Extracting the Subhalo Mass Function from Strong Lens Images with Image Segmentation

Detecting substructure within strongly lensed images is a promising route to shed light on the nature of dark matter. However, it challenging task, which traditionally requires detailed lens modeling and source reconstruction, taking weeks analyze each system. We use machine-learning circumvent need for develop neural network both locate subhalos in an image as well determine their mass using technique segmentation. The trained with single subhalo located near Einstein ring across wide range apparent magnitudes. then able resolve masses $m\gtrsim 10^{8.5} M_{\odot}$. Training this way allows learn gravitational lensing light, remarkably, detect entire populations substructure, even locations further away from than those used training. Over magnitude, false-positive rate around three false per 100 images, coming mostly lightest detectable that signal-to-noise ratio. With good accuracy low rate, counting number pixels assigned class over multiple measurement function (SMF). When measured bins $10^9M_{\odot}$--$10^{10} M_{\odot}$ SMF slope recovered error 36% 50 improves 10% 1000 Hubble Space Telescope-like noise.