UAV orthoimaging mosaic of the coastal zone by combining foreground segmentation and improved A-Star algorithm

Unmanned aerial vehicle (UAV) orthoimages of coastal zones can provide fundamental data support for maritime activities such as coastal zone environmental monitoring and other applied ocean research. However, due to the narrow field of view, a single image cannot comprehensively cover wide-ranging remote sensing needs. Therefore, UAV orthoimages require mosaicking. Coastal areas present challenges due to their complex and dynamically changing water characteristics, leading to issues such as voids, misalignments, and ghosting in current UAV orthoimage mosaicking algorithms. To address these challenges, we propose a method that combines foreground segmentation and an improved A-Star algorithm specifically tailored for coastal environments. Initially, image registration under trajectory constraints and the construction of an initial stitching network based on Voronoi diagrams are performed. Water extraction from the initial airborne RGB images is achieved using an improved UE-Net model and edge information, producing clear-edged foreground segmentation results. Finally, an optimization model for seamlines is established based on the foreground segmentation results and an improved A-Star algorithm to refine and adjust initial seamlines. Experimental validation using coastal zone images from Dachangshan Island, Changhai, Dalian, China, demonstrates the effectiveness of the proposed algorithm. Visual outcomes and structural similarity indices are compared with those obtained using Agisoft Metashape 1.8.0 software, the scale invariant feature transform-random sample consensus algorithm, the A-Star algorithm, and the Dijkstra algorithm, showing varying degrees of improvement. Overall, the proposed algorithm offers an effective solution for coastal zone remote sensing and other applied ocean research applications.