City Wall Multispectral Imaging Disease Detection Method Based on Convolutional Neural Networks

Min Li; Huiqin Wang; Ke Wang; Zhan Wang; Yuan Li

doi:10.3788/LOP223189

Laser & Optoelectronics Progress, Volume. 61, Issue 4, 0437006(2024)

City Wall Multispectral Imaging Disease Detection Method Based on Convolutional Neural Networks

Min Li¹, Huiqin Wang^1、*, Ke Wang¹, Zhan Wang², and Yuan Li³

¹School of Information and Control Engineering, Xi'an University of Architecture and Technology, Xi'an 710055, Shaanxi, China

²Shaanxi Provincial Institute of Cultural Relics Protection, Xi'an 710075, Shaanxi, China

³Xi'an Museum, Xi'an 710074, Shaanxi, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(16)

Fig. 1. Basic structure of CNN for city wall multispectral imaging disease detection

Download full size

Fig. 2. Basic process of convolution

Download full size

Fig. 3. Network structure with dropout added

Download full size

Fig. 4. Flow chart of wall disease detection

Download full size

Fig. 5. City wall multispectral imaging data. (a) City wall sampling cube data; (b) multispectral imaging data for each channel

Download full size

Fig. 6. MNF analysis results of wall surface multispectral data

Download full size

Fig. 7. RGB images of city wall surface disease. (a) Biological disease; (b) strong salting out disease; (c) weak salting out disease; (d) basal brick

Download full size

Fig. 8. Spectral reflectance of city wall surface diseases

Download full size

Fig. 9. Images of target area on wall surface. (a) RGB image of the target area; (b) multispectral image of the target region at 740 nm;(c) target area marking condition

Download full size

Fig. 10. Visual display of wall surface diseases. (a) False-color image of the target area; (b) prediction results of the proposed algorithm; (c) prediction results of RF algorithm; (d) prediction results of SVM algorithm; (e) prediction results of KNN algorithm

Download full size

Table 1. CNN model structure
View table
Table 1. CNN model structure
Layer Configuration Output shape
Conv2d+ReLU （15，3×3）（3，3，15）
Conv2d+ReLU （45，3×3）（1，1，45）
Dropout 0.25 （1，1，45）
Flatten — 45
Dense+ReLU 30 30
Dropout 0.5 30
Dense 4 4

Table 2. Eigenvalue and cumulative contribution rate of each component of MNF

View table

Table 2. Eigenvalue and cumulative contribution rate of each component of MNF

Component	Eigenvalue	Cumulative contribution rate /%
1	168.68	49.97
2	50.56	64.95
3	45.68	78.48
4	19.13	84.15
5	10.68	88.61
6	9.06	90.25
7	6.75	92.30
8	6.40	94.19
9	3.97	95.37
10	3.02	96.26
11	2.95	97.13
12	2.54	97.89
13	2.09	98.51
14	1.89	99.07
15	1.80	99.60
16	1.35	100.00

Table 3. Number of training sets and test sets for each sample
View table
Table 3. Number of training sets and test sets for each sample
Type Biological disease Strong salting out Weak salting out Basal brick
Training set 4039 3325 2780 3774
Test set 1731 1425 1192 1618

Table 4. Detection results of four algorithms for the pure area of city wall disease
View table
Table 4. Detection results of four algorithms for the pure area of city wall disease
Type Classification results of four algorithms
CNN RF SVM KNN
Biological disease
Strong salting out
Weak salting out
Basal brick

Table 5. Index comparison before and after minimum noise fraction method in convolutional network

View table

Table 5. Index comparison before and after minimum noise fraction method in convolutional network

Type

Biological disease /%

Strong salting out /%

Weak salting out /%

Basal

brick /%

Overall accuracy /%

Kappa /10²

Train

time /s

Predict time /s

Original

98.96

98.60

91.02

87.27

94.17

92.18

510.12

913.41

MNF

98.50

98.04

90.35

85.66

93.28

91.00

141.57

219.24

Table 6. Classification results of wall diseases by different algorithms

View table

Table 6. Classification results of wall diseases by different algorithms

Parameter	RF	SVM	KNN	CNN
Biological disease rate /%	97.00	95.96	93.47	98.50
Strong salting out rate /%	96.63	93.61	88.70	98.04
Weak salting out rate /%	85.00	83.98	79.95	90.35
Basal brick rate /%	84.98	84.30	80.41	85.66
Overall accuracy /%	91.25	89.84	85.79	93.28
Kappa /10²	88.27	86.38	80.94	91.00

Tools

Get Citation

Copy Citation Text

Min Li, Huiqin Wang, Ke Wang, Zhan Wang, Yuan Li. City Wall Multispectral Imaging Disease Detection Method Based on Convolutional Neural Networks[J]. Laser & Optoelectronics Progress, 2024, 61(4): 0437006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites