摘要
现有的石窟表面风化程度评估多使用人工局部测量方法,这种方法存在效率较低且评估结果易受主观因素影响等问题,鉴于此,提出了一种多光谱成像与随机森林算法相结合的石窟表面风化智能量化评估方法。通过多光谱成像提取的石窟表面光谱信息对风化类型及风化程度进行表征;利用多光谱特征数据重组和标准化处理建立训练、测试及预测样本;基于最小相对熵理论设计损失函数,训练随机森林算法模型,提取不同风化类型及风化程度样本数据的光谱特征;利用训练后具有特征感知能力的分类模型对石窟多光谱图像每个像素点的风化类型及风化程度进行智能预测评估;使用混淆矩阵和Kappa系数对评估结果进行精度评价。以陕西省延安市清凉山万佛寺万佛窟为例对所提方法进行验证,实验结果表明:目标石窟表面强盐析风化区域所占比例为5.15%,弱盐析风化区域所占比例为27.88%,微盐析风化区域所占比例为27.39%,积尘风化区域所占比例为39.58%,评估结果与实际风化状况基本一致,评估的准确率为98.49%,Kappa系数为0.98。所提方法可以实现像素级的精细化评估。
Existing artificial local measurement methods are often used to evaluate the degree of surface weathering of grottoes.However,such methods are inefficient and the evaluation results are easily affected by subjective factors.In this paper,an intelligent quantitative evaluation method for grotto surface weathering based on multispectral imaging and random forest algorithm was proposed.Multispectral imaging was used to extract the the surface spectral information of grotto to characterize the type and degree of weathering.The multispectral feature data were reorganized and normalized to establish training,testing,and prediction samples.Based on the theory of minimum relative entropy,a loss function was designed to train a random forest algorithm model,and the spectral characteristics of samples with different weathering types and degrees were extracted.The weathering degree of each pixel in multispectral images of grottoes was predicted and evaluated using a classification model with feature perception ability after training.The confounding matrix and Kappa coefficient were used to evaluate the accuracy of the results.The proposed method was verified taking the Wanfo temple grottoes,Qingliang mountain,Yan'an city,Shaanxi Province as an example.Results show that the target grottoes’strong salting-out weathering surface area ratio was 5.15%,weak salting-out weathering area ratio was 27.88%,slight salting-out weathering area ratio was 27.39%,and strong dust weathering zone ratio was 39.58%.The evaluation results were basically in accord with actual weathering conditions.Accuracy was 98.49%and the Kappa coefficient was 0.98.The proposed method can realize pixel-level refined evaluation.
作者
曹赤鹏
王慧琴
王可
王展
张刚
马涛
Chipeng Cao;Huiqin Wang;Ke Wang;Zhan Wang;Gang Zhang;Tao Ma(School of Information and Control Engineering,Xi'an University of Architecture and Technology,Xi'an,Shaanxi 710055,China;Shanxi Provincial Institute of Cultural Relics Protection,Xi'an,Shaanxi 710075,China)
出处
《光学学报》
EI
CAS
CSCD
北大核心
2020年第22期195-205,共11页
Acta Optica Sinica
基金
国家重点研发计划(2019YFC1520500)
国家自然科学基金青年科学基金(61701388)
陕西省科技厅科技合作项目(2020KW-012)
陕西省教育厅智库项目(18JT006)
西安市科技局高校人才服务企业项目(GXYD10.1)。
关键词
光谱学
石窟风化
多光谱成像
特征重组
最小相对熵
预测评估
随机森林
spectroscopy
grottoes weathering
multispectral imaging
recombinant characteristic
minimum relative entropy
prediction evaluation
random forest
作者简介
王慧琴,E-mail:hqwang@xauat.edu.cn。
