Single-pixel imaging through non-homogeneous turbid media with adaptive illumination.

The presence of scattering media limits the quality of images obtained by optical systems. Single-pixel imaging techniques based on structured illumination are highly tolerant to the presence of scattering between the object and the sensor, but very sensitive when the scattering medium is between the light source and the object. This makes it difficult to develop single-pixel imaging techniques for the case of objects immersed in scattering media. We present what we believe to be a new system for imaging objects through inhomogeneous scattering media in an epi-illumination configuration. It works in an adaptive way by combining diffuse optical imaging (DOI) and single pixel imaging (SPI) techniques in two stages. First, the turbid media is characterized by projecting light patterns with an LED array and applying DOI techniques. Second, the LED array is programmed to project light only through the less scattering areas of the media, while simultaneously using a digital micromirror device (DMD) to project light patterns onto the target using Hadamard basis coding functions. With this adaptive technique, we are able to obtain images of targets through two different scattering media with better quality than using conventional illumination. We also show that the system works with fluorescent targets.

Improving the reliability of deep learning computational ghost imaging with prediction uncertainty based on neighborhood feature maps.

Defect inspection is required in various fields, and many researchers have attempted deep-learning algorithms for inspections. Deep-learning algorithms have advantages in terms of accuracy and measurement time; however, the reliability of deep-learning outputs is problematic in precision measurements. This study demonstrates that iterative estimation using neighboring feature maps can evaluate the uncertainty of the outputs and shows that unconfident error predictions have higher uncertainties. In ghost imaging using deep learning, the experimental results show that removing outputs with higher uncertainties improves the accuracy by approximately 15.7%.