Comparison of Paint Classification Methods Based on Spectral Fusion

Kunshan Gu; Jifen Wang

doi:10.3788/LOP202158.2230002

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 22 >Page 2230002 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 22, 2230002 (2021)

Comparison of Paint Classification Methods Based on Spectral Fusion

Kunshan Gu and Jifen Wang^*

Author Affiliations

School of Investigation, People's Public Security University of China, Beijing 100038, China

show less

DOI: 10.3788/LOP202158.2230002 Cite this Article Set citation alerts

Kunshan Gu, Jifen Wang. Comparison of Paint Classification Methods Based on Spectral Fusion[J]. Laser & Optoelectronics Progress, 2021, 58(22): 2230002 Copy Citation Text

show less

Four infrared spectra of 50 paint samples. (a) Original spectra; (b) first derivative spectra; (c) second derivative spectra; (d) third derivative spectra

Fig. 1. Four infrared spectra of 50 paint samples. (a) Original spectra; (b) first derivative spectra; (c) second derivative spectra; (d) third derivative spectra

Download full size

Fig. 2. Effect of K value selection on error rate

Download full size

Fig. 3. Overall classification and recognition rate of five paint samples under 10 spectral data models

Download full size

Fig. 4. Statistical results of minimum classification errors of four kernel functions

Download full size

Fig. 5. Recognition rate of paint samples from different spectral data sets by SVM

Download full size

Fig. 6. The average classification and recognition rate of all kinds of samples by SVM

Download full size

Fig. 7. Discriminant analysis diagram

Download full size

Spectral type	Accuracy /%
Spectral type	Y1	Y2	Y3	Y4	Y5
Original spectra (OG)	14	100	88	84	14
1st derivative spectra (FD)	57	100	100	74	14
2nd derivative spectra (SD)	57	100	100	89	43
3rd derivative spectra (TD)	57	100	100	74	29
G1	29	100	100	79	14
G2	29	100	100	89	29
G3	43	100	100	84	29
G4	71	100	100	89	43
G5	57	100	100	79	29
G6	43	100	100	84	29
Average accuracy	45.7	100	98.8	82.5	27.3

Table 1. Classification and recognition rate of paint samples for each spectral data when K=1

Spectral type	Accuracy /%
	Wilks’ Lambda		Unexplained variance		Mahalanobis distance		Smallest F ratio		Rao’s V
	Training set	Test set	Training set	Test set	Training set	Test set	Training set	Test set	Training set	Test set
OG	94	78	94	88	94	88	94	86	92	76
FD	90	84	94	90	94	86	100	90	92	72
SD	96	90	96	90	100	96	100	90	90	84
TD	98	90	98	88	100	96	98	94	98	86
G1	94	88	92	86	100	98	100	90	92	72
G2	96	90	96	90	100	96	100	90	90	84
G3	98	88	96	88	100	96	100	92	98	86
G4	90	86	100	90	100	96	100	94	96	82
G5	96	84	96	88	98	94	100	100	88	80
G6	92	84	96	88	100	98	100	96	96	86

Table 2. The recognition rate of training sets and test sets of five discriminant analysis models under different spectral data

Function	Eigenvalue	Canonical correlation	Test of function	Wilks’ Lambda	P
F₁	90.092	0.994	1 through 4	0.000	0.000
F₂	14.707	0.968	2 through 4	0.002	0.000
F₃	11.233	0.958	3 through 4	0.028	0.000
F₄	1.876	0.808	4	0.348	0.001

Table 3. Discriminant function characteristic table

Kunshan Gu, Jifen Wang. Comparison of Paint Classification Methods Based on Spectral Fusion[J]. Laser & Optoelectronics Progress, 2021, 58(22): 2230002

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information