Machine Vision and Deep Learning for Classification of Radio SETI\n Signals

Abstract

We apply classical machine vision and machine deep learning methods to\nprototype signal classifiers for the search for extraterrestrial intelligence.\nOur novel approach uses two-dimensional spectrograms of measured and simulated\nradio signals bearing the imprint of a technological origin. The studies are\nperformed using archived narrow-band signal data captured from real-time SETI\nobservations with the Allen Telescope Array and a set of digitally simulated\nsignals designed to mimic real observed signals. By treating the 2D spectrogram\nas an image, we show that high quality parametric and non-parametric\nclassifiers based on automated visual analysis can achieve high levels of\ndiscrimination and accuracy, as well as low false-positive rates. The (real)\narchived data were subjected to numerous feature-extraction algorithms based on\nthe vertical and horizontal image moments and Huff transforms to simulate\nfeature rotation. The most successful algorithm used a two-step process where\nthe image was first filtered with a rotation, scale and shift-invariant affine\ntransform followed by a simple correlation with a previously defined set of\nlabeled prototype examples. The real data often contained multiple signals and\nsignal ghosts, so we performed our non-parametric evaluation using a simpler\nand more controlled dataset produced by simulation of complex-valued voltage\ndata with properties similar to the observed prototypes. The most successful\nnon-parametric classifier employed a wide residual (convolutional) neural\nnetwork based on pre-existing classifiers in current use for object detection\nin ordinary photographs. These results are relevant to a wide variety of\nresearch domains that already employ spectrogram analysis from time-domain\nastronomy to observations of earthquakes to animal vocalization analysis.\n