Non-Destructive Identification of Hazardousbone Fragments Embedded in the Frozen-Thawed Pork Based on Multispectral Imaging

Hua-feng ZHANG; Wu WANG; Yu-rong BAI; Yi-ru LIU; Tao JIN; Xia YU; Fei MA

doi:10.3964/j.issn.1000-0593(2021)09-2892-06

Journals >Spectroscopy and Spectral Analysis >Volume 41 >Issue 9 >Page 2892 > Article

Spectroscopy and Spectral Analysis
Vol. 41, Issue 9, 2892 (2021)

Non-Destructive Identification of Hazardousbone Fragments Embedded in the Frozen-Thawed Pork Based on Multispectral Imaging

Hua-feng ZHANG^*, Wu WANG^{1; 2; *;}, Yu-rong BAI^1;, Yi-ru LIU^1;, Tao JIN^1;, Xia YU^1;, and Fei MA^{1; 2; *;}

Author Affiliations

1. School of Food and Biological Engineering, Hefei University of Technology, Hefei 230009, China

show less

DOI: 10.3964/j.issn.1000-0593(2021)09-2892-06 Cite this Article

Hua-feng ZHANG, Wu WANG, Yu-rong BAI, Yi-ru LIU, Tao JIN, Xia YU, Fei MA. Non-Destructive Identification of Hazardousbone Fragments Embedded in the Frozen-Thawed Pork Based on Multispectral Imaging[J]. Spectroscopy and Spectral Analysis, 2021, 41(9): 2892 Copy Citation Text

show less

Fig. 1. Flow chart of the identification of FT-LPS-BFs based on MSI technology

Download full size | View in the Article

Fig. 2. Mean spectra of ROIs-1 and ROIs-2

Download full size | View in the Article

Fig. 3. Two-dimensional scatter plots of PCA by using (a) ROIs-1 full spectra, (b) ROIs-2 full spectra and (c) ROIs-2 key spectra

Download full size | View in the Article

Fig. 4. Images of typical samples in SVM models based on image information

Download full size | View in the Article

特征	最小值	最大值	均值	标准差
长度/cm	1.15	2.56	1.71	0.50
厚度/cm	0.16	0.58	0.28	0.11

Table 1. Size of bone fragments

View in the Article

类型	分类模型	样品	训练集			测试集			准确率 /%
类型	分类模型	样品	灵敏度	精确率	特异性	灵敏度	精确率	特异性	准确率 /%
Ⅰ	SVM	FT-LPS	0.84	0.89	0.95	0.47	0.41	0.67	55.56
		FT-LPS-SBF	0.96	0.84	0.91	0.47	0.78	0.93
		FT-LPS-IBF	0.78	0.85	0.93	0.73	0.58	0.73
	NN	FT-LPS	0.98	0.99	0.98	0.70	0.75	0.54	71
		FT-LPS-SBF	0.96	0.96	0.92	0.55	0.96	0.86
		FT-LPS-IBF	0.93	0.99	0.98	0.85	0.86	0.79
Ⅱ	SVM	FT-LPS	1.00	1.00	1.00	1.00	1.00	1.00	100
		FT-LPS-SBF	1.00	1.00	1.00	1.00	1.00	1.00
		FT-LPS-IBF	1.00	1.00	1.00	1.00	1.00	1.00
	NN	FT-LPS	1.00	1.00	1.00	1.00	1.00	1.00	100
		FT-LPS-SBF	1.00	1.00	1.00	1.00	1.00	1.00
		FT-LPS-IBF	1.00	1.00	1.00	1.00	1.00	1.00
Ⅲ	SVM	FT-LPS	0.98	1.00	1.00	1.00	1.00	1.00	100
		FT-LPS-SBF	0.98	1.00	1.00	1.00	1.00	1.00
		FT-LPS-IBF	1.00	0.96	0.98	1.00	1.00	1.00
	NN	FT-LPS	1.00	1.00	1.00	1.00	0.88	0.93	100
		FT-LPS-SBF	1.00	1.00	1.00	1.00	1.00	1.00
		FT-LPS-IBF	1.00	1.00	1.00	0.87	1.00	1.00

Table 2. Performance parameters of SVM and NN models based on (Ⅰ) ROIs-1 full spectra, (Ⅱ) ROIs-2 full spectra and (Ⅲ) ROIs-2 key spectra

View in the Article

模型	真实类	训练集		测试集		总精度/%
模型	真实类	FT-LPS	FT-LPS-BF	FT-LPS	FT-LPS-BF	总精度/%
SVM	FT-LPS	49	1	15	0	100
	FT-LPS-BF	0	100	0	30
NN	FT-LPS	50	0	15	0	95.56
	FT-LPS-BF	0	100	2	28

Table 3. Confusion matrixes of SVM and NN models for the classification of FT-LPS and FT-LPS-BFs based on key wavelengths

View in the Article

模型	真实样品	判别结果		灵敏度	精确率	特异性	总精度 /%
模型	真实样品	FT-LPS	FT-LPS-BF	灵敏度	精确率	特异性	总精度 /%
SVM	FT-LPS	59	6	0.91	0.91	0.95	93.8
	FT-LPS-BF	6	124	0.95	0.95	0.91
NN	FT-LPS	59	6	0.91	0.89	0.95	93.33
	FT-LPS-BF	7	123	0.95	0.95	0.91

Table 4. Results of SVM and NN models based on image information

Download Citation

Save the article for my favorites

Paper Information