光谱学与光谱分析, 2019, 39 (1): 166, 网络出版: 2019-03-17  

基于多波段漫反射光谱古陶瓷窑口无损鉴定

Ancient Ceramic Kiln Non-Destructive Identification Based on Multi-Wavelength Diffuse Reflectance Spectroscopy
作者单位
1 上海大学通信与信息工程学院, 上海 200444
2 新型显示技术及应用集成教育部重点实验室, 上海 200072
3 中国科学院上海硅酸盐研究所, 上海 201899
4 上海大学文化遗产保护基础科学研究院, 上海 200444
摘要
古陶瓷是历史的遗存, 具有不可再生性, 因而理想的古陶瓷分析技术应该是无损的。 为客观、 有效地对古陶瓷窑口进行无损鉴定, 提出了一种基于紫外、 可见光和近红外的多波段漫反射光谱无损鉴定方法。 针对传统单一波段古陶瓷窑口鉴定对目标特征描述不足的问题, 即在可见光波段区域, 漫反射光谱数据可反映古陶瓷的颜色特征, 但在同一窑口烧制的陶瓷也会有不同的颜色属性, 仅仅根据可见光波段的漫反射反射率来鉴定窑口来源是不合理的, 在紫外与近红外波段, 古陶瓷内部分子与此波段光发生作用后的漫反射光谱数据, 可反映古陶瓷携带的丰富样品结构和物质属性信息, 结合紫外与近红外光谱漫反射光谱数据可有效提高特征的表达, 因此提出利用紫外、 可见光和近红外的多波段特征提取方法。 在实验过程中, 基于多波段线性特征融合窑口平均鉴定准确率为92.9%, 相比于单波段的窑口鉴定平均准确率91.1%提高了1.8%, 实验结果验证了所提多波段方法相对单波段方法的有效性; 在特征提取过程中, 常用小波变换对光谱信号进行处理, 但由于古陶瓷漫反射光谱波信号在紫外、 可见光与近红外波段形波动大, 频率变化大, 因此, 在小波基的选取上存在很大困难, 提出利用自适应时频分析特征提取方法, 其特点是可自适应分配不同频率子波本征模态函数, 通过选择合适的本征模态函数来提取古陶瓷不同波段的光谱特征, 但在分解过程中存在过分解现象, 即虚假的本征模态函数, 将所有样本与分解的本征模态函数的平均相关系数和平均方差贡献率作为选择本征模态函数的标准, 实验结果表明, 随着分解阶次的递增平均相关系数和平均方差贡献率递减, 当分解阶次为4时, 相关系数和方差贡献率都为0.30, 但当分解阶次为5时, 相关系数和方差贡献率仅为0.15和0.18, 因此选择4阶分解, 用于不同波段的特征提取; 所提取的特征给与分类器进行分类时, 不同波段的特征对分类的准确率贡献不同, 因此在此基础上, 计算不同光谱特征的散布矩阵, 利用类内与类间散布矩阵的迹, 计算特征融合时不同波段特征的权重, 自适应分配权重并进行非线性特征融合, 权重越大, 表明该类特征对鉴别的贡献越大, 非线性特征融合时, 平均鉴定准确率为94.5%, 相比于线性特征融合鉴定平均准确率92.9%提高了1.6%; 其中分类器采用k最近邻分类器对来自不同窑口的古陶瓷进行无损分类识别。 通过客观定量地将该方法与同类方法进行对比, 朱旭峰等利用非线性特征融合方法, 窑口平均鉴定准确率为86.97%, 该方法比其高7.53%。 刘峰等采用基于协方差阵进行特征级融合多波段方法, 窑口平均鉴定准确率为89.63%, 该方法比其高4.87%。 实验结果表明所提方法有效、 可行, 可作为古陶瓷窑口鉴定的有效辅助鉴定方法。
Abstract
Ancient porcelain is a remnant of history and has non-regenerability, so the ideal ancient ceramic analysis technique should be nondestructive. In order to objectively and effectively identify ancient ceramics kiln, a non-destructive method has been developed based on ultraviolet, visible and near-infrared diffuse reflectance spectroscopy to identify ancient ceramic kiln objectively and effectively. In view of the lack of description of the target characteristics in the traditional single-band ancient ceramic kilns, for example, the diffuse reflectance spectroscopy data can reflect the color characteristics of ancient ceramics in the visible region, but the ceramics fired in the same kiln will have different color property, only based on the diffuse reflectance of the visible light band to identify the source of the kiln unreasonable, in the ultraviolet and near-infrared and ultraviolet light band, the ancient ceramic interior molecules and the band light after the diffuse reflectance spectral data reflect the ancient rich sample structure and material properties carried by the ceramic can effectively improve the expression of the features by combining the UV and near-infrared spectral reflectance spectroscopy data. Therefore, we propose a multi-band feature extraction method using ultraviolet, visible and near infrared. During the actual experiment, the average identification accuracy based on multi-band linear feature fusion kiln is 92.9%, which is 1.8% higher than the average accuracy of 91.1% for single-band kiln identification. The experimental results verify that the multi-band method is effective; In the process of feature extraction, wavelet transform is often used to process the spectral signal. However, since the ancient ceramic reflection spectrum wave signal is in the ultraviolet region, the waveform of the diffuse reflection spectrum in the visible and near-infrared region is not only fluctuating but also changing greatly in frequency. Therefore, it is very difficult to select the wavelet basis. In this paper, The feature extraction method is characterized by adaptively distributing the intrinsic mode functions of different frequency wavelets, selecting the appropriate intrinsic mode functions to extract the spectral characteristics of different wavelength bands of the ancient ceramics, but there exists an over-decomposition phenomenon in the decomposition process, that is to say, the intrinsic mode function of the false component. The average correlation coefficient and the mean contribution rate of all the intrinsic mode functions of all the samples and the decomposition are taken as the criteria for selecting the intrinsic mode function. The experimental results show that with the decomposition order increases, the average correlation coefficient and the mean variance contribution rate decrease, and when the decomposition order is 4, the contribution rate of correlation coefficient and variance are 0.30, but when the decomposition order is 5, the contribution rate of correlation coefficient and variance is only 0.15 and 0.18. Therefore, the fourth-order decomposition is chosen for feature extraction in different bands. On this basis, calculate the distribution matrix of different spectral features, use intra-class and class scatter matrix traces, calculate the weight of different band features when the feature is fused, the greater the weight, indicating that the greater the contribution of such features to the identification; Finally, the k nearest neighbor classifier is used to classify the ancient ceramics from different kilns. By comparing objectively and quantitatively the proposed method with similar methods, Zhu Xufeng used non-linear feature fusion method, and the average identification accuracy of kiln is 86.97%, and the method of this paper is 7.53% higher than this method. Liu Feng used covariance matrix to solve the feature weight of multi-band method, and the average identification accuracy of kiln is 89.63%, and the method of this paper is 4.87% higher than this method. The experimental results show that the proposed method is effective and feasible. It can be used as an effective auxiliary appraisal method for ancient ceramic kiln identification.

李净, 管业鹏, 李伟东, 罗宏杰. 基于多波段漫反射光谱古陶瓷窑口无损鉴定[J]. 光谱学与光谱分析, 2019, 39(1): 166. LI Jing, GUAN Ye-peng, LI Wei-dong, LUO Hong-jie. Ancient Ceramic Kiln Non-Destructive Identification Based on Multi-Wavelength Diffuse Reflectance Spectroscopy[J]. Spectroscopy and Spectral Analysis, 2019, 39(1): 166.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!