Method and apparatus for characterisation of constituents in a physical sample from electromagnetic spectral information

Computational apparatus with self-learning for characterization of one or more constituents in a physical sample, each constituent consisting of one or combinations of chemical elements and/or their isotopes, molecules and/or their conformations or states, wherein it is configured to implement a method for characterization of one or more constituents in a physical sample from electromagnetic spectral information of such physical sample, each constituent consisting of one or combinations of chemical elements and/or their isotopes, molecules and/or their conformations or states, the method comprising the following steps:

• obtaining the resolution of electromagnetic spectral information,
• extracting, where the spectral resolution corresponding to the spectral information is sub-optical, one or more spectral lines from said electromagnetic spectral information,
• projecting said spectral lines into a sample point of a deterministic feature space, such deterministic feature space consisting of a multiple dimension vector space comprising a plurality of spectral lines with a predetermined vector basis, such plurality of spectral lines with a predetermined vector basis:
• being stored in a database and having been obtained by sub-optical spectral resolution extraction, and
• corresponding to a plurality of known constituents,
• wherein each dimension of said multiple dimension vector space is a prediction feature of the one or more constituents in said physical sample to which the electromagnetic spectral information corresponds, such prediction feature providing determination of quantity, classification and/or identification of one or more constituents in said physical sample,
• wherein the apparatus preferably further comprises a spectroscopy device able to induce a plasma state in a physical sample, said spectral information being obtained from said spectroscopy device, the spectroscopy device preferably consisting of a LIBS device;
• and
• wherein the method further comprises the following steps:
• selecting a minimum of neighbouring sample points within said deterministic feature space, corresponding to a known constituent quantity, such that the projected sample point maximize the covariance with the corresponding minimum neighbouring points,
• selecting, from the minimum neighbouring points, exclusive, interference and unique spectral lines, thereby deriving a local feature space within the neighbouring sample points, and
• predicting the quantification of the constituent from the physical sample to be quantified by correlating the known constituents from the local feature space, taking into consideration the projected spectral lines of said physical sample to be quantified into the covariance feature space of the selected neighbouring sample points.

Non-transitory storage media including program instructions executable to carry out a method for characterization of one or more constituents in a physical sample from electromagnetic spectral information of such physical sample, each constituent consisting of one or combinations of chemical elements and/or their isotopes, molecules and/or their conformations or states, the method comprising the following steps:

• obtaining the resolution of electromagnetic spectral information,
• extracting, where the spectral resolution corresponding to the spectral information is sub-optical, one or more spectral lines from said electromagnetic spectral information,
• projecting said spectral lines into a sample point of a deterministic feature space, such deterministic feature space consisting of a multiple dimension vector space comprising a plurality of spectral lines with a predetermined vector basis, such plurality of spectral lines with a predetermined vector basis:
• being stored in a database and having been obtained by sub-optical spectral resolution extraction, and
• corresponding to a plurality of known constituents,
• wherein each dimension of said multiple dimension vector space is a prediction feature of the one or more constituents in said physical sample to which the electromagnetic spectral information corresponds, such prediction feature providing determination of quantity, classification and/or identification of one or more constituents in said physical sample;
• and
• wherein the method further comprises the following steps:
• selecting a minimum of neighbouring sample points within said deterministic feature space, corresponding to a known constituent quantity, such that the projected sample point maximize the covariance with the corresponding minimum neighbouring points,
• selecting, from the minimum neighbouring points, exclusive, interference and unique spectral lines, thereby deriving a local feature space within the neighbouring sample points, and
• predicting the quantification of the constituent from the physical sample to be quantified by correlating the known constituents from the local feature space, taking into consideration the projected spectral lines of said physical sample to be quantified into the covariance feature space of the selected neighbouring sample points.