Lunar impact craters identification and age estimation with Chang'E data by deep and transfer learning

DeepCraters is a pipeline for training a convolutional neuralnetwork (CNN) to identify impact craters on the Moon and training a dual-channel impact crater age estimation model to classify the impact craters identified before.

The DeepCraters pipeline trains a impact craters identified model and age estimatiom model using data derived from CE-1 and CE-2 DOM and DEM image data and catalogue of craters. The code is divided into two parts. The first trains a impact craters identified model with R-FCN [the details can be found in the subfile craters_detection/]; The second trains a dual-channel impact crater classification model using craters with constrained ages and craters detected without age.

• I. http://moon.bao.ac.cn/ [CE-1 and CE-2 DOM and DEM data can be obtained from this website]
• II. https://planetarynames.wr.usgs.gov/Page/MOON/target [recognized lunar impact craters regulated by the IAU can be obtained from this website]
• III. https://www.lpi.usra.edu/lunar/surface/ [lunar craters with constrained ages aggregated by the LPI can be obtained from this website]
• IV. https://astrogeology.usgs.gov/search/map/Moon/Geology/Lunar_Geologic_GIS_Renovation_March2013 [Lunar 1:5M Geologic Map, 38 stratigraphic information of craters extracting from, can be obtained from this website]

NOTE: All the craters with constrained ages in 'II' are contained in 'III'.

The craters used for 'ctaters detection' can be find in /craters_detection/data_list/, and the craters with constrained ages used for 'age estimatiom' can be find in /age_estimation/data_list/.

Each part of DeepCraters has the corresponding script:

• part 1 (craters_detection):
  RFCN_ROOT/experiments/scripts/moon_rfcn_end2end.sh for build and train the detection model,
  RFCN_ROOT/tools/rfcn_test_Moon_Detect.py to generate a crater atlas of study area.
  The more details can be find in README for Craters Detection.

• part 2 (age_estimation):
  train_moon_age_estimation.py for build and train the age estimation model,
  pred_dete_moon_age_estimation.py to predict the age of all craters detected in part 1.
  The more details can be find in README for Age Estimation.

We recommend copying these scripts into a new working directory (and appending this repo to your Python path) instead of modifying them in the repo.

DeepCraters is free software made available under the MIT License. For details see the LICENSE.md file.