云环境下基于并行支持向量机的高光谱影像分类研究

发布时间：2018-06-07 10:20

本文选题：高光谱影像分类 + 云计算　；参考：《福建师范大学》2014年博士论文

【摘要】：高光谱遥感影像具有波段多、数据量大、数据不确定性和监督分类时易受Hughes现象影响等特点,由此对现有的图像信息分析处理技术提出了更高的要求。支持向量机(SVM)是一种基于统计学习理论且已被众多实验所证实的有效学习机制,能较好地解决小样本、非线性、高维数等问题,并已被成功地应用于高光谱分类领域；但对于大规模高光谱影像的分类问题,SVM传统算法(串行)的训练和预测效率低下,而单机和传统分布式环境也难以提供处理海量数据所需的强大并行运算能力和足够的内存空间。有鉴于此,本文引入并行支持向量机(PSVM)和云计算技术,设计出一种基于云计算的并行支持向量机(Cloud-PSVM)分类模型,提出云环境下Cloud-PSVM的增量学习算法和参数的全局优化策略,并将Cloud-PSVM应用于土地利用分类领域,构建基于Hadoop平台的高光谱影像分类云服务。整个研究从计算模式、分类方法和服务模式这三方面入手,旨在保证分类精度的前提下提高高光谱影像分类的效率,推动大规模高光谱影像地物信息提取与机器解译的规模化和智能化。主要研究内容与成果如下： (1)为有效地提高Hyperion高光谱影像的空间分辨率,设计出一种改进型的Gram-Schmidt高光谱影像融合方法,实现了Hyperion高光谱影像与同一遥感平台及同一时相的ALI高空间分辨率影像的高效融合；提出一种基于光谱-地形,以及纹理特征的组合径向基核函数(MRBF),并构建出一种基于MRBF的二叉决策树多类SMO (BDT-SMO)分类器,可有效地提高高光谱融合影像的分类精度。 (2)构建Hadoop云储存平台,采用Hadoop分布式文件系统(HDFS)和Hbase数据库实现大规模高光谱融合影像数据和样本数据的分布式存储,通过合理选择分割策略、存取机制和数据组织形式,可有效地提高大规模融合影像和样本数据的存取效率。 (3)为有效地提高大规模训练样本的并行学习效率,提出一种基于交叉样本的改进型混合并行支持向量机(YBJCF-PSVM)模型,并与GPU技术相结合,以提高单节点的并行学习能力。此外,设计出一种基于MapReduce和YBJCF-PSVM模式的Cloud-PSVM分类器。 (4)将Cloud-PSVM应用于土地利用分类领域。采用MapReduce模式对实验区高光谱融合影像进行并行特征提取,并通过Cloud-PSVM分类器对大规模样本进行并行训练与预测。实验结果表明,Cloud-PSVM分类器能在保证分类精度的前提下较大程度地提高高光谱融合影像的分类效率。此外,为能有效地提高土地利用分类结果的发布效率,还设计并实现了一种基于Hadoop的高光谱融合影像分类的云服务。 (5)在云计算环境下设计出一种基于MapReduce和壳向量的SVM增量学习算法(MapReduce-HASVM),可有效地提高Cloud-PSVM分类器的泛化能力和扩展性。此外,还提出一种基于云计算和并行遗传算法(PGA)的Cloud-PSVM参数分布式全局优化策略,可有效地提高Cloud-PSVM分类器的分类精度和核参数的优化效率。
[Abstract]:Hyperspectral remote sensing images have many characteristics such as multi band, large amount of data, uncertainty of data and the influence of Hughes phenomenon when supervised classification. Thus, higher requirements are put forward for the existing image information analysis and processing technology. Support vector machine (SVM) is an effective learning mechanism based on statistical theory and has been proved by many experiments. It can solve the problems of small sample, nonlinear, high dimensional number and so on, and has been successfully applied to the hyperspectral classification field. But for the classification of large-scale hyperspectral images, the training and prediction efficiency of the traditional SVM algorithm (serial) is low, while the single machine and the traditional distributed environment are also difficult to provide the powerful parallel processing of massive data. In view of this, this paper introduces a parallel support vector machine (PSVM) and cloud computing technology to design a parallel support vector machine (Cloud-PSVM) classification model based on cloud computing, and proposes a global optimization strategy for incremental learning algorithms and parameters of Cloud-PSVM in the cloud environment, and applies Cloud-PSVM to the land. The Hadoop Platform Based Hyperspectral Image Classification cloud service based on the classification field is constructed. The whole study starts with the three aspects of the computing mode, the classification method and the service mode. The purpose is to improve the efficiency of the hyperspectral image classification under the premise of guaranteeing the classification accuracy, and promote the scale of the large-scale high spectral image information extraction and the machine interpretation. The main research contents and results are as follows:
(1) in order to effectively improve the spatial resolution of Hyperion hyperspectral images, an improved Gram-Schmidt hyperspectral image fusion method is designed, which realizes the efficient fusion of Hyperion hyperspectral images with the same remote sensing platform and ALI high spatial resolution images in the same phase, and proposes a kind of spectral terrain and texture features. Combined radial basis kernel function (MRBF), and a MRBF based two fork decision tree multi class SMO (BDT-SMO) classifier, which can effectively improve the classification accuracy of hyperspectral fusion images.
(2) constructing the Hadoop cloud storage platform, using the Hadoop distributed file system (HDFS) and the Hbase database to realize the large-scale hyperspectral fusion image data and the distributed storage of the sample data. Through the rational selection of the segmentation strategy, the access mechanism and the data organization form, the access efficiency of the large-scale fusion image and the sample data can be effectively improved.
(3) in order to effectively improve the parallel learning efficiency of large-scale training samples, an improved hybrid parallel support vector machine (YBJCF-PSVM) model based on cross samples is proposed and combined with GPU technology to improve the parallel learning ability of single node. In addition, a Cloud-PSVM classifier based on MapReduce and YBJCF-PSVM mode is designed.
(4) Cloud-PSVM is applied to the field of land use classification. The MapReduce model is used to carry out parallel feature extraction for hyperspectral fusion images in the experimentation area, and the large-scale samples are trained and predicted by the Cloud-PSVM classifier. The experimental results show that the Cloud-PSVM classifier can greatly improve the classification accuracy. In addition, in order to effectively improve the efficiency of the distribution of land use classification results, a cloud service based on Hadoop Based Hyperspectral fusion image classification is also designed and implemented.
(5) a SVM incremental learning algorithm (MapReduce-HASVM) based on MapReduce and shell vector is designed under the cloud computing environment, which can effectively improve the generalization ability and extensibility of the Cloud-PSVM classifier. In addition, a distributed global optimization strategy for Cloud-PSVM parameters based on cloud computing and parallel genetic algorithm (PGA) is proposed, which can be effectively proposed. Classification accuracy and optimization efficiency of kernel parameters for high Cloud-PSVM classifier.
【学位授予单位】：福建师范大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】