在保证分类结果清晰、准确的前提下，为了提高分类执行效率，本文基于图形处理器（graphicprocessing unit, GPU）及并行优化，提出一种基于归一化光谱向量的高光谱图像实时性非监督分类方法。利用高光谱图像的空间一致性有效提高分类精度，同时，利用归一化光谱向量简化了像元间相似性的计算公式，统一了图像内像元处理方式，并利用GPU 并行技术有效提高计算速度。首先，利用GPU 并行处理方法计算空间相邻像元间光谱向量相似性，根据高斯拟合取得安全阈值；然后利用光谱角作为像元光谱相似测度，将相似像元划为同质区；最后以同质区内各像元平均光谱向量表述同质区光谱特征，根据安全阈值合并相似的同质区完成分类。用AVIRIS 数据评估了该方法性能。本文的理论分析和实验结果显示，与现有非监督分类方法相比，该方法分类精度更高，同时，算法本身运行速度更快。

Laser shock peening is a well-known method for extending the fatigue life of metal components by introducing near-surface compressive residual stress. The surface acoustic waves (SAWs) are dispersive when the near-surface properties of materials are changed. So the near-surface properties (such as the thickness of hardened layers, elastic properties, residual stresses, etc.) can be analyzed by the phase velocity dispersion. To study the propagation of SAWs in metal samples after peening, a more reasonable experimental method of broadband excitation and reception is introduced. The ultrasonic signals are excited by laser and received by polyvinylindene fluoride (PVDF) transducer. The SAW signals in aluminum alloy materials with different impact times by laser shock peening are detected. Signal spectrum and phase velocity dispersion curves of SAWs are analyzed. Moreover, reasons for dispersion are discussed.