A Spatial Arrangement Preservation-Based Stitching Method via Geographic Coordinates of UAV for Farmland Remote Sensing Image

High-quality panoramic images captured by unmanned aerial vehicle (UAV) for farmland remote sensing play a crucial role in monitoring crop growth. However, the stitching algorithms that rely on local image matching often face challenges due to high similarity in color and texture of farmland remote sensing images. In this work, for the first time, we explore the spatial arrangement reflected by these geographic coordinates of each captured image during the UAV flight to obtain the high-quality panoramic images. We theoretically propose a novel algorithm for preserving the spatial arrangement structure based on the triangular similarity transformation, which ensures that the spatial arrangement of all images in the stitched image remains invariant to that before stitching. Moreover, we incorporate the proposed spatial arrangement preservation (SAP) module as an energy term into the global similarity prior (GSP) stitching model and thus obtain the proposed SAP-GSP stitching method. In a wheat-dominated experimental field, we collected UAV remote sensing images at various flight altitudes and multiple growth stages over a three-year period, which formed the image stitching dataset (Wheat-UAV). We evaluated the proposed SAP-GSP algorithm from both subjective and objective perspectives on Wheat-UAV dataset. The experimental results conclusively demonstrate that the SAP-GSP algorithm outperforms representative stitching algorithms in terms of preserving spatial arrangement and achieving natural-looking global stitched images. The proposed SAP-GSP offers a novel and effective technique for robustly capturing large-scale UAV remote sensing panoramic images in agricultural fields.