DeepMoon Supplemental Materials

This database contains supplemental materials to the journal article, "Lunar Crater Identification via Deep Learning", which trained a convolutional neural network (convnet) to automate the identification of lunar craters.  The database is meant to enable users to reproduce our work, or apply our model and data to new applications.  For more information about the project please see the associated journal article.  For the code, DeepMoon, and instructions on how to use the files in this database, see the GitHub link below. For a brief description of each file in this database, see the file list below. 

Code
--------
DeepMoon is available at https://github.com/silburt/DeepMoon

Files
-------

dev_craters.hdf5 - Pandas HDFStore of crater locations and sizes for images in the validation dataset.

dev_images.hdf5 - Input DEM images and output targets of the validation dataset.  Also includes each image's longitude/latitude bounds, and the pixel bounds of the global DEM regions cropped to make each image.

LunarLROLrocKaguya_118mperpix.png - LRO LOLA and Kaguya Terrain Camera DEM Merge, downsampled to 118 m/pixel and 8 bits/pixel.  The original file can be found at: https://astrogeology.usgs.gov/search/map/Moon/LRO/LOLA/Lunar_LRO_LrocKaguya_DEMmerge_60N60S_512ppd.

model_keras1.2.2.h5 - Keras model weights for the DeepMoon CNN, compatible with Keras version 1.2.2.

model_keras2.h5 - Keras model weights for the DeepMoon CNN, compatible with Keras versions >= 2.0.

post-processed_sample_images.zip - Contains a set of sample images from the test dataset with the Moon DEM image, new identified craters, CNN target predictions, and ground-truth. The new craters from these images were used to estimate the post-processed false positive rate. See Instructions.txt in .zip file for more details.

post-processed_test_craters.npy - numpy file containing post-processed craters identified by our pipeline on the test set. Each crater entry is arranged as a tuple: (longitude, latitude, radii), where longitude and latitude are in degrees, and radius is in kilometres. 

test_craters.hdf5 - Pandas HDFStore of crater locations and sizes for images in the test dataset.

test_images.hdf5 - Input DEM images and output targets of the test dataset.  Also includes each image's longitude/latitude bounds, and the pixel bounds of the global DEM regions cropped to make each image.

train_craters.hdf5 - Pandas HDFStore of crater locations and sizes for images in the training dataset.

train_images.hdf5 - Input DEM images and output targets of the training dataset.  Also includes each image's longitude/latitude bounds, and the pixel bounds of the global DEM regions cropped to make each image.