Deep Learning of High-Resolution Unmanned Aerial Vehicle Imagery for Classifying Halophyte Species: A Comparative Study for Small Patches and Mixed Vegetation

Recent advances in deep learning (DL) and unmanned aerial vehicle (UAV) technologies have made it possible to monitor salt marshes more efficiently precisely. However, studies rarely compared the classification performance of DL with pixel-based method for coastal wetland monitoring using UAV data. In particular, many been conducted at landscape level; however, little is known about species discrimination very small patches mixed vegetation. We constructed a dataset based on UAV-RGB data methods five scenarios (combinations annotation type patch size) marsh Maximum likelihood, method, showed lowest overall accuracy 73%, whereas U-Net achieved over 90% all scenarios. As expected, comparison methods, approach most accurate results. Unexpectedly, there was no significant difference between two types labeling sizes this study. when comparing results detail, we confirmed that polygon-type effective mixed-vegetation than bounding-box type. Moreover, smaller size detecting vegetation patches. Our suggest combination can facilitate mapping local scale.