Related Experiment Video
Updated: Feb 26, 2026
05:57
Long-term Video Tracking of Cohoused Aquatic Animals: A Case Study of the Daily Locomotor Activity of the Norway Lobster Nephrops norvegicus
Published on: April 8, 2019
7.3K
Performance of neural networks for localizing moving objects with an artificial lateral line
Luuk H Boulogne1, Ben J Wolf, Marco A Wiering
1Institute of Artificial Intelligence and Cognitive Engineering, University of Groningen, 9700 AK Groningen, Netherlands.
Bioinspiration & Biomimetics
|July 15, 2017
Summary
Neural networks can mimic fish's lateral line system to detect object location using flow sensors. Different network architectures show varying performance and noise robustness, with more sensors improving accuracy.
Area of Science:
- Biomimicry
- Robotics
- Signal Processing
Background:
- Fish utilize their mechanoreceptive lateral line organ to sense water flow and detect nearby objects.
- This biological system comprises an array of flow detectors along the fish's body.
Purpose of the Study:
- To investigate the use of neural networks for localizing objects based on simulated flow sensor data.
- To compare the performance, applicability, and noise robustness of various neural network architectures.
Main Methods:
- Simulated excitation patterns from artificial flow velocity sensors were used as input.
- Different neural network architectures, including Extreme Learning Machine, Echo State Networks, and Multilayer Perceptron, were trained and tested.
- Performance was evaluated based on localization accuracy and robustness to input noise.
Main Results:
- Extreme Learning Machine achieved the best performance (0.4% mean Euclidean error) under high signal-to-noise conditions (46 dB).
- Echo State Networks improved performance under lower signal-to-noise conditions, while Multilayer Perceptron demonstrated the highest noise robustness.
- Localization accuracy decreased when objects were close to the array or at its sides; more sensors enhanced performance and robustness.
Conclusions:
- Neural networks can effectively extract object location from flow sensor data, inspired by fish lateral line systems.
- The choice of neural network architecture impacts performance and noise robustness, with specific architectures suited for different conditions.
- Increasing sensor array density improves localization capabilities and resilience to noise, offering potential for improved artificial sensing systems.
