Related Experiment Video
Updated: Jun 15, 2026
09:19
Measuring the Structure, Composition, and Change of Underwater Environments with Large-area Imaging
Published on: April 18, 2025
Distortion-tolerant 3D recognition of underwater objects using neural networks
Robert Schulein1, Cuong Manh Do, Bahram Javidi
1Department of Electrical and Computer Engineering, University of Connecticut, 371 Fairfield Road, Unit 2157, Storrs, Connecticut 06269-2157, USA.
Summary
This study introduces a novel method for recognizing underwater objects using 3D integral imaging and neural networks. The system demonstrates robust object recognition despite water distortion and occlusion.
Area of Science:
- Computer Vision
- Optical Imaging
- Robotics
Background:
- Underwater object recognition is challenging due to optical distortions like scattering and occlusion.
- Neural networks excel at image recognition tasks, with prior applications in 2D and 3D holographic imaging.
- Integral imaging offers a passive 3D imaging technique.
Purpose of the Study:
- To develop and evaluate a distortion-tolerant method for underwater object recognition.
- To investigate the effectiveness of neural networks in conjunction with 3D integral imaging for this task.
- To assess the system's robustness under various challenging underwater conditions.
Main Methods:
- Utilized three-dimensional (3D) integral imaging for capturing object data.
- Developed and implemented neural network classification architecture for recognition.
- Tested the system with rotation-variable 3D objects under simulated adverse water conditions (scattering, occlusion).
Main Results:
- Successfully demonstrated distortion-tolerant recognition of 3D objects in water.
- Validated the system's robustness against variable scattering levels and physical occlusion.
- Achieved reliable recognition performance despite challenging environmental factors.
Conclusions:
- Neural networks combined with 3D integral imaging provide an effective solution for underwater object recognition.
- This approach offers a significant advancement for passive 3D imaging and recognition in disturbed aquatic environments.
- Represents the first known application of neural networks for passive 3D integral imaging of underwater objects.
