Gaussian Mixture Model and Self-Organizing Map Neural-Network-Based Coverage for Target Search in Curve-Shape Area

Area of Science:

• Robotics and Automation
• Artificial Intelligence
• Search and Surveillance

Background:

• Multi-UAV systems are increasingly used for complex search tasks.
• Efficiently covering curve-shaped areas with detection rewards and constraints is challenging.
• Prior target probability mapping is crucial for effective search strategies.

Purpose of the Study:

• To develop an optimized search strategy for multiple UAVs in curve-shaped areas.
• To maximize cumulative detection rewards while ensuring maneuverability and obstacle avoidance.
• To propose a novel algorithm integrating Gaussian Mixture Models and Self-Organizing Maps for this problem.

Main Methods:

• Approximating prior target probability maps using a 1-D Gaussian Mixture Model (GMM).
• Utilizing Self-Organizing Map (SOM) neural networks for coverage search and segment allocation.
• Employing a modified Dubins guidance vector field for path planning with obstacle avoidance.

Main Results:

• The GMM-SOM approach effectively extracts high-value curve segments.
• Prioritization and allocation of segments to UAVs are balanced considering multiple factors.
• Optimized UAV paths are generated, respecting maneuverability and obstacle constraints.
• Performance was validated using a coastline map scenario.

Conclusions:

• The proposed GMM-SOM method provides an effective solution for multi-UAV target search in curve-shaped regions.
• The integration of GMM and SOM enhances search efficiency and reward maximization.
• The method demonstrates robust performance in complex environments, such as coastline mapping.