遥感影像中存在着昆合像元,软分类技术将这些像元按照一定的百分比划分为不同的地物类别,亚像元定位技术利用在每个混合像元中所获得的百分比信息,得到一个锐化后的分类影像.像元分解成不同的子像元,代表不同的地物类别成分.进化Agent技术结合一种空间邻域的假设关系,通过繁殖和扩散两种行为模式,分配给每一个亚像元一个确定的位置,从而达到定位的效果.利用合成影像和退化的真实影像进行实验,通过与传统的硬分类进行精度比较,证明进化Agent技术是一种简单易行的亚像元定位算法.

In remote sensed images, mixed pixels are always presen.t Soft classification defines themembership degree of these pixels fordifferent land cover classes. Sub-pixelmapping is a technique designed to use the information contained in thesemixed pixels to obtain a sharpened image. Pixels are divided into sub-pixels, representing the land coverclass fractions. Evolutionary agent algorithms, combinedwith the assumption of spatialdependence, assign a location to every sub-pixel through two reactive behaviors which arebreeding and diffusion. In thepaper, the algorithm is tested on degraded synthetic and real imagery. Theobtained accuracy measures are highercomparedwith conventionalhard classifications, so it isproved to be a simple and effectivemethod for sub-pixel mapping.