Author Affiliations
1School of Information and Control Engineering, Xi'an University of Architecture and Technology, Xi'an, Shaanxi 710055, China2Shanxi Provincial Institute of Cultural Relics Protection, Xi'an, Shaanxi 710075, Chinashow less
Fig. 1. Characterization of reflection spectrum of grottoes surface with different weathering types and degrees
Fig. 2. Spectral reflectance of grotto surface with different weathering types and degrees
Fig. 3. Characteristics of the first derivative of spectral reflectance of grotto surface with different weathering types and degrees
Fig. 4. Technical block diagram of weathering types and degrees evaluation method
Fig. 5. Experimental process
Fig. 6. Multispectral imaging data collection on grotto surface. (a) RGB image of target collection area; (b) 640 nm multi-spectral image of target area; (c) distribution of sampling points
Fig. 7. Multi-spectral data reconstruction flow chart
Fig. 8. Normalized spectral reflectance data of grotto surface with different weathering types and degrees
Fig. 9. Characteristics of the first derivative of standardized spectral reflectance date of grotto surface with different weathering types and degrees
Fig. 10. Assessment results of pure weathering area
Fig. 11. Assessment results of four algorithms on overall weathering types and degrees of grotto surface. (a) True weathering types and degrees; (b) RF algorithm; (c) SVM algorithm; (d) SAM algorithm; (e) CNN algorithm
Algorithm | Train accuracy /% | Test accuracy /% | Prediction accuracy /% | Kappa coefficient |
---|
RF | 99.91 | 99.89 | 98.49 | 0.98 | SVM | 97.63 | 94.53 | 90.28 | 0.86 | SAM | 98.72 | 97.78 | 62.65 | 0.51 | CNN | 99.99 | 99.64 | 59.61 | 0.57 |
|
Table 1. Comparison of prediction results of four algorithms
Algorithm | Class | Classification ratio /% | Total /% |
---|
Wstrong | Wslight | Wdust | Wweak |
---|
| Wstrong | 99.92 | 0 | 0 | 0 | 14.05 | | Wslight | 0 | 100 | 0.11 | 0.21 | 10.05 | RF | Wdust | 0 | 0 | 99.89 | 0 | 53.05 | | Wweak | 0.08 | 0 | 0 | 99.79 | 22.86 | | Total | 100 | 100 | 100 | 100 | 100 | | Wstrong | 89.04 | 0 | 0 | 4.31 | 15.78 | | Wslight | 4.28 | 97.32 | 4.38 | 16.97 | 8.35 | SVM | Wdust | 0 | 0.53 | 94.11 | 0.57 | 52.55 | | Wweak | 6.68 | 2.16 | 1.50 | 78.15 | 23.32 | | Total | 100 | 100 | 100 | 100 | 100 | | Wstrong | 80.13 | 0.05 | 0 | 47.45 | 28.81 | | Wslight | 0.01 | 95.68 | 0.06 | 0 | 25.29 | SAM | Wdust | 0 | 0 | 99.94 | 4.15 | 21.11 | | Wweak | 19.86 | 4.26 | 0 | 48.40 | 23.79 | | Total | 100 | 100 | 100 | 100 | 100 | | Wstrong | 44.12 | 4.41 | 0 | 3.62 | 10.79 | | Wslight | 55.88 | 95.90 | 1.39 | 95.86 | 21.14 | CNN | Wdust | 0 | 0 | 98.44 | 0.52 | 28.74 | | Wweak | 0 | 0 | 0.16 | 0 | 39.34 | | Total | 100 | 100 | 100 | 100 | 100 |
|
Table 2. Comparison of confusion matrix of four algorithms