Identifying substances stored in containers utilizing a portable Raman probe
The present disclosure relates to a portable Raman probe that uses axicon lenses (“axicons”), grating axicons (“graxicons”), and other optical elements to collect spectra from substances behind non-opaque barriers using off-axis techniques. The portable Raman probe can also include a Raman spectrometer to analyze the collected spectra, and a spectra refinement machine-learning model to generate refined spectra from raw or noisy collected samples.
1 . A portable Raman probe comprising:
a grating axicon surrounding a collection lens that accepts a path of light and directs the path of light out of the portable Raman probe toward a focal point, wherein the grating axicon includes a substantially planar first surface that receives light and a substantially planar second surface opposite the substantially planar first surface that transmits the light out of the portable Raman probe toward the focal point;
the collection lens that receives excited light that scatters or fluoresces off of a substance beyond a non-opaque container barrier; and
a spectra refinement machine-learning model that generates a refined spectrum from a noisy spectrum associated with the excited light collected by a spectrometer by reducing noise and fluorescence from noisy spectra.
2 . The portable Raman probe of claim 1 , wherein the spectra refinement machine-learning model:
determines a filter edge at a beginning of the noisy spectrum; and
filters out the noisy spectrum before the filter edge.
3 . The portable Raman probe of claim 1 , wherein the spectra refinement machine-learning model removes broadband background fluorescence from the noisy spectrum or noise from the noisy spectrum.
4 . The portable Raman probe of claim 1 , wherein the spectra refinement machine-learning model comprises a U-Net neural network architecture that processes the noisy spectrum as a 1-dimensional data set.
5 . The portable Raman probe of claim 1 , further comprising generating the spectra refinement machine-learning model utilizing different augmented data for each training cycle.
6 . The portable Raman probe of claim 1 , wherein:
the noisy spectrum does not include any distinct spectral peaks; and
the refined spectrum includes multiple distinct spectral peaks.
7 . The portable Raman probe of claim 1 , wherein:
the grating axicon surrounds the collection lens; and
the grating axicon includes an axicon lens that is encoded to redirect the path of light from a first direction within the portable Raman probe to a second direction outside of the portable Raman probe that intersects with itself at the focal point.
8 . The portable Raman probe of claim 1 , further comprising an optical element that generates the path of light as a ring of light having a fixed non-expanding width, wherein directing the path of light out of the portable Raman probe toward the focal point includes light-shaping accomplished by a planar diffractive surface associated with the grating axicon.
9 . The portable Raman probe of claim 1 , further comprising providing the refined spectrum to a library comparator algorithm for identifying one or more chemical components making up the substance.
10 . The portable Raman probe of claim 1 , wherein the focal point is associated with the substance that is beyond the non-opaque container barrier.
11 . A method for identifying substances stored in non-opaque containers, comprising:
emitting light via a grating axicon of a portable Raman probe toward a focal point associated with a substance located beyond a non-opaque container barrier, wherein the grating axicon includes a substantially planar first surface that receives the light and a substantially planar second surface opposite the substantially planar first surface that transmits the light out of the portable Raman probe toward the focal point;
receiving, by a collection lens of the portable Raman probe, excited light that scattered or fluoresced off of the substance beyond the non-opaque container barrier; and
generating a refined spectrum from a noisy spectrum associated with the excited light collected by a spectrometer by utilizing a spectra refinement machine-learning model that is trained to reduce noise and fluorescence from noisy spectra.
12 . The method of claim 11 , further comprising determining a spectral analysis of the substance based on the refined spectrum.
13 . The method of claim 12 , further comprising matching the spectral analysis of the substance to a known substance to identify the substance stored in a non-opaque container having the non-opaque container barrier.
14 . The method of claim 11 , wherein the spectra refinement machine-learning model:
determines a filter edge at a beginning of the noisy spectrum; and
filters out the noisy spectrum before the filter edge.
15 . The method of claim 11 , wherein the spectra refinement machine-learning model removes noise and broadband background fluorescence from the noisy spectrum.
16 . The method of claim 11 , wherein the spectra refinement machine-learning model comprises a U-Net neural network architecture that processes the noisy spectrum as a 1-dimensional data set.
17 . The method of claim 11 , wherein:
the noisy spectrum does not include any distinct spectral peaks; and
the refined spectrum includes multiple distinct spectral peaks.
18 . The method of claim 11 , wherein:
the grating axicon surrounds the collection lens; and
the grating axicon includes grating axicon that is encoded to redirect the light from a first direction within the portable Raman probe to a second direction outside of the portable Raman probe that intersects with itself at the focal point.
19 . A portable Raman spectrometer probe comprising:
a light source that generates a ring of light;
a grating axicon surrounding a collection lens that accepts the ring of light and directs the ring of light out of the portable Raman spectrometer probe toward a focal point associated with a substance located beyond a non-opaque container barrier, wherein the grating axicon includes a substantially planar first surface that receives light and a substantially planar second surface opposite the substantially planar first surface that transmits the light out of the portable Raman spectrometer probe toward the focal point;
the collection lens that receives excited light that scatters or fluoresces off of the substance beyond the non-opaque container barrier;
a spectra refinement machine-learning model that generates a refined spectrum from a noisy spectrum associated with the excited light collected by a spectrometer by reducing noise and fluorescence from noisy spectra; and
a spectrometer that processes the refined spectrum to generate a spectral analysis of the substance.
20 . The portable Raman spectrometer probe of claim 19 , wherein the spectra refinement machine-learning model converts the noisy spectrum having no distinct spectral peaks into the refined spectrum having multiple distinct spectral peaks.