摘要高光谱遥感图像处理中，目标检测技术是其中的关键环节和研究热门。但是由于高光谱图像目标检测处理具有数据量大、算法复杂度高等特性，现有的串行执行算法效率很低，实际应用中很难满足需求。随着GPU通用计算的飞速发展，GPU的多处理核心、高存储带宽等特点，为加速高光谱目标检测算法的效率提供了新的思路。82921

论文分析了高光谱图像目标检测的概念和研究现状，基于L1/2稀疏性解混设计了高光谱目标检测方法，针对算法迭代中计算复杂度与像元的数目成线性关系的特点，利用GPU/CUDA提供的通用并行计算平台和编程模型，设计了并行优化计算方法，对算法的迭代过程进行了并行化重构，重点对涉及与像素数有关的耗时比较多的计算进行了逻辑划分和GPU任务分配。同时，基于VC++开发了高光谱遥感图像目标检测并行处理软件系统，给出了系统的详细设计和主要功能模块实现。通过实际高光谱数据的实验表明，该系统和算法在保证目标检测精度的同时，能有效提高运行速度。

毕业论文关键词  高光谱图像  目标检测  GPU/CUDA   解混  并行

毕业设计说明书外文摘要

Title        Target Detection for Hyperspectral Imagery  based on GPU architecture             

Abstract Throughout the hyperspectral remote sensing image processing, the target detection technique is one of the key link and research focus。 Due to the large amount of hyperspectral image and its high complexity of algorithm, the existing serial algorithms with low efficiency can’t meet the needs of real-time processing。 With the rapid development of GPU general-purpose computing technology, its more powerful parallel processing capability and higher memory bandwidth provide a new effective way to accelerate the target detection algorithm。

This paper designs a new method of target detection for hyperspectral imagery based on L1/2 unmixing。 Because of the linear relationship between the computational complexity and the number of pixels, we design this parallel computing method and reconstructed the iterative process in parallel based on the general-purpose parallel computing platform and programming model provided by GPU/CUDA。 We focus on more time-consuming logical pision calculations involving the number of the pixels and the GPU task distribution。 At the same time, we develop hyperspectral remote sensing image target detection system of parallel processing based on VC++,and give a detailed design of the system and the main function modules 。 The real hyperspectral data shows that the system and the algorithm ensure the target detection accuracy, and can effectively improve the operating speed simultaneously。

Keywords  Hyperspectral Imagery,  Target Detection,  GPU/CUDA,  Unmixing,  Parallel。

目   录

1  绪论 1

1。1  课题研究背景及意义 1

1。2  研究现状 1

1。3  论文主要内容及结构安排 3

2  CUDA编程模型及存储结构 5

2。1  GPU通用计算发展 5

2。2  CUDA编程基础 5

2。3  CUDA软硬件架构 8

2。4  论文实验平台 9

2。5  本章小结 9

3  基于L1/2稀疏性解混的高光谱目标检测算法并行优化