RESUMO
Spectral image fusion techniques combine the detailed spatial information of a multispectral (MS) image and the rich spectral information of a hyperspectral (HS) image into a high-spatial and high-spectral resolution image. Due to the data deluge entailed by such images, new imaging modalities have exploited their intrinsic correlations in such a way that, a computational algorithm can fuse them from few multiplexed linear projections. The latter has been coined compressive spectral image fusion. State-of-the-art research work have focused mainly on the algorithmic part, simulating instrumentation characteristics and assuming independently registered sensors to conduct compressed MS and HS imaging. In this manuscript, we report on the construction of a unified computational imaging framework that includes a proof-of-concept optical testbed to simultaneously acquire MS and HS compressed projections, and an alternating direction method of multipliers algorithm to reconstruct high-spatial and high-spectral resolution images from the fused compressed measurements. The testbed employs a digital micro-mirror device (DMD) to encode and split the input light towards two compressive imaging arms, which collect MS and HS measurements, respectively. This strategy entails full light throughput sensing since no light is thrown away by the coding process. Further, different resolutions can be dynamically tested by binning the DMD and sensors pixels. Real spectral responses and optical characteristics of the employed equipment are obtained through a per-pixel point spread function calibration approach to enable accurate compressed image fusion performance. The proposed framework is demonstrated through real experiments within the visible spectral range using as few as 5% of the data.
RESUMO
Reactions as the attack by naphthenic and hydrogen sulfide have caused corrosion problems in the petroleum industry due to they affect the crude oil heating furnaces and distillation towers at temperatures between 220 and 400⯰C. The total acid number (TAN) measurement has been used as a test to quantify the acid compounds in crude oils and has shown to be a reliable indicator of their corrosion degree. However, the standard method for the TAN measurement, ASTM D-644, involves long times, environment unfriendly wastes and high costs for each analysis. A more appropriate method for the TAN determination is implemented in this paper, by correlating Fourier transform infrared spectroscopy (FTIR) spectral data of the samples with the standard method measurements using multivariate regression models. In particular, the intensities and frequencies of their mid-infrared attenuated total reflectance (MIR-ATR) spectra (4000 - 400â¯cm-1) are used as independent variables of several principal component regression (PCR) and partial least squares regression (PLSR) models. The latter are employed to correlate the spectra with their respective TAN values so as to obtain a suitable prediction model. Twenty-six (26) samples of Colombian crude oils are used for the study with a TAN ranging from 0.1 to 6.8â¯mg KOH/g crude oil (ASTM D-664). The models are evaluated according to the coefficient of determination (R2), the root mean square error of calibration (RMSEC) and of prediction (RMSEP). The best model is obtained via PLSR using as few as four components (i.e. factors), which attains a calibration R2 of 0.981 and an RMSEC of 0.317â¯mg KOH/g crude oil, while for prediction it attains an R2 of 0.996 and an RMSEP of 0.160â¯mg KOH/g crude oil. It is observed that the functional groups COOH, CH3 and CH2 contribute the most to the prediction models. The designed methodology is faster and environmentally friendly since it does not require sample pretreatment and the use of toxic reagents, and of low-cost compared with the standard procedure since FTIR measurements can be easily taken anywhere using a hand-held or portable spectrometer and a laptop.