Biologically inspired model for crater detection

Crater detection from panchromatic images has its unique challenges when comparing to the traditional object detection tasks. Craters are numerous, have large range of sizes and textures, and they continuously merge into image backgrounds. Using traditional feature construction methods to describe craters cannot well embody the diversified characteristics of craters. On the other hand, we are gradually revealing the secret of object recognition in the primate's visual cortex. Biologically inspired features, designed to mimic the human cortex, have achieved great performance on object detection problem. Therefore, it is time to reconsider crater detection by using biologically inspired features. In this paper, we represent crater images by utilizing the C1 units, which correspond to complex cells in the visual cortex, and pool over the S1 units by using a maximum operation to reserve only the maximum response of each local area of the S1 units. The features generated from the C1 units have the hallmarks of size invariance and location invariance. We further extract a set of improved Haar features on each C1 map which contain gradient texture information. We apply this biologically inspired based Haar feature to crater detection. Because the feature construction process requires a set of biologically inspired transformations, these features are embedded in a high dimension space. We apply a subspace learning algorithm to find the intrinsic discriminative subspace for accurate classification. Experiments on Mars impact crater dataset show the superiority of the proposed method.