engleski

Crater Detection Algorithms: A Survey of the First Decade of Intensive Research

The volume of acquired data increases significantly with each new lunar and planetary mission. Therefore, manual extraction of all the relevant information present in the images is no longer possible. Impact craters are frequent features on almost all planetary surfaces and provide insight into geological history. Hence, it is not surprising that several research groups are working on achieving increasingly robust and reliable crater detection algorithms (CDAs). However, the most advanced image-analysis/object-recognition still does not offer an answer on how to create a CDA that is as robust as the scientific community would desire. The selection of datasets considerably influences the architecture of CDAs. Therefore, in the first step of this survey, CDAs are classified on the basis of whether they utilize optical images, topography data, or any other type of images such as radar and gravimetric images. Most of CDAs use optical images, because most available images are of this type. A large number of CDAs use topography data, because: (1) detection is much easier than from optical images ; and (2) it is straightforward to achieve automated measurements of different craters’ features. The selection of methods for CDAs is an important research question. Therefore, in the second step of this survey, CDAs are classified with respect to the most commonly used methodologies: (1) edge detectors ; (2) Hough transform ; and (3) machine learning. Of particular challenge to CDAs is their capability to process global datasets. The issues are the amount of data which need to be processed as well as inherent overall heterogeneity of images. Only a few CDAs actually demonstrated the capability of processing global optical or topographic images. Therefore, in the third step of this survey, CDAs are classified with respect to ability to process global datasets. Those CDAs that are able to process global datasets are analyzed in more detail. For the comparison of CDAs, their evaluation is of the particular importance. Therefore, in the forth step of this survey, CDAs are classified with respect to evaluation method and referenced catalogue: (1) F-ROC (Free-response Receiver Operating Characteristics), A-ROC (Approximation-of Receiver Operating Characteristics) – the most appropriate alternative to ROC ; and (2) publicly available catalogues. The CDAs have already been used intensively for cataloguing of Martian, Lunar and Phobos craters. Their usage and contributions in this practical application therefore will be elaborated in more detail in the fifth section of this survey. In summary, the primary aims of this survey are: (1) to provide a review of CDAs ; and (2) to describe usage of existing CDAs, primarily for detection of craters, their cataloguing, and automated measurements of their properties.

Crater Detection Algorithms, Image Analysis, Image Processing

nije evidentirano