基于BOA-FRNN光谱模型的彩绘颜料浓度预测研究

doi:10.3964/j.issn.1000-0593(2025)02-0322-10

Abstract
References
Related Citation (15)

Download: PDF (28551 KB)
Export: BibTeX | EndNote (RIS)

Abstract In recent years, efforts to protect cultural relics and heritage have been intensified, and the strengthening of the preservation and inheritance of historical culture has risen to the level of a national strategy. Painted cultural relics under anthropogenic, sand erosion, and photodamage, the color of cultural relics generally appeared to varying degrees of fading, discoloration, aging, shedding, and loss of disease, so that it is now difficult to see the original face of the mural painting,digital protection and restoration has become an important means of protection and inheritance of painted cultural relics. Based on the spectral reflectance of color fingerprints, this study has taken the spectral reflectance of the external manifestation of pigment composition change as the entry point andused digital methods to map the concentration of painted pigments. To quickly and accurately identify mineral pigment concentration in color painting, the Bayesian Optimization Algorithm (BOA) was used to find the optimal hyperparameters of the Feed-forward Regression Neural Network (FRNN), and a BOA-FRNN spectral model was constructed topredict pigment composition and concentration distribution mapping. Firstly, the color chart of Dunhuang mineral pigments with different concentration gradients was drawn by traditional Chinese painting techniques, and the visible spectral reflectance and chromaticity information of the color chartwas obtained by Ci64UV integrating sphere spectrophotometer.Secondly, the correlation database of pigments' spectral reflectance, chromaticity values, concentration, pigment particle size, and ingredients was constructed based on the measured data. Finally, the pigment concentration was predicted. The results were compared using the two-constant Kubelka-Munk model, BP network model, support vector machine (SVM) regression algorithm, FRNN network model, and BOA-optimized SVM. To improve the accuracy of concentration prediction and model stability, BOA was proposed to optimize the network structure, activation function, and regularization strength of FRNN. Root Mean Squared Error (RMSE) was used as the fitness function, and the optimal regression parameters were selected through iteration to train the model.The results of pigment data prediction showed that the optimal combination model was BOA-FRNN. Experimental data show that the BOA-FRNN proposed in this paper has higher accuracy. The determination coefficient R² of the model test set was 99.55%, theroot mean square error RMSE was 1.805%. The results show that the Dunhuang pigment color database can select the required spectral reflectance more accurately and quickly, thus improving the model's efficiency and simplifying the algorithm's complexity. BOA searches for the optimal hyperparameters of FRNN and can quickly obtain the global optimal solution by iteratively updating the optimal position of hyperparameters. Compared with K-M, BP, SVM, FRNN, and BOA-SVM, the prediction accuracy and model stability are significantly improved, which meets the accuracy requirements of pigment concentration detection and is a feasible new method for fast pigment mapping.

Key words：Painted artifacts; Spectral reflectance; Pigment color database; Spectral prediction model; Concentration mapping

Received: 2024-02-06 Accepted: 2024-07-26

ZTFLH:

O433

Corresponding Authors: FAN Shuo E-mail: 13581064038@163.com

Cite this article:

LIU Zhen,FAN Shuo,LIU Si-lu, et al. Research on Prediction of Pigment Concentration in Color Painting Based on BOA-FRNN Spectral Model[J]. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(02): 322-331.

URL:

https://www.gpxygpfx.com/EN/10.3964/j.issn.1000-0593(2025)02-0322-10 OR https://www.gpxygpfx.com/EN/Y2025/V45/I02/322