Machine learning for exoplanet detection in high-contrast spectroscopy

Abstract

Context. The advent of high-contrast imaging instruments combined with medium-resolution spectrographs allows spectral and temporal dimensions to be combined with spatial dimensions to detect and potentially characterize exoplanets with higher sensitivity. Aims. We developed a new method to effectively leverage the spectral and spatial dimensions in integral-field spectroscopy (IFS) datasets using a supervised deep-learning algorithm to improve the detection sensitivity to high-contrast exoplanets. Methods. We began by applying a data transform whereby the four-dimensional (two spatial dimensions, one spectral dimension, and one temporal dimension) IFS datasets are replaced by four-dimensional cross-correlation coefficient tensors obtained by cross-correlating our data with young gas giant spectral template spectra. Thus, the spectral dimension is replaced by a radial velocity dimension and the rest of the dimensions are retained ‘as is’. This transformed data is then used to train machine learning (ML) algorithms. We trained a 2D convolutional neural network with temporally averaged spectral cubes as input, and a convolutional long short-term memory memory network that uses the temporal data as well. We compared these two models with a purely statistical (non-ML) exoplanet detection algorithm, which we developed specifically for four-dimensional datasets, based on the concept of the standardized trajectory intensity mean (STIM) map. We tested our algorithms on simulated young gas giants inserted into a SINFONI dataset that contains no known exoplanet, and explored the sensitivity of algorithms to detect these exoplanets at contrasts ranging from 10−3 to 10−4 for different radial separations. Results. We quantify the relative sensitivity of the algorithms by using modified receiver operating characteristic curves (mROCs). We discovered that the ML algorithms produce fewer false positives and have a higher true positive rate than the STIM-based algorithm. We also show that the true positive rate of ML algorithms is less impacted by changing radial separation than the STIM-based algorithm. Finally, we show that preserving the velocity dimension of the cross-correlation coefficients in the training and inference plays an important role in ML algorithms being more sensitive to the simulated young gas giants. Conclusions. In this paper we demonstrate that ML techniques have the potential to improve the detection limits and reduce false positives for directly imaged planets in IFS datasets, after transforming the spectral dimension into a radial velocity dimension through a cross-correlation operation and that the presence of the temporal dimension does not lead to increased sensitivity.