No-reference quality assessment for neutron radiographic image based on a deep bilinear convolutional neural network

Neutron Imaging and Learning Algorithms: New Perspectives in Cultural Heritage Applications

“…To obtain good NI data, quality improvements to the radiographic images are developed in previous works thanks to specific algorithms to find the best possible settings to the detriment of phase sensitivity and spatial resolution [ 22 ]. The data acquisition could be optimized through experimental procedures related to the beam characteristics and collimation, optical and detection systems, and secondary effects such as multiple scattered neutrons, and γ radiation background at the beamlines [ 23 , 24 , 25 ]. Indeed, the neutron source characteristics, such as the neutron fluence rate, could have statistical fluctuations in terms of time and space, which gives rise to the Gaussian noise, or mixed Poisson–Gaussian noise, which can affect the image quality [ 23 ].…”

“…The data acquisition could be optimized through experimental procedures related to the beam characteristics and collimation, optical and detection systems, and secondary effects such as multiple scattered neutrons, and γ radiation background at the beamlines [ 23 , 24 , 25 ]. Indeed, the neutron source characteristics, such as the neutron fluence rate, could have statistical fluctuations in terms of time and space, which gives rise to the Gaussian noise, or mixed Poisson–Gaussian noise, which can affect the image quality [ 23 ]. In the data analysis process, one of the challenges in cultural heritage applications is the segmentation of the internal regions because the objects are generally composed of different materials where the edges are not sharp due to corrosion and degradation processes, or the penetration of one phase in another is very common, such as the case of the liquid vase content in the ceramic or stone matrix ( Figure 1 ).…”

“…Recently, Deep Learning techniques are applied in the NI context to address the sensitivity-resolution trade problem [ 23 ]. The latter is based on a set of algorithms that attempt to model high-level abstractions in data by using model architectures, with unsupervised or semi-supervised feature learning and hierarchical feature extractions [ 26 ].…”

“…Deep Learning (DL) models are successfully applied in different stages of data processing from the acquisition to the data analysis in several contexts and with different investigation techniques, spanning from X-ray computed tomography (XCT) to Neutron Imaging [ 23 , 26 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 ].…”

No‐reference image quality assessment based on multiscale feature representation

“…The FR‐IQA and RR‐IQA algorithms assume that all or part of the information of the reference image is available for IQA tasks. However, reference images are usually unavailable in many fields, such as remote sensing [3] and neutron radiography [4]. NR‐IQA method only takes the distorted image to be assessed as input and thus is more practical yet more challenging.…”