提高TM图像的分类精度,是图像处理及应用领域中一个很重要的研究课题。本文在总结已有成果基础上,首先利用现有的统计分类技术,对待分类图像进行预分类,并检测出“不确定”像元。然后综合光谱、地理、土壤类型、早期判别结果、目视判读经验等各种知识和信息,充分发挥专家系统的推理判断能力,对“不确定”像元的类别作进一步判别,使得整幅图像的分类精度得到改善。并据此初步建立了一个土地利用的分类系统。试验证明,这种分类方法的精度比仅用单一多光谱信息的统计分类法(最大似然法)提高约8％。

An important research topic in remotely-sensed image processing and applicatoin is to improve the classification accuracy of TM Landsat image. In this paper, a survey is made on the current classification methods. A new method for TM Landsat image classification using rule-based expert system is discussed.In this method, an image is pre-classified by statistical classification method and "uncertain" pixels are detected. Then "uncertain" pixels are accurately using various types of knowledge and information such as spectrum, topography, soil map, the previous classification result and visual interpretation experiences etc. So the whole image has been classified more accurate-ly.Using this method, we have designed and implemented a preliminari land use classification system. Experiment proved that the accuracy of this new clasification method is 8 percent higher than that of the statistical classification (Maximum likelihood) which only uses multi-spectral information.