Active-learning-for-crater-detection-using-digital-elevation-model-and-CNN
To improve the performance of Crater detection algorithm.
Abstract
The craters are the primary morphological structure on the moon or mars. these are halloweed out area produced by the impact of meteors. Supervised algorithms are widely used approach to detect craters from the planetary surface data. Digital Elevation Models (DEM) specifying the 3D features on the planetary surface are generated by radar scanning. DEMs and surface imagery are now commonly available for planetary surfaces, which provides new opportunities to enhance the performance of the crater detection systems. With the 3D features provided by DEMs, It’s very easy to detect craters with decent performance. However, live detection of craters is not possible as producing DEMs is itself a very time consuming process. Supervised techniques like CNN, Haar features etc. can be used for live detection of craters after using trained models. To achieve high accuracy, we must have abundant number of training samples to train the model. Finding and labelling craters in a large number of satellite imagery to generate training samples to is a very tedious task. DEMs can be deliberately used to find craters using unsupervised techniques like SVM and label them to generate training samples to apply in CNN model.