Sensory evaluation method for spectral data of mainstream smoke

Patent metadata

Field	Value
Publication number	US-11525774-B2
Application number	US-202117595852-A
Country	US
Kind code	B2
Filing date	Mar 4, 2021
Priority date	Jan 14, 2021
Publication date	Dec 13, 2022
Grant date	Dec 13, 2022

Abstract

A sensory evaluation method for spectral data of mainstream smoke includes: performing a data enhancement on spectral data of mainstream smoke of a plurality of cigarettes; extracting a shallow spectral characteristic from the spectral data of the mainstream smoke of each cigarette; obtaining a shallow sensory quality result of the spectral data of the mainstream smoke of each cigarette based on the spectral data of the mainstream smoke of each cigarette and the shallow spectral characteristic; extracting deep spatial characteristics from the spectral data of the mainstream smoke of each cigarette; obtaining a deep sensory quality result based on the spectral data of the mainstream smoke of each cigarette and the deep spatial characteristics; obtaining a comprehensive sensory quality result according to the shallow sensory quality result and the deep sensory quality result. The sensory evaluation method achieves accurate screening of unknowns in the mainstream smoke.

First claim

What is claimed is:

1. A sensory evaluation method for spectral data of mainstream smoke, comprising:
performing a data enhancement on spectral data of mainstream smoke of each cigarette of a plurality of cigarettes;
extracting a shallow spectral characteristic from the spectral data of the mainstream smoke of each cigarette after the data enhancement;
obtaining a shallow sensory quality result of the spectral data of the mainstream smoke of each cigarette based on the spectral data of the mainstream smoke of each cigarette from which the shallow spectral characteristic is extracted and the shallow spectral characteristic;
extracting deep spatial characteristics from the spectral data of the mainstream smoke of each cigarette from which the shallow spectral characteristic is extracted;
obtaining a deep sensory quality result of the spectral data of the mainstream smoke of each cigarette based on the spectral data of the mainstream smoke of each cigarette from which the deep spatial characteristics are extracted and the deep spatial characteristics; and
obtaining a comprehensive sensory quality result of the spectral data of the mainstream smoke of each cigarette according to the shallow sensory quality result and the deep sensory quality result.

2. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 1 , wherein the spectral data of the mainstream smoke comprises mid-infrared spectral data.

3. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 1 , wherein the step of performing the data enhancement on the spectral data of the mainstream smoke of each cigarette of the plurality of cigarettes comprises:
performing a horizontal flipping on the spectral data of the mainstream smoke of each cigarette;
performing a random cutting on the spectral data of the mainstream smoke of each cigarette after the horizontal flipping;
performing a physical perturbation on the spectral data of the mainstream smoke of each cigarette after the random cutting; and
performing a component perturbation on the spectral data of the mainstream smoke of each cigarette after the physical perturbation.

4. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 1 , wherein the step of extracting the shallow spectral characteristic from the spectral data of the mainstream smoke of each cigarette after the data enhancement comprises:
eliminating an outlier data point from the spectral data of the mainstream smoke of each cigarette of the plurality of cigarettes through a Hotelling T 2 statistic of a spectral vector to eliminate outlier data from the spectral data of the mainstream smoke; and
performing a denoising on the spectral data of the mainstream smoke of each cigarette from which the outlier data is eliminated, by at least one of a second-order differential, a Karl Norris derivative filter, a multivariate scatter correction (MSC) and a wavelet transform (WT).

5. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 4 , wherein the step of obtaining the shallow sensory quality result of the spectral data of the mainstream smoke of each cigarette based on the spectral data of the mainstream smoke of each cigarette from which the shallow spectral characteristic is extracted and the shallow spectral characteristic comprises:
inputting the spectral data of the mainstream smoke of each cigarette from which the shallow spectral characteristic is extracted into a first sensory classification model constructed in advance to obtain the shallow sensory quality result of the spectral data of the mainstream smoke of each cigarette.

6. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 5 , wherein the first sensory classification model is constructed based on principal component analysis (PCA) in combination with a nonlinear support vector machine (SVM), and a method for constructing the first sensory classification model comprises:
performing characteristic selection on the spectral data of the mainstream smoke of each cigarette after the denoising based on the PCA to extract a characteristic peak of each component of the mainstream smoke in the spectral data; and
training, based on the nonlinear SVM, the spectral data of the mainstream smoke of each cigarette from which the characteristic peak is extracted, to obtain the first sensory classification model.

7. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 1 , wherein the step of extracting the deep spatial characteristics from the spectral data of the mainstream smoke of each cigarette from which the shallow spectral characteristic is extracted comprises:
extracting, based on a deep residual convolutional neural network (CNN), the deep spatial characteristics from the spectral data of the mainstream smoke of each cigarette from which the shallow spectral characteristic is extracted.

8. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 7 , wherein the step of obtaining the deep sensory quality result of the spectral data of the mainstream smoke of each cigarette based on the spectral data of the mainstream smoke of each cigarette from which the deep spatial characteristics are extracted and the deep spatial characteristics comprises:
inputting a plurality of deep spatial characteristics extracted based on the deep residual CNN into an SVM in a stack manner to obtain a second sensory classification model; and
inputting the spectral data of the mainstream smoke of each cigarette from which the deep spatial characteristics are extracted into the second sensory classification model to obtain the deep sensory quality result of the spectral data of the mainstream smoke of each cigarette.

9. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 7 , wherein a method for determining a network parameter of the deep residual CNN comprises:
based on a fixed spectral data set of the mainstream smoke of each cigarette, taking an optimal classification error as a first objective function to obtain a first optimal network parameter;
taking a maximum computing efficiency as a second objective function to obtain a second optimal network parameter; and
selecting a balance point of a convolution kernel size of a first optimal convolution kernel corresponding to the first optimal network parameter and a second optimal convolution kernel corresponding to the second optimal network parameter to be a final network parameter of the deep residual CNN.

10. The sensory evaluation method for the spectral data of the mainstream smoke according to claim 1 , wherein the step of obtaining the comprehensive sensory quality result of the spectral data of the mainstream smoke of each cigarette according to the shallow sensory quality result and the deep sensory quality result comprises:
respectively comparing the shallow sensory quality result and the deep sensory quality result with an expert evaluation result to obtain a shallow modeling accuracy rate corresponding to the shallow sensory quality result and a deep modeling accuracy rate corresponding to the deep sensory quality result;
determining a weight of the shallow sensory quality result and the deep sensory quality result according to the shallow modeling accuracy rate and the deep modeling accuracy rate; and
performing a weighted summation on the shallow sensory quality result and the deep sensory quality result to obtain the comprehensive sensory quality result.

Assignees

• China Tobacco Yunnan Ind Co Ltd

Inventors

• Li Chao
• Fan Duoqing
• Wang Mingfeng
• Wang Hui
• Li Exian
• Xiong Wen
• Ye Ling
• Chen Fangrui
• Wei Yuling

Classifications

• G06F2218/04

  Denoising · CPC title

• G06F2218/12

  Classification; Matching · CPC title

• G06F18/2433

  Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection · CPC title

• G06F18/241

  relating to the classification model, e.g. parametric or non-parametric approaches · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title