As part of the PRISMA project, we aim to create a neural network for detecting transient atmospheric events caused by meteoroids entering the atmosphere. The current approach to detection is through motion detection using the open source software FRIPON / Freeture. However, this approach is inefficient as this technology is not selective and generates different events (detection) than the ones it should monitor, leading to false positives. The project proposed by N3 in collaboration with the National Institute of Astrophysics in Turin provides for the development and training of a neural network that detects in real time the entry of meteoroids into the atmosphere.