Dense extreme inception network-based edge detection with deep reinforcement learning for object localization in an...

Area of Science:

• Robotics and Automation
• Computer Vision
• Marine Technology

Background:

• The underwater environment presents unique challenges for observation and exploration due to factors like pressure, light attenuation, and temperature.
• Accurate object detection (OD) is crucial for understanding marine ecosystems and managing underwater infrastructure.
• Existing imaging technologies and algorithms struggle with the complexities of underwater imagery.

Purpose of the Study:

• To develop an advanced technique for robust underwater object detection and classification.
• To address the limitations of current methods in accurately identifying objects in challenging aquatic conditions.
• To introduce a novel approach combining edge detection and deep reinforcement learning for enhanced underwater situational awareness.

Main Methods:

• A Dense Extreme Inception Network-based Edge Detection with Deep Reinforcement Learning for Object Localisation (DEINED-DRLOL) technique was proposed.
• The Dense Extreme Inception Network for Edge Detection (DexiNed) was utilized for edge map prediction.
• Object detection was performed using the YOLOv5 method, followed by Q-Reinforcement Learning (QRL) for classification.

Main Results:

• The DEINED-DRLOL technique demonstrated effective edge detection and object classification in underwater environments.
• Experimental results on an underwater OD dataset showed superior performance compared to existing models.
• The proposed method achieved a high accuracy of 92.67% in underwater object detection tasks.

Conclusions:

• The DEINED-DRLOL technique offers a significant advancement in underwater object detection capabilities.
• This approach enhances the ability to explore and monitor marine environments more effectively.
• The study highlights the potential of integrating advanced deep learning techniques for complex underwater imaging challenges.