基于高光谱成像技术与异构集成学习的龟甲药材生长年限鉴别  

Identification for Different Growth Years of Plastrum Testudinis via Hyperspectral Imaging Technique and Heterogeneous Ensemble Learning

作　　者：位云朋 胡会强 毛晓波[1] 赵宇平[2] 张蕾 盛文涛 WEI Yun-peng;HU Hui-qiang;MAO Xiao-bo;ZHAO Yu-ping;ZHANG Lei;SHENG Wen-tao(Department of Biomedical Engineering,School of Electrical and Information Engineering,Zhengzhou University,Zhengzhou 450001,China;National Resource Center for Chinese Materica Medica,China Academy of Chinese Medical Sciences,Beijing 100020,China;School of Pharmacy,Jiangxi University of Chinese Medicine,Nanchang 330004,China;Hubei Jingshan Shengchang Turtle Breeding Farm,Jinmen 431800,China)

机构地区：[1]郑州大学电气与信息工程学院生物医学工程系,河南郑州450001 [2]中国中医科学院中药资源中心,北京100020 [3]江西中医药大学药学院,江西南昌330004 [4]湖北京山盛昌乌龟原种场,湖北荆门431800

出　　处：《光谱学与光谱分析》2024年第9期2613-2619,共7页Spectroscopy and Spectral Analysis

基　　金：国家中医药管理局中医药创新团队及人才支持计划项目(ZYYCXTD-D-202205);中央本级重大增减支项目课题(2060302-2101-26);国家重点研发计划项目(2017YFC1702901)资助。

摘　　要：龟甲是常见的集药、食两用的中药之一,富含维生素、氨基酸、胶原蛋白及大量矿物质成分,被广泛应用于贫血、骨质疏松、免疫力低下等临床症状的医疗与日常饮片炮制。研究表明,龟甲的生长年限越长,其滋阴有效部位及微量元素含量越充足。由于对生长规律认识不足、培育不规范等因素,市场上普遍存在以次充好的现象。目前对龟甲生长年限的鉴别主要通过经验法与理化手段。经验法具有较强的主观性,不利于推广应用;理化技术操作周期长,会破坏样本的完整性。考虑到传统经验、理化检验等鉴别方法的局限性,该研究构建了一种基于高光谱成像技术的龟甲年限鉴别模型。以不同生长年限的龟甲药材为研究对象,采用高光谱成像系统采集原始龟甲药材在可见近红外(VNIR)与短波红外(SWIR)透镜下的高光谱图像,并建立基于支持向量机(SVM)、逻辑回归(LR)与K近邻(KNN)分类策略的异构集成学习模型。结果表明,基于VNIR与SWIR融合波段下的高光谱图像包含更丰富的光谱信息,采用异构集成学习模型可以有效地对龟甲年限实现精确鉴别。模型在龟甲背甲与腹甲样本的测试集准确率分别达到96.14%与93.82%,表明龟甲背甲对其生长年限的鉴别更有优势。考虑到快速性检测的因素,采用波段选择方法剔除冗余特征,降低龟甲药材图像的复杂度,并采用特征波段表征龟甲药材的光谱信息,进一步提升模型分类性能。结果表明,模型在波段数目为32时的背甲样本可以达到96.35%的分类准确率,超过了全波段光谱数据的鉴别精度,表明波段选择策略对提取有效光谱信息的可行性。基于高光谱成像技术的异构集成学习模型可以快速、准确地鉴别龟甲药材的生长年限,为龟甲及其他药材属性的检测提供新的技术参考。Plastrum Testudinis is a popular traditional Chinese medicine(TCM)with abundant medicinal and edible value,and it is widely applied to clinical medical treatment and medicinal slice preparation.Studies show that the contents of trace elements in Plastrum Testudinis are directly proportional to its growth years.However,due to inexperience and nonstandard breeding,adulterated Plastrum Testudinis medicines are on the market.Because of the limitation of empirical and chemical-based methods,a heterogeneous ensemble learning(HEL)method based on a hyperspectral imaging technique is proposed to identify the growth years of Plastrum Testudinis.First,the Plastrum Testudinis samples with different growth years are taken as research objects.The original hyperspectral images of visible near-infrared ray(VNIR)and short-wave infrared ray(SWIR)lenses are captured on the hyperspectral imaging system.Then,the heterogenous ensemble learning(HEL)model is constructed based on support vector machine(SVM),logistic regression(LR),and K-nearest neighbors(KNN).Results show the fused hyperspectral images of VNIR and SWIR include more abundant spectral information.The HEL model can achieve satisfactory prediction ability by identifying the different growth years of Plastrum Testudinis samples.In addition,considering the detection efficiency,an unsupervised band selection is employed to reduce the complexity,eliminate the redundant bands in hyperspectral images,and improve the classification performance further.When the number of selected spectral bands is 32,the classification accuracy reaches 96.35%.Experimental results demonstrate that the HEL model based on hyperspectral imaging can accurately and rapidly identify the different growth years of Plastrum Testudinis samples and provide a novel technique reference for the attributes identification of TCM.

关 键 词：龟甲 高光谱图像 波段选择 集成学习 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]