|
|
|
| Detection of Nitrite in Sausages Based on Feature Extraction of Hyperspectral Images Using Principal Component Analysis |
| CHEN Xiaodong, GUO Peiyuan* |
| School of Computer and Information Engineering, Beijing Technology and Business University, Beijing 100048, China |
|
|
|
Abstract This paper proposes a rapid and nondestructive method based on hyperspectral imaging system to detect the nitrite
content in sausages. Firstly, the hyperspectral data in the wavelength range of 400–1 000 nm of 45 sausage samples were
obtained and analyzed by principal component analysis. The third principal component (PC3) was selected for investigation,
and four characteristic wavelengths were obtained, i.e. 402.47, 483.04, 642.27 and 961.82 nm. Furthermore, by combing the
characteristic wavelengths and spectra, the data in the wavelength range of 800–950 nm were definitively chosen to build
the models for quantitative analysis of nitrite in sausages based on the average spectra obtained from the region of interest
(ROI) and chemical measurements using partial least squares regression (PLSR) and genetic algorithm (GA)-optimized back
propagation neural network (BPNN). Results showed that the determination coefficient (R2) and root mean square error
of cross-validation (RMSECV) of the PLSR model were 0.899 and 0.291, respectively, whereas those of the BPNN
model were 0.918 and 0.365, respectively. As the number of samples increased, the BPNN model was increasingly
superior to the PLSR model.
|
|
|
|
|
|
|
|
|
