基于顶点成分分析的高光谱图像端元提取算法

Endmembers Extraction for Hyperspectral Images Based on Vertex Component Analysis

在基于几何学的端元提取这类算法中,VCA以其全自动和端元提取速度快的特点而受到广泛的关注,但它仅利用了高光谱图像的光谱信息,易受异常像元影响,且抗噪性能较差。论文根据高光谱图像中地物在空间上具有成片分布的特点,提出利用空间信息来改进VCA算法端元提取的质量。该约束使VCA算法逐次选择的纯像元位于空间一致区域,并将该区域的均值作为端元。仿真数据和实测数据上实验结果表明,改进的VCA算法有效克服了异常像元的影响,并提高了端元提取的精度。

Among the endmember extraction algorithms based on geometric approaches, VCA is widely utilized {or its full automation and fast execution speed. But VCA only utilizes spectral imformation, which is susceptible to anomaly pixels and noise. According to the materials continuous spatial distribution characteristics, an improved method for VCA with spatial information is proposed. In the algorithm, VCA is promoted to sequentially select pure pixels in homogeneous areas, and the mean spectra is taken as the final endmember. Experimental results on simulated and real hyperspectral data demonstrate that the proposed algorithm not only overcomes the influence of the anomaly pixels, but also increases the precision of the results.