Multi-Dimensional Underwater Point Cloud Detection Based on Deep Learning.

Area of Science:

• Marine robotics
• Underwater sensing technologies
• Environmental monitoring

Background:

• Traditional sensing methods face limitations in underwater environments due to signal attenuation and low light.
• Detecting submerged debris, such as abandoned tires, is crucial for marine ecosystem health.
• Advanced acoustic sensing offers a viable solution for underwater object detection.

Purpose of the Study:

• To adapt and apply 2D and 3D detection technologies for identifying abandoned tires underwater.
• To evaluate the effectiveness of deep learning models for underwater tire detection using acoustic data.
• To process and analyze underwater point cloud data for environmental remediation applications.

Main Methods:

• Utilized a 3D acoustic sensor (BV5000) to collect underwater point cloud data.
• Pre-processed data to remove noise and seabed interference.
• Generated 2D bird's eye view images and 3D point cloud segments for analysis.
• Employed deep learning models: Faster R-CNN and YOLOv3 for 2D detection, and PointNet and PointConv for 3D classification.

Main Results:

• Both 2D and 3D deep learning approaches demonstrated good accuracy in detecting underwater abandoned tires.
• The pre-processing steps successfully isolated tire data from background noise and seabed.
• Acoustic sensing combined with deep learning proved effective for underwater object identification.

Conclusions:

• 2D and 3D deep learning methods are effective for detecting abandoned tires using underwater acoustic point cloud data.
• The study validates the use of acoustic sensors and advanced processing for marine debris monitoring.
• This research contributes to developing technologies for underwater environmental cleanup and management.