利用卫星图像结构信息提高城市地物模式识别精度的研究

针对城市地物分布特征及其光谱特点，对卫星遥感数据的空间结构信息应用于城市分类进行了研究。分类结果与传统的统计模式识别方法（Bayes法）相比较简洁、快速，精度提高近2个百分点。

STUDY OF IMPROVING ACCURACY FOR URBAN LANDCOVER MODE RECOGNITION BY USING STRUCTURE INFORMATION OF SATELLITE IMAG

When studying a task of urbanlization, the precision of mode recognition is too low to apply actually using the common statistical classification way (supervised way as Bayes). The investigative way called the Combined Spectral and Structure Classification (shortening as CSSC) more conveniently merge the spatial structure information with the spectral information of satellite image to classify, and compared with the traditional statistical way(as Bayes),the CSSC way can improve the classification precision of the identical area well considering 4 comparative precision index as Division accuracy, Error ratio, Bhattacharrya distance matrix and Kappa coefficient. This study using spatial structure reclassify the result of fuzzy classification based on spectral entirely again and reduce the fuzzy of those pixels in a mass in a furthest degree, hence, it carry out successfully the study aim of improving the precision of city land use classification and acquire several results as followings: (1) The CSSC way can conveniently merge the self possessed spatial structure information with the spectral information of satellite image to improve the classification precision. (2) The used arithmetic in this study is easy to comprehensive theoretically and carry out in practice. The result of applying in the more land cover types and more mixed pixel area like urban is very distinct. (3) Compared with the traditional statistical way(as Bayes), the CSSC way can occupy less memory ,have quick speed and spend less. It must be pointed out the CSSC way in this study only choose lesser land cover types, so it need to study more deeply for the future.