DeepLabV3+/Efficientnet Hybrid Network-Based Scene Area Judgment for the Mars Unmanned Vehicle System

Area of Science:

• Robotics
• Artificial Intelligence
• Planetary Science

Background:

• Mars exploration missions face challenges due to the planet's complex and hazardous environment.
• Traditional unmanned ground vehicles struggle with efficient navigation and decision-making in Martian terrain.

Purpose of the Study:

• To develop an advanced scene area judgment system for Mars unmanned vehicles.
• To enhance the safety and efficiency of Mars exploration missions through intelligent navigation.

Main Methods:

• A hybrid deep learning network combining DeepLabV3+ and Efficientnet was proposed.
• DeepLabV3+ was utilized for feature extraction from Martian imagery.
• Efficientnet processed extracted features to categorize scene areas into safe, reportable, and dangerous zones.

Main Results:

• The DeepLabV3+/Efficientnet hybrid network achieved a high accuracy of 99.84% in scene area judgment.
• Performance significantly surpassed the standalone Efficientnet network, with an approximate 18% accuracy improvement.
• The system successfully enabled Mars unmanned vehicles to perform distinct actions (pass, report, send) based on categorized areas.

Conclusions:

• The DeepLabV3+/Efficientnet hybrid network is effective for real-time scene area judgment in Mars exploration.
• This AI-driven approach enhances the safety and operational capabilities of Mars unmanned vehicles.
• The developed system contributes to more efficient and secure robotic exploration of Mars.