MULTI-MODAL ACCURACY COMPARISON FOR DEVELOPMENT OF MULTI-SENSOR-BASED SHIP DETECTION ALGORITHM

Abstract

In recent years, as satellites have become miniaturized, the satellite development cycle has been shortened, and a large number of various satellites are being operated due to the development of microsatellite constellations. Therefore, in accordance with the trend of the new space era, combining multiple satellite data, rather than using a single satellite, can increase the time resolution and be useful for monitoring and detecting objects. In this study, before developing a ship detection algorithm using multiple satellites, the accuracy of multi-modal methods suitable for ship detection in dual-pol SAR images was compared. To ensure diversity in the training data, all training images were composed of Gray and Parula colormaps(in Matlab), and six multi-modals were configured to compare detection accuracy. As a results, it was confirmed that the detection accuracy was the highest in the 1) model that fused the VH image of the Gray colormap and the VV image of the Parula colormap and 2) model that fused the VH and VV images of the Parula colormap.