Birkl, GerhardKuijper, ArjanMittenbühler, MarcelWerhan, DanielDanielWerhan2026-01-152026-01-152025https://publica.fraunhofer.de/handle/publica/503208Neutral atom quantum computers are a promising platform for quantum computing. They trap Rydberg atoms in optical dipole traps arranged in an array structure. To allow efficient interaction between the atoms, they need to be moved to predefined structures that create a computational grid. To allow this, the state of the traps needs to be determined. To evaluate the results of the performed quantum operations, the states of the traps need to be determined in the same way. This thesis investigates different approaches for atom detection for this specific application. Observations are performed by fluorescence imaging with a CCD camera, which leads to noisy observations. This noise can interfere with atom detection and cannot be easily reduced during image acquisition. So far, Difference of Gaussians (DoG) has been used in the experiment setup. The aim was to investigate whether it is possible to improve this approach or if different denoising techniques yield better results. For this purpose, different approaches for denoising were investigated, which include filters and dimensionality reduction. The investigated denoising approaches have been demonstrated to be able to improve the atom detection in the experiment, or shown promising results that may be investigated further. Afterward, the resulting denoised observations were clustered using different implementations of Gaussian Mixture Models. The aim of this is to not use hard clustering, which assigns the intensity values into rigid groups, but instead calculate a probability for each point of interest of belonging to each used cluster. Extracting these probability values from the fitted models allows determining a prediction of the states of the traps present in an image. This has been demonstrated to be a good way to predict the state of the traps present in a given image. Additionally, it was investigated whether this prediction can be improved using a novel approach using Hidden Markov Models (HMM). One problem with the current approach is determining the probability values for the prior of having an atom. This value is only easily determinable for the initial observations, and not after movement operations were performed. Hidden Markov Models allow for a better modeling of the transitions between the trap states. It has been investigated how these transitions can be modeled, and whether a Hidden Markov Model can improve the atom detection for later states of the experiment.enBranche: Infrastructure and Public ServicesResearch Line: Computer vision (CV)LTA: Scalable architectures for massive data setsQuantum computingHidden Markov Model (HMM)Pattern recognitionNoise reductionbachelor thesis