Towards automatic quality assessment of tomograms of cataclysmic variable stars

Astronomy provides important challenges for computer sciences, since there are many astronomical phenomena that must be studied through computational means. One of them is cataclysmic variable stars (CV). These phenomena must be studied through indirect observation techniques, since modern instruments are not able to directly obtain information about their structure and behavior. One of such techniques, Doppler tomography, uses a search algorithm to generate an image, called tomogram that depicts the relevant structures of a cataclysmic variable star. One important drawback of this algorithm is that it lacks any criteria to decide when to stop the search. This paper proposes an approach to automatically stop the algorithm based on the quality of the tomogram. The approach is to process each tomogram with the Sobel operator and then calculate the standard deviation (SD) of the result. The SD values of all of the tomograms generated during the search are introduced into a feed-forward neural network that indicates which tomograms have the best scientific quality. The neural network training data was created with the assessment of an expert astronomer.