A hybrid Bi-LSTM and RBM approach for advanced underwater object detection.

Area of Science:

• Marine Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Deep-sea resource development necessitates efficient underwater exploration.
• High-stress underwater environments pose significant challenges for autonomous operations.
• Accurate underwater object detection (UOD) is crucial for marine robotics and resource management.

Purpose of the Study:

• To develop and evaluate a hybrid model for enhanced Underwater Object Detection (UOD).
• To improve the efficiency and accuracy of autonomous underwater exploration systems.
• To address the challenges of UOD in complex, high-stress marine environments.

Main Methods:

• A hybrid model combining Bi-directional Long Short-Term Memory (Bi-LSTM) and Restricted Boltzmann Machine (RBM) was proposed.
• Bi-LSTM was utilized for capturing long-term dependencies and bidirectional sequence processing.
• RBMs were employed for effective hierarchical and abstract feature learning.

Main Results:

• The BiLSTM-RBM model demonstrated superior performance on brackish and URPC 2020 datasets.
• High accuracies were achieved, including 98.5% for big fish detection in the brackish dataset.
• The model successfully identified star fish with 98% accuracy in the URPC dataset.

Conclusions:

• The BiLSTM-RBM model is highly suitable for robust Underwater Object Detection (UOD).
• This hybrid approach offers a significant advancement for autonomous underwater exploration and deep-sea resource development.
• The model effectively captures complex patterns, mitigates the vanishing gradient problem, and handles variable-length sequences.