Pattern Recognition Using SVM for the Classification of the Size and Distance of Trans-Neptunian Objects Detected by Serendipitous Stellar Occultations

Abstract

Abstract We present a new pipeline based on the Support Vector Machine algorithm to confirm the detection and perform classification of small solar system objects by serendipitous stellar occultations. This pipeline is designed to analyze light curves and to identify the occultation events and the classification of the occulting bodies according to their size, typically from a fraction to a few kilometers, and their distance from the Sun, typically a few tens of astronomical units. The input light curves for this pipeline were obtained from the event simulator for the Trans-Neptunian Automated Occultation Survey (TAOS II). We explore parameters affecting occultation light curves such as spectral type, apparent magnitude and finite angular size of the occulted star, angle from opposition, and readout cadence for the observations; also we assumed a Poisson noise distribution as expected from the TAOS II project. We find that occultation events, especially by trans-Neptunian objects with diameters ≥2 km are detected with 99.99%, 99.53%, and 86% efficiency for stars with a visual apparent magnitude of 12, 14, and 16, respectively at 0.05 s of exposure time. In terms of size and distance classification the overall accuracy is 94%. However, for smaller objects the confirmation and classification depends mostly upon the signal-to-noise ratio.