| dc.description.abstract | The multimode fibre (MMF) is a promising tool for high-resolution, minimally invasive imaging, attributed to its miniaturized structure, extensive data transmission density, and exceptional light collection capabilities. Speckle patterns are generated at the MMF's output end as the input light undergoes inter-modal coupling and modal dispersion during propagation through the fibre. Utilizing the MMF as an imaging probe necessitates prior knowledge of the relationship between the input-output ends via fibre calibration. Generation of desired optical fields at one end of the MMF can be achieved using wavefront shaping. However, MMF is highly sensitive to varying states of deformation. MMF-based imaging techniques face challenges in flexibility and real-time versatility, particularly in applications requiring high-resolution imaging of dynamic activities. This thesis explores the integration of compressive imaging and deep learning methods in a multimodal MMF-based imaging system. Firstly, a multimodal operation is demonstrated utilizing a single MMF, which allows the MMF to serve not only as an optical imaging probe but also as a photocurrent detector, with rapid imaging speed enabled by compressive imaging. Wavefront shaping at the distal end of the MMF is discussed by applying a convolutional neural network with a compact architecture to predict multichannel binary patterns for amplitude-only modulations. The versatility and effectiveness of the proposed neural network are further evaluated for image transmission through a perovskite optical fibre under various conditions. Finally, the potential of building a flexible, multifunctional, and high speed MMF imaging probe is verified in a MMF coupler which provides a monitor end for evaluating the fibre configuration and distal end mapping. A generative adversarial network facilitates real-time calibration due to its scalability with data under varying fibre configurations and its accurate predictions of fibre end mappings. The proposed novel MMF imaging methods may enhance research interest in this field and pave the way for a versatile MMF imaging probe capable of multifunctional operations. | en_US |
