detected correctly classified delineated. We then conclude R-CNN a robust model multi-scale various allows research evolutionary aspects aeolian landform.