Methods of tobacco classification via hyperspectral imaging
Methods of classifying tobacco include imaging tobacco with an imaging system to obtain an image and classifying the tobacco as very low nicotine (VLN) or traditional based on the obtained image. The imaging system includes a hyperspectral camera and an image processing system.
1 . A method of classifying tobacco, the method comprising:
placing a reflectance material on the tobacco for use in preprocessing;
imaging tobacco with a hyperspectral imaging system comprising a hyperspectral camera and an image processing system to obtain an image;
defining a first region of interest for the reflectance material;
defining a second region of interest for the tobacco;
after the image is obtained by the hyperspectral camera, removing pixels with a response less than a threshold response within the second region of interest;
creating a mean spectral vector for each pixel along an x-axis of the first region of interest by averaging the first region of interest along a y-axis corresponding to each pixel along the x-axis; and
classifying the tobacco as very low nicotine (VLN) or traditional based on the obtained image and the mean spectral vector.
2 . The method of claim 1 , further comprising:
placing the tobacco on a conveyor belt configured to pass underneath the hyperspectral camera.
3 . The method of claim 2 , wherein the imaging the tobacco occurs while the tobacco is moved linearly underneath the hyperspectral camera via the conveyor belt and movement of the tobacco is tracked by the image processing system to provide a consistent image.
4 . The method of claim 2 , wherein the classifying the tobacco includes analyzing the image in real time as the tobacco is moved linearly underneath the hyperspectral camera.
5 . The method of claim 1 , wherein the hyperspectral camera is configured to image the tobacco with shortwave infrared (SWIR) imaging.
6 . The method of claim 5 , wherein the SWIR imaging operates between 900 nanometers (nm) and 2500 nm.
7 . The method of claim 1 , wherein the classifying includes extracting relevant features from the obtained image.
8 . The method of claim 1 , wherein the classifying is performed by a machine learning algorithm.
9 . The method of claim 8 , wherein the machine learning algorithm is at least one of logistic regression or linear discriminant analysis.
10 . The method of claim 8 , further comprising:
training the machine learning algorithm via a plurality of images of tobacco with a known classification of VLN tobacco or traditional tobacco.
11 . The method of claim 1 , wherein the hyperspectral camera images the tobacco to construct a two-dimensional image of a surface of the tobacco for each spectral wavelength captured by the hyperspectral camera.
12 . The method of claim 1 , wherein the image includes a plurality of pixels, each of the plurality of pixels includes a plurality of spectral measurements.
13 . The method of claim 12 , wherein each of the plurality of pixels includes at least 160 spectral measurements, the at least 160 spectral measurements being defined by the hyperspectral camera.
14 . The method of claim 1 , wherein the classifying the tobacco as VLN or traditional is non-invasive.
15 . The method of claim 1 , wherein a rectangular area of the tobacco is imaged by the hyperspectral camera.
16 . The method of claim 15 , wherein the rectangular area is a 12″ by 30″ area.
17 . The method of claim 1 , further comprising:
correcting each of the pixels within the second region of interest with the mean spectral vector to create a mean spectra for the tobacco.
18 . The method of claim 17 , wherein the tobacco is classified as VLN or traditional based on the mean spectra for the tobacco.
19 . The method of claim 17 , wherein the correcting each of the pixels within the second region of interest includes discarding a pixel of the second region of interest if there is no corresponding element of the mean spectral vector.