Image formation is one of the most important aspect of our everyday life. Conventional optical Imaging (and Sensing) exploits light, reaching the detection system from a target or a scene of interest, mainly unscattered. However, there are many practical situations in which unscattered light may be undetectable, insufficient or mispresented. Nonetheless, if the considered system allows it, it could be still possible to exploit scattered light in order to extract relevant information. Problems arise from the fact that, in these cases, light propagation may undergo severe alterations, thus leading to challenging, and sometimes ill- posed, problems.In this thesis, two main scenarios involving scattered light are studied and addressed by means of artificial neural networks. Over the last period, these powerful data-driven algorithms have been extensively employed in many scientific contexts for their ability to solve even complex problems implicitly. Precisely this characteristic is exploited, in the present work, in a non-line- of-sight scenario in order to simultaneously locate and identify people hidden behind a corner. Moreover, a complex-valued neural network algorithm is implemented and applied to the problem of transmission of images through a multimode fibre, demonstrating high-speed and high-resolution image restoration even without the need for any phase measurements. Finally, due to its formulation based on the physics of multimode fibres, a direct comparison is proposed between the same algorithm and a more standard approach.
【 预 览 】
附件列表
Files
Size
Format
View
Artificial neural networks for scattered light imaging
PDF
download
878987797
028-85220240
