J. Spectral Imaging 12, a2 (2023). doi:10.1255/jsi.2023.a2
Due to the increasing amount of plastic waste and high-quality demands on recycled plastic interest for in-line composition estimation in plastics has grown the last few years. This study investigates pigment blue 15 : 3 with varying concentrations in LDPE. Samples are investigated with two industrial hyperspectral imaging systems where one has the hyperspectral range from 450 nm to 1050 nm and the other from 950 nm to 1750 nm. A model based on peak ratios of selected bands and model based on a principal component analysis have been tested. The models only predict pigment concentrations between 40.0 wt% and 1.7 × 10–3 wt% if both spectral ranges are combined. Unknown samples containing pigment concentration ranging from 20 wt% to 0.31 wt% were predicted and correlated to the actual pigment concentrations (R2 = 0.977) and the PC-based model outperforms the peak ratio model. The studied approach can be a part of the solution to the plastic challenge and can be transferred to other applications where concentration determination is key.
]]>J. Spectral Imaging 12, a1 (2023). doi:10.1255/jsi.2023.a1
We present a simple, but novel, hybrid approach to hyperspectral data cube reconstruction from computed tomography imaging spectrometry (CTIS) images that sequentially combines neural networks and the iterative expectation maximisation (EM) algorithm. We train and test the ability of the method to reconstruct data cubes of 100 � 100 � 25 and 100 � 100 � 100 voxels, corresponding to 25 and 100 spectral channels, from simulated CTIS images generated by our CTIS simulator. The hybrid approach utilises the inherent strength of the Convolutional Neural Network (CNN) with regards to noise and its ability to yield consistent reconstructions and make use of the EM algorithm’s ability to generalise to spectral images of any object without training. The hybrid approach achieves better performance than both the CNNs and EM alone for seen (included in CNN training) and unseen (excluded from CNN training) cubes for both the 25- and 100-channel cases. For the 25 spectral channels, the improvements from CNN to the hybrid model (CNN + EM) in terms of the mean-squared errors are between 14 % and 26 %. For 100 spectral channels, the improvements between 19 % and 40 % are attained with the largest improvement of 40 % for the unseen data, to which the CNNs are not exposed during the training.
]]>