DNN-based Signal Processing for Liquid Argon Time Projection Chambers

We investigate a deep learning-based signal processing for liquid argon time projection chambers (LArTPCs), which are fine-grained, fully-active particle detectors widely used in neutrino experiments. This approach frames signal region-of-interest (ROI) identification as an image segmentation task. To ensure robustness against detector variations in real data, we train models on two types of variation datasets: samples with simulation-level detector variations and samples with image-level augmentations. This method improves performance on detector features like malfunctioning wires and maintains robustness even without variation samples. Our results demonstrate that this method offers a scalable and efficient performance, supporting its deployment for current and future neutrino data processing.