DM-Free Curvelet Based Denoising for Astronomical Single Pulse Detection

Rotating radio transients (RRATs) are sporadically emitting pulsars which are detected only through single pulse search. Detecting these single pulses in RRATs observation with high detection accuracy is a challenge due to the background noise. It is better to conduct the single pulse detection directly on the raw time-frequency observation than on the de-dispersed data, because de-dispersion process takes very intensive computation. In this paper, we propose to accomplish this idea by treating two-dimensional (2D) time-frequency data as images and develop a curvelet based denoising approach after studying the characteristics of the RRATs pulses and the noise. The denoising approach estimates the range of curvature (orientations) and width (scales) that describe the RRATs pulses and reconstructs cleaner images from the selected orientations and scales. The proposed denoising approach does not require prior knowledge of exact dispersion measures (DM) value. In addition, a framework of detecting the single pulses from the time-frequency data, named HOG-SVM, is also proposed to further evaluate the curvelet based denoising approach. Compared with the other four denoising approaches, the proposed curvelet based method leads to better detection results, with detection accuracy being increased to 98.7% by HOG-SVM.