Efficient spectral matching, particularly for multicomponent spectra

[0001] This document concerns an invention relating generally to identification of unknown spectra obtained from spectrometer measurements, and more specifically to identification of unknown spectra via comparison of the unknown spectra to reference spectra.

[0002] A molecular spectrometer (sometimes referred to as a spectroscope) is an instrument wherein a solid, liquid, or gaseous sample is illuminated, often with non-visible light such as light in the infrared region of the spectrum. The light from the sample is then captured and analyzed to reveal information about the characteristics of the sample. As an example, a sample may be illuminated with infrared light having known intensity across a range of wavelengths, and the light transmitted and/or reflected by the sample can then be captured for comparison to the illuminating light. Review of the captured spectra can then illustrate the wavelengths at which the illuminating light was absorbed by the sample. To illustrate, FIG. 1 illustrates an exemplary spectrum (denoted U) presented as a plot of light intensity versus light wavelength. The spectrum, and in particular the locations and amplitudes of the “peaks” therein, can be compared to libraries of previously-obtained reference spectra (denoted as L₁, L₂, . . . L_N) to obtain information about the sample, such as its composition and characteristics. In essence, the spectrum serves as a “fingerprint” for the sample and for the substances therein, and by matching the fingerprint to one or more known fingerprints, the identity of the sample might be determined.

[0003] However, as when identifying a fingerprint, it can prove difficult and time-consuming to find a match for an unknown spectrum in a reference library. Even where an unknown spectrum is obtained from a sample having the same composition as the one from which a reference spectra was captured, an exact match is unlikely owing to differences in the measurement conditions between the unknown and reference spectra (e.g., differences in intensity/amplitude, differences in wavelength scaling/binning, different background noise levels, etc.). Further, while reference spectra are usually obtained from pure substances, unknown spectra often aren't. The unknown spectrum will therefore not match a single one of the reference spectra L₁, L₂, . . . L_N, and will rather match a combination of two or more of these spectra. In such a combination, the spectra are effectively “overlaid” with each other, though each may have a different weight depending on the relative concentrations of the substances from which they originate. It should be appreciated that if one wishes to compare an unknown spectrum U to all possible combinations of one or more reference spectra L₁, L₂, . . . L_N, this will typically be an exceedingly large number, particularly where a large reference library may have tens of thousands of entries (N being equal to the number of these entries). The computational time needed to perform these comparisons can be further magnified if quantitative analysis is to be performed as well as qualitative analysis, i.e., where the relative proportions of component spectra within the unknown spectrum are to be determined as well as their identities. Such quantitative analysis may require that regression be performed between a combination of reference spectra versus the unknown spectrum to determine the weighting that each reference spectrum should have to result in a combination which best matches the reference spectrum. As a result, exhaustive spectral matching can sometimes take hours—or even days—to perform, even where dedicated computers or other machines with high-speed processors are used.

[0004] The invention, which is defined by the claims set forth at the end of this document, is directed to methods and systems which can at least partially alleviate the aforementioned problems, and provide accurate spectral matches with fewer computations (and thus with greater speed). A basic understanding of some of the features of preferred versions of the invention can be attained from a review of the following brief summary of the invention, with more details being provided elsewhere in this document. To assist in the reader's understanding, the following review makes reference to the accompanying illustrations, which are briefly reviewed in the “Brief Description of the Drawings” section following this Summary section of this document.

[0005] Once an unknown spectrum is obtained from a spectrometer, a database, or another source, candidate spectra within the unknown spectrum can be identified in the following manner. Initially, comparison spectra—i.e., reference spectra for comparison—are accessed from one or more spectral libraries or other sources. The unknown spectrum is then compared to at least some of the comparison spectra to determine the degree to which the unknown spectrum corresponds to the comparison spectra. This step is schematically illustrated at 200 in FIG. 2, wherein the unknown spectrum U is compared to comparison spectra L₁, L₂, . . . L_N. If the unknown spectrum U corresponds to a comparison spectrum to a desired degree—as by meeting or exceeding some user-defined or preset correspondence threshold—the comparison spectrum is regarded as being a candidate spectrum B(1). Thus, so long as the correspondence threshold is not set too high, several candidate spectra B(1)₁, B(1)₂, . . . B(1)_M will usually be identified. (If no candidate spectra are identified, the correspondence threshold can be set to a lower value.) Note that each candidate spectrum will represent a single component, i.e., a single reference spectrum obtained from a spectral library or other source (which generally represents the spectrum of a pure component/substance).

[0006] Next, the possibility that the unknown spectrum might have arisen from a multi-component mixture is considered. New comparison spectra are generated, with each comparison spectrum being a combination of one of the previously identified candidate spectra and one of the comparison spectra from the spectral libraries or other sources. The unknown spectrum is then again compared to at least some of these new comparison spectra to determine the degree to which the unknown spectrum corresponds to the new comparison spectra. This step is schematically illustrated at 210 in FIG. 2, wherein the unknown spectrum U is compared to new comparison spectra:
B(1)₁+L₁, B(1)₁+L₂, . . . B(1)₁+L_N
(i.e., the first of the previously identified candidate spectra from 200 combined with each of the comparison spectra from the spectral libraries or other sources);
B(1)₂+L₁, B(1)₂+L₂, . . . B(1)₂+L_N
(i.e., the second of the previously identified candidate spectra from 200 combined with each of the comparison spectra from the spectral libraries or other sources); and so forth, until the unknown spectrum U is compared to new comparison spectra:
B(1)_M+L₁, B(1)_M+L₂, . . . B(1)_M+L_N
(i.e., the last of the previously identified candidate spectra from 200 combined with each of the comparison spectra from the spectral libraries or other sources).

[0007] Where these comparisons find that one of the new comparison spectra has a desired degree of correspondence to the unknown spectrum U (as by meeting or exceeding the correspondence threshold), the new comparison spectrum is regarded to be a new candidate spectrum. These new candidate spectra are depicted in FIG. 2 at step 210 as B(2)₁, B(2)₂, . . . B(2)_M. (It should be understood that if desired, M in step 210 need not be equal to M in step 200, i.e., the number of candidate spectra in step 210 need not be the same as the number of candidate spectra in step 200.) Here each candidate spectrum B(2)₁, B(2)₂, . . . B(2)_M represents two components, i.e., two combined reference spectra obtained from a spectral library or other source.

[0008] The foregoing step can then be repeated one or more times, with each repetition using the candidate spectra identified in the foregoing step to generate new comparison spectra. This is exemplified by step 220 in FIG. 2, wherein the candidate spectra B(2)₁, B(2)₂, . . . B(2)_M from step 210 are used in combination with the comparison spectra L₁, L₂, . . . L_N from the spectral libraries or other sources to generate new comparison spectra. Comparison of the unknown spectrum with these new comparison spectra in turn identifies new candidate spectra B(3)₁, B(3)₂, . . . B(3)_M (wherein M again need not be equal to M in steps 210 and/or 200). Repetition may cease when the candidate spectra include some desired number of components, e.g., once the new comparison spectra include a desired number of combined comparison/reference spectra obtained from a spectral library or other source. This condition is illustrated in the flowchart of FIG. 3, wherein box 300 is equivalent to step 200 of FIG. 2, box 330 is equivalent to steps 210 and 220 of FIG. 2 (as well as future repetitions of these steps), and the condition box 340 evaluates the number of components c in the candidate spectra and ceases repetition once some maximum number C is reached. Alternatively or additionally, repetition may cease when some desired number of candidate spectra are identified; when one or more candidate spectra are identified which match the unknown spectrum by at least some qualifying correspondence value (the qualifying correspondence value being greater than the threshold correspondence value), as discussed below; or when other suitable conditions arise.

[0009] At least some of the candidate spectra may then be presented to a user, with the candidate spectra preferably being presented to the user in ranked order such that those candidate spectra having greater correspondence to the unknown spectrum are presented first (as depicted at step 350 in FIG. 3). An exemplary format for an output list of candidate spectra that might be presented to a user is illustrated in FIG. 4. Here, details regarding the unknown spectrum are given in the heading of the output list, followed by details of the candidate spectra. The first listed candidate spectrum—listed with a rank/index of 1—is a spectrum for polystyrene film, and has a match metric (roughly equal to a “percentage of match”) of 99.58 versus the unknown spectrum. The spectral library or other source of this candidate spectrum is also listed (here “User Example Library”), as is its location within the library/source (at “Source Index” no. 2, i.e., it is the second spectrum provided in the “User Example Library”). The second listed candidate spectrum is actually a combination of three spectra from spectral libraries or other sources—a Toluene (Transmission Cell) spectrum, an ABS Plastic (ATR Corrected) spectrum, and Polytetrafluouroethylene Film spectrum—wherein these spectra, when combined in appropriate proportions (as discussed below), yield a 68.97 match metric with the unknown spectrum. Their cumulative match metrics are also presented, with Toluene having a 56.96 match metric, Toluene and ABS together having a 68.92 match metric, and Toluene, ABS, and Polytetrafluouroethylene collectively yielding the 68.97 match metric. Again, the libraries or other sources of these spectra are provided along with an indication of the location of each spectrum within its library/source.

[0010] Additional metrics are also preferably provided with the output list, in particular, the weight of each comparison spectrum (each component/reference spectrum) within the candidate spectrum, i.e., the scaling factor used to adjust each comparison spectrum to obtain the best match with the unknown spectrum. For example, the first listed candidate spectrum (Polystyrene Film) has a weight of 5.4195, meaning that the unknown spectrum is estimated to have 5.4195 times the polystyrene content of the sample from which the candidate spectrum was obtained. The second listed candidate spectrum contains different weights of toluene, ABS, and polytetrafluouroethylene, with these weights being determined by regression analysis of the comparison spectra versus the unknown spectrum during the aforementioned comparison step (i.e., the various component/reference spectra within a comparison spectrum are proportioned to attain the best match to the unknown spectrum during comparison). Thus, the user may be provided with an at least approximate quantization of the components within the unknown spectrum.

[0011] The methodology above can be said to find “best-match” reference spectra, combine the best-match spectra to other reference spectra, and then identify further best-match spectra from these combinations (with the methodology iteratively continuing from the foregoing combination step). It is therefore seen that rather than comparing all possible combinations of reference spectra L₁, L₂, . . . L_N, the methodology can consider far fewer combinations, basically by pruning out the reference spectra which have less similarity to the unknown spectrum. As a result, the methodology returns high-quality matches in far shorter time than in methods that consider all combinations, particularly where large numbers of reference spectra are used and where the unknown spectrum is reviewed for larger combinations of component/reference spectra—in some cases, returning results in minutes where hours were previously needed.

[0012] Further advantages, features, and versions of the invention are reviewed below, in conjunction with the accompanying drawings.

[0017] Expanding on the discussion above, reference (comparison) spectra for use in the invention can be obtained from one or more spectral libraries or other sources. The spectral libraries used in the invention may be commonly available or proprietary libraries, and such libraries may each contain any number of reference spectra (i.e., the library may consist of a single reference spectrum, or conversely may include many thousands of reference spectra). Further, such reference spectra may be derived from actual measurements, from theory and mathematical computation, or from combinations of experimental and computational data.

[0018] Prior to performing the aforementioned comparisons between the unknown and comparison spectra, the invention may perform one or more transforms on one or both of the unknown and comparison spectra to expedite and/or increase the accuracy of the comparison process, or otherwise enhance data processing. As examples, the invention might perform one or more of data smoothing (noise reduction), peak discrimination, resealing, domain transformation (e.g., transformation into vector format), differentiation, or other transforms on spectra. The comparison itself may also assume a variety of forms, as by simply comparing intensities/amplitudes across similar wavelength ranges between unknown and comparison spectra, by converting the unknown and comparison spectra into vectorial forms and comparing the vectors, or by other forms of comparison.

[0019] Additionally, the methodology described above can be modified to further expedite the identification of candidate spectra. As one example of such a modification, when generating a new comparison spectrum by combining a previously-identified candidate spectrum and a comparison spectrum obtained from a spectral library or other source, the combination might be skipped or discarded (i.e., deleted or not counted as a potential new candidate spectrum) if the candidate spectrum already contains the comparison spectrum. To more specifically illustrate, consider the situation where comparison spectrum L₁, which is obtained from a spectral library, is selected as B(1)₁ in step 200 (FIG. 2) owing to a sufficient match with unknown spectrum U. In the next iteration at step 210, the new comparison spectrum B(1)₁+L₁ can be skipped or discarded since it is equivalent to L₁+L₁ (i.e., reference spectrum L₁ combined with itself, which will merely again result in L₁). Thus, by avoiding the generation and/or use of comparison spectra which have redundant component spectra, the methodology can reserve computation time for comparison spectra which are more likely to yield new candidate spectra.

[0020] As another example of a modification that can be implemented to expedite the identification of candidate spectra, if a candidate spectrum matches the unknown spectrum by at degree greater than or equal to some “qualifying” correspondence value this qualifying correspondence value being greater than the threshold correspondence value—the comparison spectra therein (i.e., its component spectra) can be excluded from any later generation of new comparison spectra. In essence, this measure takes the approach that if a candidate spectrum is already a very good match for an unknown spectrum (e.g., if it has a qualifying correspondence value of above 95%), this may be sufficient, and there is no significant need to determine whether the match might be made even higher if the candidate spectrum was combined with other spectra.

[0021] Another modification that can be made to expedite the identification of candidate spectra applies in the special case where one or more of the components of the unknown spectrum are known—for example, when monitoring the output of a process which is intended to generate a material having known components in a predetermined quantity. In this case, during the first round of comparison (step 200 in FIG. 2, step 300 in FIG. 3), the candidate spectra B(1)₁, B(1)₂, . . . B(1)_M can simply be set to the spectra for the known components. Executing the remainder of the method will then serve to identify any additional components (i.e., impurities) that may be present, as well as the relative proportions of the various components.

[0022] It was previously noted that the correspondence threshold—i.e., the degree of match required between the unknown spectrum and a comparison spectrum for the comparison spectrum to be deemed a candidate spectrum—will yield no candidate spectra if set too high. Typically, a value of 90% correspondence is suitable for the correspondence threshold, though this value might be better set lower or higher depending on the details of the spectra under consideration. It is also possible to set the correspondence threshold to zero (or to a value near zero), in which case a candidate spectrum will result from each comparison spectrum. For example, if the correspondence threshold was set to zero in step 200 of FIGS. 2-3, M=N and B(1)₁, B(1)₂, . . . B(1)_M would each correspond to one of L₁, L₂, . . . L_N. In this case, some of the candidate spectra will in reality be poor candidates because of poor matching with the unknown spectrum. It is then useful to rank the candidate spectra in order of highest correspondence to lowest correspondence, and then first consider those candidate spectra with highest correspondence when performing any subsequent steps. In this case, to reduce computations, it can be useful to discard the candidate spectra with lowest correspondence when performing any subsequent steps. For example, one might keep only the top 10%, 25% or 50% of the candidate spectra having highest correspondence, and to use these in subsequent steps.

[0023] It is expected that the invention will be implemented in spectral identification software for use in computers or other systems (e.g., spectrometers) which receive and analyze spectral data. Such systems may include portable/handheld computers, field measurement devices, application specific integrated circuits (ASICs) and/or programmable logic devices (PLD) provided in environmental, industrial, or other monitoring equipment, and any other systems wherein the invention might prove useful.

[0024] Additionally, while the invention has generally been described as being usable in the context of spectral matching for molecular spectrometers, it may alternatively or additionally be used in mass spectroscopy, X-ray spectroscopy, or other forms of spectroscopy. It might additionally be useful in other forms of measurement analysis wherein signals are measured versus reference values, in which case such signals and reference values may be regarded as “spectra” in the context of the invention.

[0025] It should be understood that the foregoing discussion merely relates to preferred versions of the invention, and the invention is not intended to be limited to these versions. Rather, the invention is only intended to be limited only by the claims set out below, with the invention encompassing all different versions that fall literally or equivalently within the scope of these claims.