结合超分辨率重建的神经网络亚像元定位方法

遥感影像中普遍存在着混合像元，如何分析和解译混合像元一直是人们研究的热点。亚像元定位方法是将混合像元分解成为亚像元，并赋予不同的端元组分，以提高影像整体分类精度的一种技术。本文在神经网络亚像元定位模型的基础上，结合超分辨率重建理论，提出一种新型的BPMAP模型，在每一个类别的组成分图像与亚像元定位图像之间建立起高、低分辨率的观测模型，采用最大后验估计（MAP）算法对BP神经网络的定位结果进行约束，最终确定混合像元内部各组分合适的空间位置。通过对模拟的简单图像和长江三峡地区的ETM影像进行实验，结果表明，与神经网络模型相比，本文方法能够更加有效地解决亚像元定位的问题，进一步消除定位过程中产生的误差，提高精度。

Sub-pixel mapping method based on ANN and super-resolution reconstructed model

Mixed pixels are always the case in remote sensed images, and how to analysis and explain mixed pixels is of importance in remote sensing applications. Sub-pixel mapping is a technique designed to obtain the spatial distribution of the classes inside the pixels with information of different endmembers to improve the accuracy of the classification. In this paper, a new BPMAP model is introduced by combination of the neural network and super-resolution reconstructed technology. The spatial distribution of the sub-pixel can be determined by establishing of observation model between the high-resolution and the low-resolution images after the neural network mapping; with restricted by Maximum A Posteriori (MAP) algorithm. The proposed model was tested on both simple synthetic image and ETM image in the three Gorges area. Results indicate that this method can mapping sub-pixel efficiently, and better performance was observed compared to that of the original ANN model.