Datasets
Standard Dataset
Benchmark dataset with annotations for Boundary Delineator for Martian Crater Instances
- Citation Author(s):
-
DanyangLiu
Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences - Submitted by:
- Danyang Liu
- Last updated:
- Mon, 07/08/2024 - 15:59
- DOI:
- 10.21227/b21g-ya15
- Data Format:
- License:
13 Views- Categories:
- Keywords:
Abstract
Detection of impact craters on the surface of Mars is a critical component in the study of Martian geomorphology and the evolution of the planet. As one of the most distinguishable geomorphic units on the Martian surface, accurate determination of the boundaries of impact craters provides valuable information in mapping and research efforts. The topography on Mars is more complex than that of the moon, making detection of real impact crater boundaries a challenging task. Instead of detecting the real boundaries, the majority of current techniques concentrate on replacing impact craters with circles or points. Real boundaries are more challenging to identify than simple circles. To solve the above challenges, a Boundary Delineator for Martian Crater Instances (BDMCI) using fusion data was proposed. First, optical images, DEM, and slope of processed DEM were used to combine the fusion data. Second, a sample dataset for the real impact crater boundaries was created, and sample regions were chosen using geospatial prior knowledge and the optimization strategy for the proposed BDMCI. Thirdly, the model was trained for various scales using deep learning. To fix the fractures at the junctions of patches, a variety of post-processing methods were devised. The proposed BDMCI was also used to expand the catalog of Martian impact craters in the large-scale region between 65°S and 65°N. The results of this study provide valuable insights into the geomorphology of Mars and demonstrate the potential of deep learning algorithms in planetary science research.
The proposed BDMCIused fusion data integrated from the optical image, DEM, and slope of processed DEM. The optical image is the Mars global mosaic, obtained from THEMIS Day IR in June 2010. The Thermal Emission Imaging System (THEMIS) is an instrument on board the Mars Odyssey spacecraft. The resolution is 100 meters per pixel. The scale of the optical image is 65°S-65°N, 0°E-360°E. The blended DEM data is derived from the Mars Orbiter Laser Altimeter (MOLA), an instrument aboard NASA’s Mars Global Surveyor (MGS), and the High-Resolution Stereo Camera (HRSC), an instrument aboard the European Space Agency’s Mars Express (MEX) spacecraft. The resolution is 200 meters per pixel. The scale of DEM is global. Due to the scale of the optical image, we clipped DEM to the same scale. The projections of optical image and DEM are both SimpleCylindrical_Mars. The slope of processed DEM is a slope data after the filled DEM reduced original DEM. Impact craters can be thought of the depression caused by impacting. Virtually, if it rains from the sky, the impact crater will be filled by the rain. The difference in elevation after rain could indicate the boundaries of impact craters. Based on this reason, the process of DEM was designed.The slope of “Filled grid minus Original DEM” highlighted the impact craters and reduced the stripe noise of data, compared with slope of slope. Also, the resolution, scale, and projection of slope of processed DEM are the same as those of DEM. Taking fusion data as a base map, the accurate boundaries of impact craters were marked by polygons according to the comprehensive assessment. In this sample dataset, the impact craters whose diameters larger than 2 km were preserved. The size of the grid is 1024 pixels. These grids were randomly divided into a training set, an evaluation set and a test set in proportion. As a result, the training set had 316 grids. The evaluation set had 90 grids. The test set had 45 grids. In addition, these three sets did not overlap each other. The fusion data and sample dataset were clipped by the divided grids. In the form of MS COCO (Microsoft Common Objects in Context) dataset, images and annotations were fed to the models.
