论文部分内容阅读
提出了利用贝叶斯推理选择混合像元内端元的模型。考虑到端元光谱的不确定性,基于贝叶斯推理和线性光谱混合模型得到了像元内端元集合的后验概率表达式。在获得端元共存的先验知识基础上,结合端元光谱的正态分布函数,通过最大后验概率得到最佳的端元集合。通过对包含147431个像元的ETM+影像试验表明,相对于IDRISI软件的MRES和PG算法,该算法可削减至少70%的冗余端元,使端元选择错误导致的分解误差降低至少28%。结果表明,由于充分考虑端元光谱的不确定性和端元的共存性,通过贝叶斯推理可以大幅度提高端元选择的正确率,从而改善混合像元的分解精度。
A Bayesian inference model was proposed to select the endmember of mixed pixels. Taking into account the uncertainty of the endmember spectrum, a posteriori probability expression of the set of endmember elements in the pixel is obtained based on a Bayesian inference and linear spectral mixture model. Based on the a priori knowledge of the coexistence of the endmember, the best endmember set is obtained by the maximum a posteriori probability combined with the normal distribution function of the endmember spectrum. Experiments with 147431 pixels ETM + images show that the proposed algorithm can reduce the resolution error by at least 28% compared with the IDRISI MRES and PG algorithms by at least 70% redundant endmember. The results show that Bayesian reasoning can greatly improve the accuracy of endmember selection and improve the decomposition accuracy of mixed pixels, due to the uncertainty of the endmember spectrum and the coexistence of endmember.