Abstract: In marine applications, large water areas lead to more difficulties compared to land on the corresponding point matching, which restricts the UAV remote sensing image registration precision and convergence speed. This paper proposes an improved algorithm for the marine UAV remote sensing applications. First, remove water using principal component analysis (PCA) and water threshold method, and obtain the block image without water. Then, feature points extraction and the corresponding points matching of these block images were performed using the affine scale invariant feature transform algorithm (Affine-Sift, ASIFT). The experiment results show that the improved algorithm can increase 30%~50% points, the accuracy increases by about 20%, and time is not increased. This method is suitable for the sequence of UAV remote sensing image registration on the marine, and it can provide effective technical support for the remote sensing monitoring of island and coastline.

Key words: UAV, remote sensing, marine application, island monitoring, registration of image sequences