Fig. 1. Schematic diagram of the process for anti-disturbance cross-scene multispectral imaging for classification of pigments in painted cultural relics

Fig. 2. Spectral image of the mural in the 660 nm channel

Fig. 3. Flowchart of anti-spectral disturbance

Fig. 4. Schematic diagram of the overlapping region of two shots. (a) Sketch of the mural sub-lens shooting; (b) multispectral images of overlapping region of sub-lens; (c) spectral image and grayscale histogram of 660 nm overlapping region

Fig. 5. Basic image and histogram specification images. (a) Basic image Lij; (b) histogram specification image Nij

Fig. 6. Anti-spectral perturbation results. (a) Basic imag Sij; (b) anti-spectral perturbation image Fij; (c) Sij histogram；(d) Fij histogram；(e) two-shot multi-spectral image stitching after anti-spectral perturbation

Fig. 7. Painted cultural relic pigment classification network structure with cross-scene domain shift attributes

Fig. 8. Diagram of encoder

Fig. 9. Structure diagram of feature extractor

Fig. 10. Flowchart of classification and identification of pigments in painted cultural relics

Fig. 11. Schematic diagram of multispectral imaging system

Fig. 12. Experimental data of anti-spectral perturbation by sub-shot. (a) Schematic diagram of shot shooting of simulated murals; (b) shot 1 captures multispectral images at 620 nm; (c) shot 2 captures multispectral images at 620 nm; (d) multispectral images before resisting spectral perturbation at 620 nm; (e) multispectral images after resisting spectral perturbation at 620 nm

Fig. 13. Ground truth and marking diagram of simulated mural. (a) Ground truth of simulated mural; (b) marking diagram of simulated mural

Fig. 14. OA and Kappa coefficients of each algorithm before and after anti-spectral disturbance

Fig. 15. Images of simulated pigment plate and simulated mural at 620 nm. (a) Multispectral image of pigment plate at 620 nm; (b) interest marker image of pigment board points; (c) simulation mural of single exposure shot at 620 nm

Fig. 16. Generalization ability experiment. (a) Generated data distribution; (b) data distribution after processing by the discriminator

Fig. 17. Results of cross-scene classification between paint panels and simulated murals. (a) Ground truth; (b) DHCNet^[23]; (c) FDSSC^[24]; (d) SSRN^[25]; (e) SSKR^[26]; (f) HPDM-SPRN^[27]; (g) BASSNet^[28]; (h) DFFN^[29]; (i) proposed method

Fig. 18. Schematic diagram of the scene shooting of the Mingwang statue in Dule Temple

Fig. 19. Colored and labeled images of mural. (a) Mural color image; (b) shot 1 training samples

Fig. 20. Classification effect of different methods on murals. (a) Proposed method; (b) DHCNet^[23]; (c) FDSSC^[24]; (d) SSRN^[25]; (e) SSKR^[26]; (f) HPDM-SPRN^[27]; (g) BASSNet^[28]; (h) DFFN^[29]; (i) SVM

Fig. 21. Schematic diagram of XRF test points and multispectral images of the Thirteenth Venerable

Fig. 22. XRF analysis results of the Thirteenth Venerable. (a) XRF analysis result of point 1; (b) XRF analysis result of point 2; (c) XRF analysis result of point 3; (d) XRF analysis result of point 4

Fig. 23. Pigment multispectral image training set

Fig. 24. Diagram of the pigment distribution of the Thirteenth Holiness