基于近红外高光谱技术的敦煌莫高窟壁画起甲病害风险评估方法

Assessing the degree of flaking of the murals in the Dunhuang Mogao Grottoes using near-infrared hyperspectral imaging

针对敦煌莫高窟壁画起甲病害的发生发展过程,本实验采取了近红外高光谱技术进行分析和评估。高光谱技术由于其非接触、光谱信息丰富能够无损地应用于壁画的病害检测上。首先通过高光谱成像系统采集不同起甲病害程度下的壁画高光谱图像,提取出壁画从未发生起甲到碎片脱落的病害周期内各个阶段对应的光谱信息,从而建立起甲病害光谱数据库;然后分别使用偏最小二乘回归(PLSR)、主成分分析与支持向量机、主成分分析与人工神经网络三种评估模型对此光谱数据库进行训练,并且通过平均误差、决定系数的大小决定评估模型的优劣。实验结果显示PLSR取得了较好的评估结果,同时利用PLSR评估模型中的加权回归系数得到光谱中的重要波段;之后运用PLSR评估模型对未标注风险情况的壁画进行逐像素点的病害风险评估,获得壁画的起甲风险评估图。实验结果显示基于近红外高光谱技术能够对敦煌壁画起甲病害进行风险评估。

A hyperspectral imaging system operated in the near- infrared （NIR） region was used for non -contact measurement of the degree of flaking of the murals in the Dunhuang Mogao Grottoes. First, depending on the degree of flaking on the murals in a small area, the degree of flaking as being of four types were categorized ： normal, slight, serious and complete. Second, hyperspectral images of the site were acquired and the spectral signatures were extracted. Then the site with corresponding spectral data were modeled using partial least squares regression, principal component analysis and support vector instrument, principal component analysis and artificial neural networks to evaluate its degree of flaking. The degree of flaking values were evaluated with coefficients of determination （R- squared） and root mean square errors. Experimental results show that the partial least squares regression （PLSR） method gives very good results. The weighted regression coefficients of the resulting PLSR models were used to identify the most important wavelengths aim to reduce the high dimensionality of the hyperspectral data. By using these important wavelengths, an algorithm was developed to evaluate every pixel in the image for visualizing the degree of flaking in all portions of the samples. The results show that NIR hyperspectral imaging system is a potential technique for non - destructive evaluation of the degree of flaking of the Dunhuang Mogao Grottoes murals.