摘要
限制遥感图像分类精度的一个主要原因是混合像元问题。因此像元分解也一直是遥感研究的一个热点问题 ,本文针对传统像元分解方法的缺点 ,首先对影像进行纯净像元提取 ,再对混合像元进行分解。在提取纯净像元时 ,利用ETM影像的全色波段和单波段不同的分辨率选取训练样本 ,从而克服了传统像元分解中需要两种影像或地面实测资料的缺点。然后用BP神经网络对混合像元进行分解。同时用民乐县的ETM影像作了试验。又利用对应的土地利用图作了验证 ,取得了比较好的效果。
Unmixing pixel is always an important research aspect of remote sensing. In this paper, artificial neural network (ANN) algorithm is used, and the percent of the different landuse is estimated from Landsat ETM image in Heihe, Ganshu. ETM were orthorectified using a digital photogrammetric software package with ground control points collected by differential GPS. The topographical and atmosphere corrections are made and landuse types (water (f1), vegetation (f2), and soil (f3), building (f4)) are offered from landuse map. The samples for training and testing ANN algorithms are got from the endmembers supervised classification of ETM date with different matrix combination . The training and test samples are randomly selected from the ETM image. Then the thematic information are extracted from the image and the rest unclassified pixels unmixed. Experimental results indicated that unmix pixel by improved ANN has performed better than the unsupervised classification,comparing with the landuse statistics.
出处
《遥感信息》
CSCD
2004年第2期27-30,共4页
Remote Sensing Information
基金
国家重点基础研究发展规划 ( 973 )项目 ( 2 0 0 1CB3 0 940 4)
海外青年学者合作研究基金 ( 40 12 80 0 1)
教育部科学技术重点项目( 2 0 0 1)
