Morphological Crater Classification via Convolutional Neural Network with Application on MOLA data

The only approach for a surface age dating is the impact crater count. In order to facilitate this process, many automatic approaches have been proposed for the impact crater detection. However, the origin and the morphological features of those impact craters can influence the accurate crater count. In this article, we propose a novel approach for crater morphological classification. The developed method is based on a study of a 3D triangulated mesh of Mars' sample. We use a curvature analysis and local quantization method in combination with a convolution neural network to automatically classify impact craters in three categories: valid, secondary and degraded craters.