摘要高光谱遥感图像在资源勘探、城市规划、环境监测等诸多领域得到了广泛应用。由于传感器分辨率的限制和地物的复杂多样性，导致高光谱图像中存在混合像元现象。因此高光谱图像的端元提取是高光谱遥感定量化应用的重要步骤。N-FINDR算法是一种非常流行的端元提取算法，已经得到了广泛的应用。但是，一方面由于高光谱图像数据量庞大，另一方面N-FINDR算法具有计算复杂度大、串行实现耗时长的缺点，因此，迫切需要对N-FINDR算法进行分布式优化，解决高光谱数据处理的单机瓶颈问题，提高算法的运行效率。73278

本文在研究Hadoop 分布式系统关键技术和NFINDR算法的基础上，基于HDFS文件系统和MapReduce编程模型设计实现了一个高光谱图像端元提取的分布式并行优化算法，给出了相应的存储优化设计和Map-Reduce过程，并利用实际高光谱图像进行了实验验证。实验表明，该方法在保证端元提取精度的基础上，有效提高了算法执行效率。与相应的串行算法相比，在本文实验平台上并行处理像元数为 的高光谱数据时，加速比可达到31倍。

毕业论文关键词  高光谱图像  端元提取  N-FINDR算法  云计算  Hadoop

毕业设计说明书中文摘要

毕业设计说明书外文摘要

Title        Distributed Parallel Optimization Method of      Hyperspectral Images Mixed Pixel Unmixing   

Abstract

Hyperspectral remote sensing image is widely applied in various fields such as resource exploration, town planning, environment monitoring and so on。 The low sensor resolution and the persity of the surface material hinders result in   phenomenon of mixed pixel。 So extracting endmembers from hyperspectral images is an important step for the application of remote sensing quantitative technology。 N-FINDR is a very popular algorithm, which is widely used。 However, on the one hand, hyperspectral image has a large amount of data, on the other hand, the N-FINDR algorithm has a disadvantage of high computational complexity。 Therefore, to resolve the bottleneck problem of a single machine and improve the efficiency of the algorithm, a distributed parallel optimization algorithm of N-FINDR is urgently needed。

On the basis of studying the key technologies of Hadoop distributed system and N-FINDR algorithm, a distributed parallel optimization for extracting endmember from hyperspectral image based on HDFS file system and MapReduce programming model is developed in this paper。 This paper also designed the corresponding storage optimization and the process of Map-Reduce, and verified the result by using real  hyperspectral image。 Experimental results demonstrate that  the optimization algorithm proposed in this paper can effectively improve the efficiency of N-FINDR under a premise of certain precision。 Comparing with the serial algorithm, the parallel optimization algorithm implemented on this experimental platform can reach up to 31 times speedup with the hyperspectral image containing the number of pixels of  。

Keywords  hyperspectral image，endmember extraction，N-FINDR，cloud computing， Hadoop

目   次

1  绪论 1

1。1  研究背景与意义 1

1。2  国内外研究现状 2

1。3  论文的主要内容 3

2  相关概念与技术 5

2。1  高光谱图像相关概念与技术介绍 5

2。2  云计算相关概念与技术介绍