Rock Segmentation of Real Martian Scenes Using Dual Attention Mechanism-Based U-Net

Segmentation of Martian rocks is essential for future space exploration and scientific research. In this study, we propose a dual attention mechanism-based U-Net for the segmentation of Real Martian scenes. The dual attention mechanism captures both spatial and channel-wise dependencies to improve segmentation performance. The proposed method is evaluated on a dataset of Real Martian scenes, and the results show that it outperforms state-of-the-art approaches. The dual attention mechanism-based U-Net segmentation technique successfully detected the Real Martian scenes of the rock segmentation using Mars images with an accuracy of 96.38%, sensitivity of 95.36%, specificity of 95.39%, and Jaccard index of 0.86. The proposed method is helpful for rovers to find a route without any obstacles.