丝绸2026，Vol.63Issue(9)：13-22,10.DOI:10.3969/j.issn.1001-7003.2026.09.002
清代宫廷服饰绿色染色工艺与色彩重建
Green dyeing process and color reconstruction of the Qing Dynasty court costumes
摘要
Abstract
The Qing Dynasty(1636-1912)marked one of the most sophisticated periods in the development of traditional Chinese dyeing technology,particularly renowned for its mature overdyeing techniques which produced a rich and highly symbolic color palette for court attire.However,the accurate reconstruction of these historical colors remains a significant challenge due to the complex,nonlinear relationships inherent in natural dye systems and the loss of detailed craft knowledge.This study aims to establish a systematic,evidence-based,and intelligent framework for reconstructing the specific green hues found in Qing Dynasty court costumes,so as to move beyond empirical approximation toward scientifically verifiable reproduction. The research employed a comprehensive multi-evidence methodology,integrating historical archival analysis,scientific examination of artifacts,controlled laboratory dyeing experiments,and advanced computational modeling.A Qianlong-period Kesi formal robe,housed in the Beijing Art Museum,served as the primary case study.Initial high-performance liquid chromatography-mass spectrometry(HPLC-MS)analysis of micro-samples from the robe's green sections successfully identified the original dye compositions,revealing two predominant historical overdyeing systems:indigo(Indigofera tinctoria)paired with Chinese cork tree(Phellodendron chinense),and indigo combined with pagoda tree bud(Styphnolobium japonicum). To generate robust datasets for predictive modeling,we designed and executed multi-factor dyeing experiments grounded in quantitative data extracted from the Imperial Weaving and Dyeing Bureau archives.For the indigo/cork tree system,four key parameters were varied:Phellodendron concentration(5％-25％owf),indigo-to-Phellodendron ratio(2.6-3.0),dyeing temperature(50℃to boiling),and dyeing time(15-55 min).For the more complex indigo/pagoda tree bud system,which involves a pre-mordanting step,seven parameters were investigated:alum concentration(3％-15％owf),mordanting temperature(40℃to boiling),mordanting time(10-30 min),pagoda tree bud concentration(30％-150％owf),dyeing temperature(50℃to boiling),dyeing time(15-55 min),and the indigo-to-bud ratio(1-4).A total of 90 and 180 experimental sets were practically dyed for each system,respectively.The CIELAB color coordinates(L*,a*,and b*)of all resulting silk samples were measured using a spectrophotometer under D65 standard illumination. Subsequently,machine learning algorithms were applied to build high-precision color prediction models,with the experimental parameters as inputs and the corresponding CIELAB values as outputs.A comparative evaluation demonstrated that the Gaussian process regression(GPR)model achieved superior performance for the four-parameter indigo/Phellodendron system,yielding test-set root mean square error(RMSE)values of 0.43,0.79,and 0.87 for L*,a*,and b*,respectively.Conversely,for the seven-parameter indigo/pagoda tree bud system,the XGBoost model excelled,with lower test-set RMSE values of 0.54,0.67,and 0.73 for the three color channels.This performance dichotomy is attributed to the GPR model's strength in modeling continuous,smooth relationships with fewer variables,while XGBoost's tree-based structure is more adept at handling the complex,higher-order interactions present in the multi-step process involving mordanting. Leveraging these optimized models,we successfully reconstructed four specific historical greens(labeled A,B,C,and D)identified from the artifact.The color difference(ΔE)between the model's prediction and the validation dyeing result was as low as 0.92,with a minimum reconstruction ΔE of 1.14,confirming high predictive accuracy.Furthermore,acknowledging that the current appearance of the 250-year-old textile results from long-term aging,we subjected the reconstructed samples to controlled ultraviolet light aging experiments.This simulation of the color degradation pathway proved highly effective,reducing the final ΔE between the aged reconstruction samples and the current state of the historical textile.For instance,the ΔE for Color A was reduced from 8.29 to 5.91,demonstrating that modeling the aging process is crucial for approximating the present condition of cultural heritage. The principal innovations of this research are threefold.First,it pioneers the tailored application of machine learning models(GPR and XGBoost)to decipher and predict the nonlinear chemistry of specific historical overdyeing processes.Second,it introduces a novel,closed-loop reconstruction framework that systematically integrates chemical analysis,active learning-based data generation,predictive modeling,and aging simulation.Third,it delivers the set of quantitatively verified reconstruction formulas for specific greens within the Qing court palette,providing concrete technical protocols. In conclusion,this study establishes a robust,interdisciplinary,and intelligent methodology for the accurate reconstruction of historical textile colors.It provides conservators and historians with a powerful new analytical tool,offers reliable technical support for the restoration and faithful reproduction of ancient court attire,and lays a scientific foundation for the informed contemporary application of natural dyes.Future work should further enhance the reliability of reconstruction through multiple approaches,such as introducing more complex model architectures or increasing training data to improve generalization ability,and deepening the understanding of aging mechanisms.关键词
天然染料/染色工艺/色彩重建/HPLC-MS/传统色谱/清代服饰Key words
natural dyes/dyeing process/color reconstruction/HPLC-MS/traditional color palette/Qing Dynasty clothing分类
轻工纺织引用本文复制引用
张艳,刘远洋,刘剑,赵丰,苏淼..清代宫廷服饰绿色染色工艺与色彩重建[J].丝绸,2026,63(9):13-22,10.基金项目
国家社会科学基金项目(20WYSB006) （20WYSB006）
浙江省文化研究工程项目(24WH27-3Z) （24WH27-3Z）
浙江省文物保护科技项目(2025015) （2025015）
