基于城市功能区先验概率的MESMA端元迭代优化研究

One of the key questions about remote sensing application is to improve the accuracy of thematic information which extracted from remotely sensed data. As a classical spectral unmixing model in urban information extraction, multiple endmember spectral mixture analysis (MESMA) performs well in addressing within-class variability but is still affected by the between-class variability, preventing the improvement of unmixing accuracy. This study first fuses multiple sources data to build a functional zone prior probability model, and then this model is used to optimize the endmember selection to lower the impact of between-class variability. Second, MESMA land cover fraction is utilized as supplemental data to optimize functional zone prior probability estimations. Finally, functional zone optimization processes and endmember selection optimization processes are combined to build an iterative optimization model. This research is focusing on reducing the impact of spectral variability, achieving the self-optimization ability, lowering the impact of human uncertainty, and thus improving the unmixing accuracy. The results are desired to provide a reliable and accurate land cover information for the related researches and applications.

提高计算机遥感数据的专题信息提取精度，是遥感应用研究的主要问题之一。作为在城市信息提取的经典光谱分解方法，多端元光谱混合分析模型（MESMA）虽然有效降低同物异谱的影响，但受异物同谱的影响显著，严重制约着MESMA精度的进一步提高。本研究首先融合多源数据，构建像元尺度的功能区先验概率估算模型，利用城市功能区先验概率优化MESMA的端元选择，降低异物同谱的影响；其次以MESMA分解所得地物覆盖类型比例作为补充数据，优化城市功能区先验概率估算；最后组合MESMA端元优化过程和功能区优化过程，构建迭代优化模型。本研究可以有效减少光谱差异性对MESMA的影响，实现模型的自优化，降低人为不确定性，进一步提高混合像元专题信息提取的精度和可靠性。研究结果有望为相关研究应用提供准确可靠的城市地物覆盖类型信息。

提高计算机遥感数据的专题信息提取精度，是遥感应用研究的主要问题之一。作为在城市信息提取的经典光谱分解方法，多端元光谱混合分析模型（MESMA）虽然有效降低同物异谱的影响，但受异物同谱的影响显著，严重制约着MESMA精度的进一步提高。本研究首先融合多源数据，利用机器学习方法，构建功能区识别模型，通过城市功能区先验概率优化MESMA的端元选择，降低异物同谱的影响；其次以MESMA分解所得地物覆盖类型比例作为补充数据，优化城市功能区先验概率估算。本研究有效减少光谱差异性对MESMA的影响，进一步提高混合像元专题信息提取的精度，同时地表覆盖类型信息为城市功能区的识别提供了可靠的数据支撑。

内容获取失败，请点击重试