DEEP TRANSFER LEARNING ACROSS TARGETS AND SENSORS WITH SAS DATA
DP WILLIAMS · 2023
This study uses high-frequency (i.e., center frequency greater than 100 kHz) sonar data from two synthetic aperture sonar (SAS) systems, nominally named sensor α and sensor β.A large database of scene-level SAS images from sensor α was curated and divided into two disjoint sets, one for training and one for test purposes.The raw sonar data was reconstructed into imagery with the Advanced Synthetic Aperture Sonar Imagining eNgine (ASASIN) software 7 .Another database of scene-level SAS images from sensor β was also curated and divided into two disjoint sets, one for training and one for test purposes.The raw sonar data was again reconstructed into imagery with the ASASIN software.One example scene from each of the sensors is shown in Fig. 1.