| Abstract : | The non-axisymmetric features in disc galaxies including bars and spiral arms are fundamentally important in driving their secular evolution. Yet another promiscuous non-axisymmetric feature in the galactic disc is the large-scale asymmetry in the distribution of stars and gas, commonly referred to as lopsidedness. However, their formation and evolution are not very well understood. Quantitatively, lopsidedness is the normalised amplitude of the non-zero m=1 mode in the Fourier decomposition of the surface brightness distribution of the galaxy. About 30 % of the galactic disc shows global lopsidedness. In this work, we aim to augment their relatively sparse sample size by new identifications from the updated catalogs from the more recent surveys, using machine learning methods. To implement the same, we train the publicly-available ALEXNET, a convolutional neural network model for binary classification of galaxies into lopsided and non-lopsided galaxies. For training the neural network, we selected 50 galaxies images from the Sloan Digital Sky Survey Data Release 12 from Bournaud et al., 2005. After deprojecting the galaxy images to face-on view, we obtained the value of for each galaxy. Based upon the above-calculated value of and the threshold well-adopted in the literature, we labelled the galaxies as lopsided and not lopsided. Our trained model achieved an accuracy of almost 90 % on the validation set. We aim to further improve our testing accuracy by expanding the training set for our model, allowing the model to learn more efficiently. |
