| Abstract : | Fast Radio Bursts (FRBs) show a wide variety of temporal and frequency structures in their waterfall plots which could indicate different origins of the burst or emission mechanisms. CHIME and other radio telescopes have detected a large number of FRBs and the detection rate is expected to increase in the future. While morphology of FRBs is very diverse, it is not clear whether different FRBs emerge from different emission mechanisms/astrophysical channels. Rapid, automated classification of FRBs will enable multiwavelength follow-up (e.g. with ground- and space-based telescopes) and the prioritisation of rare and anomalous events. We are developing a classifier that can rapidly analyse any incoming telescope data to classify the FRBs based on their morphology. Such a classifier can also help in understanding the statistics of FRB morphology. These statistics can help support or reject various emission mechanism models for the FRB. We have created simulated datasets in the form of images (intensity as a function of time and frequency) for various observed FRB morphologies. We simulate bursts for each type while varying different burst parameters like fluence, width, scattering measure, spectral index, number of components etc. for SNR values of 10, 15, 25, 35, 50,100 to create a robust model that can classify the five broad categories we identified. We have trained a deep learning model based on CNN (Convolution Neural Network) to classify each pair of different types of FRBs. We will discuss our methodology and preliminary results on the classification. The code and example data for the framework will be made available for community usage. |
