Moodley, Chané Simone2024-02-012024-02-012024https://hdl.handle.net/10539/37490A thesis submitted in fulfilment of the requirements for the degree of Doctor of Philosophy to the Faculty of Science, School of Physics, University of the Witwatersrand, Johannesburg, 2023Quantum ghost imaging offers many advantages over classical imaging, including the ability to probe an object with one wavelength and record the image with another. The low photon fluxes, which are characteristic to quantum optics, offer the ability to probe objects with fewer photons thereby avoiding photo-damage to light sensitive structures, such as biological matter. Unfortunately, quantum ghost imaging suffers from slow image reconstruction due to sparsity and the probabilistic arrival positions of photons. Progressively, quantum ghost imaging has advanced from single-pixel scanning systems to 2-dimensional (2D) digital projective masks which offer a reduction in image reconstruction times through shorter integration times. The focus of this thesis was time-efficient quantum ghost imaging, the necessary literature is presented and discussed followed by a focus on the technical details and components required for quantum ghost imaging. Several image reconstruction algorithms using two different 2D projective mask types are showcased and the utility of each is discussed. Furthermore, a notable artefact of a specific reconstruction algorithm and projective mask combination is presented and how this artefact can be used to retrieve an image signals heavily buried under artefacts is discussed. Tests to confirm the presence of quantum entanglement were conducted and discussed along with the necessary results to confirm the of quantum entanglement. The Bell inequality was successfully violated with a Bell parameter S > 2, while a full quantum state tomography was performed with an almost perfect fidelity. This thesis was aimed at time-efficient quantum ghost imaging which was achieved through implementing a series of neural network and machine learning based approaches. These approaches consisted of speeding up the image reconstruction process, enhancing images early on in the reconstruction, establishing an optimal early stopping point and achieving image resolutions that are impractical-to-measure in real time. First a two-step deep learning approach was proposed to establish an optimal early stopping point based on object recognition, even for sparsely filled images. In step one the reconstructed image was enhanced after every measurement by a deep convolutional auto-encoder, followed by step two in which a neural classifier was used to recognise the image. This approach was tested on a non-degenerate ghost imaging setup while physical parameters such as the mask type and resolution were varied. A 5-fold decrease in image acquisition time at a recognition confidence of 75% was achieved. The significant reduction in experimental running time is an important step towards real-time ghost imaging, as well as object recognition with few photons, especially in the detection of light sensitive structures. Many computationally intense deep-learning methods have been implemented in an effort to speed up image acquisition times by retrieving image information. Often over-looked, machine learning methods can offer the same, if not better, reduction up in image acquisition time by an object recognition process. Four machine learning algorithms were implemented and trained on a uniquely generated, noised and blurred dataset of numerical digits 1 through 9. Of the tested recognition algorithms, logistic regression showed a 10× speed up in image acquisition time with a 99% prediction accuracy. Additionally, this reduction in acquisition time was achieved without any image denoising or enhancement prior to recognition thereby reducing training and implementation time, as well as the computational intensity of the approach. This method can be implemented in real-time, requiring only 1/10th of the measurements needed for a general solution, making it ideal for quantum ghost imaging and the recognition of light sensitive structures. In quantum ghost imaging the image reconstruction time depends on the resolution of the required image which scale quadratically with the image resolution. A superresolved imaging approach was proposed based on neural networks where a low resolution image was reconstructed. The low resolution image was subsequently denoised and then super-resolved to a higher image resolution. To test the approach, both a generative adversarial network as well as a super-resolving autoencoder network were implemented in conjunction with an experimental quantum ghost imaging setup, demonstrating its efficacy across a range of object and imaging projective mask types. A super-resolving enhancement of 4× the measured resolution was achieved with a fidelity close to 90% at an acquisition time of N2 measurements, required for a complete N × N pixel image solution. This significant resolution enhancement is a step closer to a common ghost imaging goal, to reconstruct images with the highest resolution and the shortest possible acquisition time. The approaches detailed here prove valuable to the community working towards time-efficient quantum ghost imaging. Not only has the image reconstruction process been reduced by up to 10×, but general image solutions have been enhanced through denoising and super-resolving capabilities. Through the introduction of these techniques and algorithms via machine intelligence, a significant step in timeefficient quantum imaging has been achieved.enQuantum imagingQuantumTime-efficient quantum imagingThesis