中文摘要：

TM遥感影像中专题信息的自动提取是目前遥感界的研究热点,也已成为遥感信息生产流程中的瓶颈环节.本文分析了耕地专题信息自动提取的研究现状,选取了吉林中部和辽宁省东北部作为研究区,采用监督分类的方法对研究区TM遥感影像进行分类,提取了玉米地专题信息,玉米地信息的提取精度为85.5%.根据遥感影像目视解译的原理,提出了基于多特征空间的遥感影像专题信息自动提取的研究方法,通过对多特征空间的数学描述和计算机处理实现遥感专题信息的自动提取，采用多特征空间的方法将玉米地信息分为光谱特征空间、形状特征空间、区域地学特征空间和干扰特征空间,应用ERDAS8.5遥感图像处理软件中的knowledge engineer模块的开发功能,设计了基于多特征空间的玉米地专题信息自动提取的专家库，使用专家库对玉米地信息进行了自动提取，提取精度为92.9%.从基于多特征空间的分类结果与监督分类结果的比较发现，基于多特征空间的自动提取方法可以提高专题信息的提取效率，对未来实现遥感影像的智能解译是一种研究方法的探索。

英文摘要：

Many researches have focused on the automatic extraction of information from Landsat/TM remote sensing images. However, many of them seemed to be low effective, and the application of remote sensing technology is limited. Firstly, this paper analysed the current situation about how to extract arable land information from remote sensing images, and middle Jilin province and northeast Liaoning Province were selected as study area. The supervised classification method was adopted in the Landsat/TM image classification and maize land in the study area was extracted from the Landsat/TM images with the precision of up to 85.5 %. Secondly, the method of automatic information extraction based on the multicharacters space in remote-sensing images was put forward. According to the method, maize fields information space was divided into several sub-spaces including spectral feature space, shape feature space, interference feature space and local geoscience-based feature space. Automatic classification of thematic information about maize land was carried out based on TM images and use of the method above. An expertise database was created for the automatic extraction of maize land from Landsat/TM images on the basis of multi-characters space with the use of knowledge engineer module from the ERDAS 8.5 software. The extraction of maize fields in the study area from the images was performed again on the basis of the expertise database. It was found that the interpretation was notably improved with the precision of 92.9%. Comparing this classification result with the traditional visual interpretation, it was concluded that the new method adopted in the paper could improve efficiency of thematic information extraction from the remote sensing images. At the same time, the method appears to have wide application perspective and have a great potential to be used in other areas. The method was also theoretically significant for automatic interpretation of remote-sensing images in the future.