Abstract:Underwater acoustic environment estimation is a challenging but important task for remote sensing scenarios. Current estimation methods require high signal strength and a solution to the fragile echo labeling problem to be effective. In previous publications, we proposed a general deep learning-based method for two-dimensional environment estimation which outperformed the state-of-the-art, both in simulation and in real-life experimental settings. A limitation of this method was that some prior information had to be provided by the user on the number and locations of the reflective boundaries, and that its neural networks had to be re-trained accordingly for different environments. Utilizing more advanced neural network and time delay estimation techniques, the proposed improved method no longer requires prior knowledge the number of boundaries or their locations, and is able to estimate two-dimensional environments with one or two boundaries. Future work will extend the proposed method to more boundaries and larger-scale environments.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to estimate the position and number of reflective boundaries in reverberant environments (such as underwater and indoor acoustic environments). Specifically, the paper aims to overcome the limitations of existing methods in the following aspects: 1. **Poor performance at low signal - to - noise ratio (SNR)**: Existing boundary estimation methods usually perform poorly in low - SNR environments. 2. **Echo - label problem**: Existing methods need to solve complex combinatorial optimization problems to distinguish different reflective boundaries, especially in the presence of missing or false echoes. 3. **Dependence on prior knowledge**: Previous deep - learning methods (such as Neuro - COTANS) require users to provide prior information about the number and position of boundaries, and the neural network needs to be retrained for different environments. To solve these problems, the paper proposes a new method based on the U - Net architecture - U - COTANS, which has the following improvements: - **No prior information required**: U - COTANS can directly estimate the number and position of boundaries from COTANS images without requiring users to provide prior information. - **Better generalization ability**: By using the image segmentation method, U - COTANS can handle a wider range of environments, not just boundaries within a specific area. - **Directly estimate the number of boundaries**: U - COTANS is able to directly estimate the number of boundaries in the environment, which is an ability that existing methods do not possess. ### Specific problems and solutions #### 1. Performance improvement at low SNR The paper points out that in low - SNR environments, time - delay estimation may have a large "global error". For this reason, U - COTANS adopts a more advanced Space - Alternating Generalized Expectation - Maximization (SAGE) algorithm for time - delay estimation, thus improving the performance under low - SNR conditions. #### 2. Echo - label problem The traditional echo - label problem is caused by the overlap between multiple - path signals. U - COTANS effectively solves this problem by generating COTANS images and using U - Net for image segmentation. The curve intersections in the COTANS images correspond to the real boundary positions, and U - Net can accurately estimate the boundaries by learning the characteristics of these intersections. #### 3. Dependence on prior knowledge Neuro - COTANS requires users to provide prior information about the number and position of boundaries, and the neural network needs to be retrained for different environments. However, U - COTANS, by introducing the image segmentation method, enables the model to directly learn the position and number of boundaries from COTANS images without requiring users to provide additional prior information. ### Summary The main goal of this paper is to develop a more robust and general - purpose boundary estimation method that can accurately estimate the position and number of reflective boundaries under various SNR conditions without relying on prior knowledge or frequently retraining the model. By introducing the U - Net architecture and image segmentation technology, U - COTANS has achieved significant improvements in performance and generalization ability.

Estimating the Number and Locations of Boundaries in Reverberant Environments with Deep Learning

A deep learning method for reflective boundary estimation

Strategic Deployment in the Deep: Principled Underwater Sensor Placement Optimization with Three-Dimensional Acoustic Map.

Acoustic Field Visualization and Source Localization Via Physics-Informed Learning of Sparse Data with Adaptive Sampling

Spatiotemporal Tracking of Ocean Current Field with Distributed Acoustic Sensor Network

Towards Robust Data-Driven Underwater Acoustic Localization: A Deep CNN Solution with Performance Guarantees for Model Mismatch

Continual Learning of Range-Dependent Transmission Loss for Underwater Acoustic using Conditional Convolutional Neural Net

Research on Underwater Acoustic Environmental Perception Based on Sparse Channel Estimation

An Underwater Wideband Sound Source Localization Method Based on Light Neural Network Structure

Learning to localize sounds in a highly reverberant environment: Machine-learning tracking of dolphin whistle-like sounds in a pool

Deep neural network for learning wave scattering and interference of underwater acoustics

A Deep Neural Network Approach to Acoustic Source Localization in a Shallow Water Tank Experiment

Deep learning for source localization and seabed classification under dynamic oceanographic environments using explosive sound signals

Sonar Dereverberation via Independent Component Analysis and Deep Learning

Training Deep Networks with Only Synthetic Data: Deep-learning-based Near-Offset Reconstruction for (Closed-Loop) Surface-Related Multiple Estimation on Shallow-Water Field Data

Predictive 3D Sonar Mapping of Underwater Environments via Object-specific Bayesian Inference

Deep Learning Based Underwater Acoustic Target Recognition: Introduce a Recent Temporal 2D Modeling Method

Design of Deep Learning Acoustic Sonar Receiver with Temporal/ Spatial Underwater Channel Feature Extraction Capability

Machine learning of acoustic propagation models for sound aware autonomous systems

Deep Learning-Based Estimation of Reverberant Environment for Audio Data Augmentation

Data-Based Acoustic Far Scattered Field Prediction in Underwater Environment