Deep Learning–Based Vegetation Feature Detection for UAV-Derived Geomorphic Change Monitoring

This study presents a data-driven framework that integrates deep learning and UAV-based remote sensing for geomorphic change detection. A Mask R-CNN model is trained to identify specific plant species from high-resolution orthoimagery, treating vegetation as spatially persistent surface features. The detected plant locations are georeferenced and represented as coordinate-based point datasets, enabling quantitative analysis of surface displacement through multi-temporal comparisons. The framework is demonstrated in the Guanziling region of southern Taiwan, a tectonically active area influenced by the Chukou Fault. Results indicate that temporal changes in the spatial distribution of detected vegetation effectively capture subtle surface deformation patterns that are difficult to observe using conventional image-based approaches. Compared with LiDAR surveys, the proposed method significantly reduces data acquisition costs while preserving essential spatial information for geomorphic analysis. Beyond monitoring applications, the resulting vegetation-based spatial datasets provide new opportunities for integration with physics-based geomorphic and geotechnical models, supporting data-driven model calibration, validation, and predictive assessment. Overall, this study highlights the potential of deep learning–enabled feature detection to advance scalable, cost-effective, and interpretable geomorphic monitoring in complex natural environments.