Abstract: The theory of particle swarm optimization is reviewed, and two technical ways of endmembers extraction are analyzed. A particle swarm optimization-based endmembers extraction algorithms for hyperspectral imagery is proposed, which is based on the theories of particle swarm optimization, convex geometry and the linear spectral mixture model. The fast implementation method of this algorithm is designed. This algorithm needn’t suppose that there are pure pixels in hyperspectral images, as well as this algorithm can preserve the shape of the endmembers’ spectrums. It carries out the experiments by simulative and AVIRIS hyperspectral image, and the results among the PSO-based algorithm, SGA and NMF are compared and analyzed. The results of experiments prove the PSO-based algorithm is more accurate than SGA and NMF.

Key words: hyperspectral imagery, particle swarm optimization, linear mixture model, endmembers extraction

摘要： 回顾了粒子群算法的基本原理，分析了端元提取算法的两种技术途径。利用粒子群优化的原理，结合凸面几何学理论和线性光谱混合模型，设计了一种粒子群优化端元提取算法，并设计了算法的快速实现方法。该算法不需要假设影像中存在纯像元，同时保持了端元光谱的形状。利用模拟数据和AVIRIS影像对该算法、SGA算法和NMF算法进行实验对比分析，实验结果证明该算法的端元提取精度优于其他二者。

关键词: 高光谱影像, 粒子群优化, 线性混合模型, 端元提取