Related Experiment Video
Updated: Dec 30, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
952
Three-dimensional multi-source localization of underwater objects using convolutional neural networks for artificial
Ben J Wolf1, Jos van de Wolfshaar1, Sietse M van Netten1
1Bernoulli Institute of Mathematics, Computer Science and Artificial Intelligence, Faculty of Science and Engineering, University of Groningen, Groningen, The Netherlands.
Journal of the Royal Society, Interface
|January 23, 2020
Summary
This study presents a novel method for tracking multiple underwater objects using simulated hydrodynamic flow. The system effectively localizes moving sources in 3D space by processing fluid velocity data with advanced algorithms.
Area of Science:
- Biomimetics and Bio-inspired Engineering
- Hydrodynamics and Fluid Dynamics
- Signal Processing and Machine Learning
Background:
- Fish utilize their lateral line system to detect and localize underwater objects by sensing hydrodynamic disturbances.
- Existing methods for underwater object localization often face challenges with multiple simultaneous sources and complex 3D environments.
Purpose of the Study:
- To develop and evaluate a signal processing method for simultaneously localizing multiple moving underwater objects in a 3D volume.
- To simulate hydrodynamic flow and leverage fish lateral line organ principles for source localization.
- To investigate the efficacy of convolutional neural networks and iterative algorithms for this task.
Main Methods:
- A two-step localization process was employed: first, a convolutional neural network (CNN) estimated source presence probability in a 2D image.
- Second, an automated iterative 3D-aware algorithm determined the precise 3D position of each source.
- The system utilized sampled fluid velocity data from two parallel lateral lines, exploring various CNN architectures and input presentation methods.
Main Results:
- Optimized CNN architectures, including multi-level amplified inputs and merged convolutional streams, significantly improved imaging performance.
- The combined system demonstrated adequate 3D localization capabilities for multiple underwater sources.
- The proposed method effectively processes complex hydrodynamic signals for source detection.
Conclusions:
- The bio-inspired sensory system, integrating hydrodynamic simulation and advanced signal processing, shows promise for robust multi-object localization in 3D underwater environments.
- This research validates the potential of combining neural networks with iterative algorithms for complex underwater sensing tasks.
- Further development could lead to enhanced underwater surveillance and navigation systems.
