Deep Learning Based Mars Terrain Classification with Edge Detection for Planetary Rovers

Accurate terrain classification is critical for autonomous rover navigation during planetary exploration missions. This study presents a hybrid deep learning framework that combines Convolutional Neural Networks (CNNs) with edge detection techniques to improve the classification of Martian terrains. Feature enhancement methods including Sobel, Canny, Laplacian filters and wavelet transforms are employed to strengthen structural features in raw images. Several CNN models VGG19, MobileNet, InceptionV3 and DenseNet were trained and tested using NASA’s Mars Science Laboratory (MSL) terrain dataset and AI4mars dataset. DenseNet achieved the highest classification accuracy of 95.62%. The system is deployed on a Raspberry Pi with a USB camera for live terrain analysis and inference results are transmitted to a cloud-based server that simulates satellite communication. Integration with the Robot Operating System (ROS) enables real-time decision-making and autonomous navigation. Experimental results confirm that the proposed system significantly enhances terrain classification and offers strong potential for intelligent, mission-ready deployment in planetary rovers.