Martian Swiss Cheese detection and volume estimation using shape from shading in very high-resolution imagery

Abstract

This study introduces a methodology for the automatic detection of Swiss Cheese-like features on Mars using high-resolution remote sensing images acquired by the Mars Orbiter Camera (MOC) and the High-Resolution Imaging Science Experiment (HiRISE). The proposed approach combines digital image processing (DIP) and mathematical morphology (MM) to identify and delineate these complex, irregular surface patterns. A key innovation in this work is the application of Shape from Shading (SfS) to reconstruct the three-dimensional characteristics of depressions in the Martian surface, specifically to extract the volume of the Swiss Cheese features. The SfS technique addresses the challenge of obtaining 3D representations from 2D remote sensing images, where varying lighting conditions and complex topography introduce significant reconstruction difficulties. The 3D reconstruction process involved integrating surface normals calculated using finite differences. While the Lambertian reflectance model was employed for its simplicity and compatibility with available data, it has limitations, particularly under Martian atmospheric conditions. The thin Martian atmosphere was not explicitly modeled, which may introduce biases, especially in areas with significant shadowing. Despite these limitations, the reconstructed 3D features qualitatively align with the expected topographic characteristics of the Swiss Cheese features. By estimating the area and volume of the detected features, the application of SfS provides valuable insights into their evolution over time. An additional analysis was conducted to estimate internal height variations within the features. By utilizing solar illumination data, the methodology assessed the average height and height differentials between shadowed and illuminated regions of the Swiss cheese. This analysis revealed internal topographic variations, highlighting the complex and dynamic nature of the Swiss Cheese features. The method was tested on 22 images, showing high performance with an average recall of 93%, precision of 79%, and an F1-Score of 85%. Temporal analysis of HiRISE images from 2007 and 2010 revealed notable increases in the area and volume of Swiss cheese, driven by sublimation-induced erosion. Although the accuracy of the estimated area and volume could be improved with complementary datasets such as field surveys or LiDAR measurements, these initial results provide a meaningful basis for understanding the dynamic nature of Martian surface features. The methodology has proven effective for generating 3D visualizations of Swiss Cheese features, though some intrinsic characteristics may be lost in the reconstruction process. The results should be interpreted with caution, as they represent a general 3D visualization rather than a precise reconstruction. Despite these challenges, the methodology offers a promising approach to studying Martian polar features, providing a foundation for future research that could refine the 3D reconstruction process and enhance the reliability of the findings.