The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLII-2
https://doi.org/10.5194/isprs-archives-XLII-2-479-2018
https://doi.org/10.5194/isprs-archives-XLII-2-479-2018
30 May 2018
 | 30 May 2018

ASTEROID (21) LUTETIA: SEMI-AUTOMATIC IMPACT CRATERS DETECTION AND CLASSIFICATION

M. Jenerowicz and M. Banaszkiewicz

Keywords: Asteroid Lutetia, Image processing, Patter recognition, Crater detection

Abstract. The need to develop an automated method, independent of lighting and surface conditions, for the identification and measurement of impact craters, as well as the creation of a reliable and efficient tool, has become a justification of our studies. This paper presents a methodology for the detection of impact craters based on their spectral and spatial features. The analysis aims at evaluation of the algorithm capabilities to determinate the spatial parameters of impact craters presented in a time series. In this way, time-consuming visual interpretation of images would be reduced to the special cases. The developed algorithm is tested on a set of OSIRIS high resolution images of asteroid Lutetia surface which is characterized by varied landforms and the abundance of craters created by collisions with smaller bodies of the solar system.The proposed methodology consists of three main steps: characterisation of objects of interest on limited set of data, semi-automatic extraction of impact craters performed for total set of data by applying the Mathematical Morphology image processing (Serra, 1988, Soille, 2003), and finally, creating libraries of spatial and spectral parameters for extracted impact craters, i.e. the coordinates of the crater center, semi-major and semi-minor axis, shadow length and cross-section. The overall accuracy of the proposed method is 98 %, the Kappa coefficient is 0.84, the correlation coefficient is ∼ 0.80, the omission error 24.11 %, the commission error 3.45 %. The obtained results show that methods based on Mathematical Morphology operators are effective also with a limited number of data and low-contrast images.