Physics-anchored masked autoencoder for efficient sonogram simulation

We present a technique for efficiently creating sonogram such as LOFAR and DEMON, by combining physical acoustic modeling and data-driven method of masked autoencoder. This technique involves two stages. First, in the given the ocean environment, the physical model accurately calculates a restricted portion of entire sonogram image. Next, the data-driven model based on masked autoencoder creates an image of remaining region. By employing an iterative decoding structure and controlling the weight of the physical loss term, the reconstruction accuracy of masked autoencoder is improved. The results are compared with those of a pure physical model, a hybrid model with original masked autoencoder, and a hybrid model with traditional PCA technique. We will discuss the practicality of the proposed technique in terms of accuracy and calculation performance, as well as the applicablity of real data using Deepship dataset. [Work supported by the Korea Research Institute for Defense Technology Planning and Advancement (KRIT) grant funded by the Korea government (Defense Acquisition Program Administration (DAPA)) (No. KRIT-CT-22-052, Physics-guided Intelligent Sonar Signal Detection Research Laboratory).]