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。

To improve the efficiency of classification while ensuring clarity and accuracy, a new algorithm of real-time unsupervised classification based on the normalized spectral vector and GPU parallel optimization is proposed. Classification accuracy is improved via the spatial coherence property, while the computing speed is improved via GPU parallel processing. The algorithm procedure starts with the spectral similarity calculation of adjacent pixels via GPU parallel processing. The safety threshold is acquired via Gaussian fitting. Similar pixels are classified as homogenous regions by similarity measurement of the spectral angle. Represented by the average spectral vectors, similar homogeneous regions restricted by the safety threshold are merged. AVIRIS data are used to evaluate the performance of the proposed algorithm. Theoretical analysis and experimental results show that the proposed algorithm outperforms current unsupervised classification algorithms in terms of classification accuracy and computing efficiency.