摘要
海洋沉积物的粒度研究有助于了解人类活动对自然环境的影响。将主成分分析(PCA)和连续投影算法(SPA)融合能够综合利用两种光谱特征提取方法的优势,获得比单一特征提取方法更丰富的特征波长,实现无关特征和干扰信息的剔除,最大限度减少特征信息的丢失,有利于沉积物粒度的分析。以青岛市东大洋村潮间带表层32份沉积物为例,将海洋沉积物划分为0.3~0.2、0.2~0.1、0.1~0.075和<0.075 mm四个不同粒径的沉积物样品,分别测定不同粒径的32份沉积物的可见-近红外反射光谱,共计128条光谱。将128条光谱数据分别以2∶1,1∶1和1∶2的比例划分建模集和检验集进行分析;采用主成分分析和连续投影融合算法(FOPAS)提取不同粒径沉积物的特征光谱,利用支持向量机算法建立粒径分类模型。结果显示,对2∶1、1∶1、1∶2比例的数据集,融合算法检验集正确率分别为83.33%、82.81%、75.29%,仅在2∶1比例下正确率低于连续投影算法检验集的正确率90.47%,其余正确率相对于单一特征提取算法均有显著的提高,表明使用融合算法提取特征光谱建立的分类模型在训练集样本量少,粒径清晰的条件下,其分类模型相较于单独使用两个特征提取算法的模型更具有优势。采用基于主成分分析和连续投影融合算法的海洋沉积物粒度分类模型,能够提高海洋沉积物粒度分类结果的正确率,建立正确率更高的粒度分类模型,对快速粒度分类提供了解决方法。
The study on the granularity of marine sediments is helpful in understanding the impact of human activities on the natural marine environment.The fusion of principal component analysis and successive projection algorithm combines the advantages of both spectral feature extraction methods.It can obtain richer feature wavelengths than a single feature extraction method,achieve rejection of irrelevant features and interference information,minimize the loss of feature information,and facilitate the analysis of sediment grain size.In this paper,32 sediments from the surface layer of the intertidal zone of East Dayang Village in Qingdao City were divided into four sediment samples with different grain sizes of 0.3~0.2,0.2~0.1,0.1~0.075 and<0.075 mm.The visible-NIR reflectance spectra of 32 sediments with different grain sizes were measured separately,with 128 spectra samples.The 128 spectral samples were divided into modeling set and test set in the 2∶1,1∶1 and 1∶2 ratio for analysis.An algorithm fused with principal component analysis and successive projection algorithm was used to extract the characteristic spectra of different grain-size sediments,and the support vector machine algorithm was used to build a grain-size classification model.The results show that the fusion algorithm test set correct rates of 83.33%,82.81%,and 75.29%at 2∶1,1∶1 and 1∶2,respectively.All the correct rates were significantly improved relative to the single feature extraction algorithm,except for the lower than 90.47%correct rate for the test set of the continuous projection algorithm at the 2∶1 ratio,indicating that the classification models were built by using the extracted feature spectra of the fusion algorithm.The classification model using the fused algorithm with the extracted feature spectra has an advantage over the model using two separate feature extraction algorithms under the condition of a small training set and clear particle size.Adopting a classification model a marine sediment particle size based on principal component analysis and continuous projection fusion algorithm can improve the correct classification rate results of marine sediment particle size,establish a particle size classification model with a higher correct rate,and provide a solution for fast particle size classification.
作者
贾宗潮
王子鉴
李雪莹
邱慧敏
侯广利
范萍萍
JIA Zong-chao;WANG Zi-jian;LI Xue-ying;QIU Hui-min;HOU Guang-li;FAN Ping-ping(Institute of Oceanographic Instrumentation,Qilu University of Technology(Shandong Academy of Sciences),Qingdao 266061,China;College of Computer Science and Technology,China University of Petroleum(East China),Qingdao 266590,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2023年第10期3075-3080,共6页
Spectroscopy and Spectral Analysis
基金
国家自然科学基金项目(32171578)
山东省自然科学基金项目(ZR2021QF028,ZR2021MD093,ZR2021MD103)
齐鲁工业大学科教产融合试点工程基础研究类项目(2022PYI008)资助。
关键词
海洋沉积物
粒度分类
主成分分析
连续投影算法
融合算法
Marine sediments
Particle size classification
Principal component analysis
Successive projection algorithm
Fusion algorithm
