Presentation #141.03 in the session Intergalactic Medium, QSO Absorption Line Systems — iPoster Session.
Quasar absorption line analysis is critical for studying gas and dust components and their physical and chemical properties as well as processes of galaxies in the early universe. Ca II absorption lines, as one of the dustiest absorbers and located at lower redshifts, are especially valuable when studying low redshift galaxies. However, currently available datasets of quasars with Ca II absorption lines are extremely limited due to the difficulty in detecting them with traditional methods. In this work, we developed an accurate and efficient approach to search for Ca II absorption lines using deep learning. To generate a large enough training set, we created a vast amount of simulated data by inserting artificial Ca II absorption lines onto original quasar spectra from the Sloan Digital Sky Survey (SDSS) whilst adopting an existing Ca II catalog as the test set. We designed a convolutional neural network for the classification task and tuned the model with the simulated data in the training and validation sets. The resulting model achieved an accuracy of 96% on the real data in the test set, which is almost 15 times higher than traditional methods. Our solution also runs thousands of times faster than traditional methods. We applied the trained neural network to quasar spectra from SDSS’s DR12 and DR7 and discovered over 600 new Ca II absorbers, making it the largest such catalog ever detected. We also confirmed over 400 known Ca II absorbers identified in previous work. New results and the design of our neural network will be presented.