Reconstructing Particle Flow Images Using Regression Models and Deep Neural Networks

This work aims to reconstruct accurate images from multiple images acquired by the Large Hadron Collider detector using advanced machine learning techniques. The importance of this task lies in enhancing the precision of image reconstruction for better analysis in high-energy physics experiments. The training data was divided into two sets: the first set focuses on predicting a true image from a sequence of four images, while the second set addresses the lateral pixel shifts in detected images, which result from deviations in particle trajectories due to magnetic fields. To tackle these challenges, we applied various deep learning models. The Multi-Layer Perceptron (MLP) outperformed others in predicting the true image from the first dataset, while the UNet with deep supervision and nested UNet showed superior performance in correcting the image shifts in the second dataset. We trained all models on GPUs to enhance training efficiency. This study introduces novel deep learning strategies to address image shift and distortion, offering significant advancements in image reconstruction tasks for high-energy physics research.