Year: 2024
Type: conference-paper
Abstract: In current broadband passive sonar systems, a human operator observes a 2D image that represents different energy levels in the bearing and time dimensions. The temporal depth of the image helps the operator to detect and identify the paths taken by different sources. However, detecting weak noise sources amid ambient sea noise and strong sources can be challenging. The purpose of this work is to increase the contrast of those images using AI techniques. The performance of AntoNet, our proposed architecture based on Unet, is evaluated for reducing ambient noise and enhancing the signal of interest. The dataset is generated by a fast sonar image simulator based on a simplified sonar beamforming model. We aim to improve pixel-wise detection through clearer 2D images with a better SNR. AntoNet leverages the temporal depth of data to produce 2D images with an enhanced SNR while having less than 1,500 parameters. Performance evaluation demonstrates AntoNet's effectiveness in single source environments, achieving higher detection performance with minimal artifacts.
Language: English
Subfield: Oceanography
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4401977917