Back to Volume
Paper: The Application of Machine Learning Methods in the CADC's Advanced Search Service
Volume: 522, Astronomical Data Analysis Software and Systems XXVII
Page: 389
Authors: Teimoorinia, H.; Kavelaars, J.; Gwyn, S.; Durand, D.; Willott, C.
Abstract: We present a Machine-Learning Network that provides quantitative information on the content of astronomical images to users of the CADC's Advanced Search service. Using pattern recognition methods, which we introduce, the users can find images that contain the desired levels of structure and source content. Our approach utilizes an artificial neural network model to classify astronomical images. For example, in this way, users can exclude 'garbage' images from the result of a search. This method can easily be used as a secondary filtering step in primary query. And it is very effective in a search, for example, when we have different images for a field (or object) and we want to select the image with the lowest sky fluctuations. Our method uses the Sextractor software package as a feature extractor and converts all the features from an image to a special two-dimensional distribution. Compressing the globally measured Sextractor parameters to a two-dimensional space significantly compresses the volume of information that needs to be stored for each image. These compressed data vectors can then be used by the network in the pattern recognition step. Our ANN system will provide users with the ability to find images with desired and interesting features within the massive number of images and spectra stored within the CADC.
Back to Volume