Deep learning model inspired by lateral line system for underwater object detection

Area of Science:

• Fluid dynamics
• Robotics
• Artificial Intelligence

Background:

• Aquatic organisms utilize lateral line systems for hydrodynamic imaging.
• Sensing flow information is crucial for navigation and object detection in fluid environments.

Purpose of the Study:

• To develop a deep learning model for object localization using flow data.
• To investigate the efficacy of flow velocity and pressure measurements for detecting objects.
• To optimize sensor arrays for efficient and accurate object localization.

Main Methods:

• Numerical simulations of a hydrofoil navigating around cylinders in a uniform flow.
• Acquisition of flow velocity and pressure data from a moving sensor array.
• Construction and evaluation of various neural network models, including a long short-term memory network.
• Application of feature selection techniques for sensor optimization.

Main Results:

• A long short-term memory network model demonstrated superior performance in object localization.
• Sensor optimization using feature selection significantly reduced the number of required sensors.
• The optimized model achieved high accuracy in predicting the positions of the hydrofoil and surrounding objects with reduced sensor count.

Conclusions:

• Deep learning models, particularly LSTMs, are effective for object localization using hydrodynamic data.
• Sensor optimization is a viable strategy to improve the efficiency of flow-based sensing systems.
• This approach offers a promising method for autonomous navigation and environmental perception in fluidic environments.