Pamplemousse:

A set of machine learning tools for Integral Field Unit Spectral Analysis

Machine learning is rapidly becoming ubiquitous in astronomy. The Pamplemousse project is a collection of tools built specifically for the analysis of spectra taken with the SITELLE instrument on the CFHT. The tools are developed by a team lead by Carter Rhea at the Université de Montréal. The core team includes Dr. Laurie Rousseau-Nepton, Dr. Simon Prunet, and Dr. Julie Hlavacek-Larrondo. The source code (and trained models) can be found at https://github.com/sitelle-signals/Pamplemousse. If you wish to use this directly with fitting spectral cubes, please consider using the analysis code Luci which has been developed specifically for the analysis of SITELLE data cubes. Luci documentation can be found at https://crhea93.github.io/LUCI/index.html. The simple-to-use installation instructions can be found here: https://github.com/crhea93/LUCI.

Please check out our examples. If you use any code here in your work, or as an inspiration, please cite us! You can find information on our pretrained networks at Network Library. These are also available on our github page at https://github.com/sitelle-signals/Pamplemousse/tree/master/PREDICTORS.

Our Projects

Click on the headers to learn about the projects using Pamplemousse!

Fig. 3 - Schematic Diagram of the Convolutional Neural Network employed in this work

The first paper explores the use of a convolutional neural network to extract the flux and velocity of underlying components. We report a standard deviation of ∼5 km/s for the velocity parameter and a standard deviation of approximately 5.5 km/s for the broadening parameter.

Fig. 5 - Artificial Neural Network created for this work.

In this work, we apply an artificial neural network to combined-filter (SN1, SN2, and SN3) SITELLE data representing typical SIGNALS large program observations. The network is designed to calculate important emission-line ratios for HII-like regions which are present in the primary SITELLE filters. Our resultsindicate that the network can potentially constrain the line ratios with greater precision than the standard line fitting technique implemented in ORCS \textbf{if the source spectral properties are well represented in the training set}. Timing analysis indicates that the network can analyze the entire cube approximately 100 times faster than the standard methods.

Fig. 6 - Line Ratio Errors for the 8 line ratios studied in this work.

In this paper (in preparation), the third of the series, we develop a convolutional neural network to classify spectra as having either a sin-gle or double line-of-sight component. This systemati cmethod will be critical for disentangling components in merger systems, HII regions, and supernova remnants.We demonstrate that the network outperforms AIC andBayesian inference model comparisons.

Fig. 8 - NGC2207-IC2163 merging system. Left: Deep SITELLE image of the N2207/IC2163 system created withORCS. This panel shows the stacked optical emission in the component galaxies. Several structures such as the spiral arms, bulges, tidal tails, merging region (green circle), and diffuse emission regions (purple circles) stand out. Right: Component map for the NGC2207/IC2163 system. White pixels correspond to double component emission. Black pixels correspond to single component emission.
...
...
...
...

Contact Us

https://github.com/sitelle-signals/Pamplemousse
Campus MIL
L'Université de Montréal, QC, CA
E: carter.rhea@umontreal.ca