Open Access

Tobacco adulteration recognition study by hyperspectral data processing under machine learning

Hongliang Zhang

Hongliang Zhang

Hebei Baisha Tobacco Co., Ltd., Baoding Cigarette FactoryWangdu, China
Search for this author on
Sciendo|Google Scholar
Zhang, Hongliang
, 
Mei Chen

Mei Chen

Hebei Baisha Tobacco Co., Ltd., Baoding Cigarette FactoryWangdu, China
Search for this author on
Sciendo|Google Scholar
Chen, Mei
, 
Xiangrong Chen

Xiangrong Chen

Hebei Baisha Tobacco Co., Ltd., Baoding Cigarette FactoryWangdu, China
Search for this author on
Sciendo|Google Scholar
Chen, Xiangrong
, 
Na Ma

Na Ma

Hebei Baisha Tobacco Co., Ltd., Baoding Cigarette FactoryWangdu, China
Search for this author on
Sciendo|Google Scholar
Ma, Na
 and 
Xiang Wei

Xiang Wei

Hebei Baisha Tobacco Co., Ltd., Baoding Cigarette FactoryWangdu, China
Search for this author on
Sciendo|Google Scholar
Wei, Xiang
  
Mar 17, 2025
Tobacco adulteration recognition study by hyperspectral data processing under machine learning's Cover Image
About this article
Previous Article
Next Article
Cite
Download Cover
Published Online: Mar 17, 2025
Received: Nov 13, 2024
Accepted: Feb 15, 2025
DOI: https://doi.org/10.2478/amns-2025-0201
Keywords
© 2025 Hongliang Zhang et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Figure 1.

Improved CNN recognition network structure
Improved CNN recognition network structure

Figure 2.

Visual results of tobacco pca
Visual results of tobacco pca

Figure 3.

The average spectral curve of three kinds of tobacco
The average spectral curve of three kinds of tobacco

Figure 4.

Different pretreatment method classification effect
Different pretreatment method classification effect

CNN model performance of different activation functions

Activation function Correct rate Average precision Average recall Average F1
correction set verification set test set
ReLU 0.977 0.963 0.925 0.96 0.936 0.937
LeakyReLU 0.989 0.997 0.913 0.848 0.844 0.845
Sigmod 0.982 0.971 0.919 0.919 0.912 0.936
Tanh 0.981 0.962 0.878 0.928 0.921 0.923

Modeling time

Methods Modeling Time(s)
SVM RF KNN This model
RAW 0.06 0.04 0.012 0.0003
SG 0.06 0.04 0.012 0.0003
GWS 0.062 0.043 0.011 0.00032
PCA 0.057 0.038 0.009 0.00021

Identification of tobacco origin

Num Model Accuracy Precision Recall F1
1 KNN 89.78% 89.58% 86.45% 87.41%
2 RF 92.78% 93.12% 92.56% 92.88^%
3 Improved CNN model 96.45% 97.42% 97.62% 97.88%
4 Artificial 81.23% 82.31% 84.51% 83.64%

Contrast between different convolution nuclei

Combination of number of convolution kernels Correct rate Average precision Average recall Average F1
correction set verification set test set
(128,128,128) 0.983 0.98 0.855 0.856 0.855 0.851
(128,128,64) 0.965 0.946 0.895 0.908 0.893 0.899
(128,64,64) 0.96 0.956 0.876 0.88 0.876 0.872
(128,64,32) 0.938 0.925 0.837 0.854 0.838 0.835
(64,64,32) 0.931 0.929 0.877 0.881 0.873 0.873
(64,32,32) 0.93 0.918 0.876 0.876 0.869 0.875
(64,32,16) 0.913 0.885 0.831 0.853 0.834 0.833
(32,32,16) 0.916 0.891 0.855 0.857 0.853 0.853

Contrast of different convolution nuclei

Convolution core size Correct rate Average precision Average recall Average F1
correction set verification set test set
(5,5,5) 0.975 0.966 0.874 0.871 0.874 0.873
(5,5,3) 0.965 0.964 0.905 0.904 0.904 0.904
(3,3,3) 0.974 0.968 0.921 0.942 0.936 0.935
(3,5,3) 0.954 0.928 0.832 0.856 0.835 0.834
(3,5,5) 0.962 0.971 0.874 0.882 0.871 0.872
Language:
English
Publication timeframe:
1 times per year
Journal Subjects:
Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other
Journal RSS Feed