Fast ROI Triggering with Autoencoders in Optical TPCs

This paper presents a novel approach for real-time data selection in optical Time Projection Chambers (TPCs), a crucial technology for rare-event searches. The core innovation lies in using an unsupervised, reconstruction-based anomaly detection strategy with convolutional autoencoders trained on pedestal images. This method allows for efficient identification of particle-induced structures and extraction of Regions of Interest (ROIs), significantly reducing the data volume while preserving signal integrity. The study's focus on the impact of training objective design and its demonstration of high signal retention and area reduction are particularly noteworthy. The approach is detector-agnostic and provides a transparent baseline for online data reduction.