地学应用中的遥感图像处理若干问题的分析

REMOTE SENSING TECHNIQUE APPLIED IN GEOSCIENCE——A REVIEW OF ITS PRESENT DEVELOPMENT

遥感技术在地理学应用中是如何从遥感影像上直观、准确的得到所需的信息，为本专业服务。文章从地学应用部门在进行遥感影像处理时遇到的几个问题：多光谱数据的选取与合成；多源信息的复合；新型图像分类器的应用；专题提取的精度等方面的进展作了分析。

emote sensing technique has been widely applied in geoscience fields. How to get the thematic information more vividly and precisely is the main focus of geoscientists. The following four questions always emerge when scientists apply remote sensing technique in their research: 1) Multispectral data selection and combination; 2) Multisource information integration; 3) New image classifiers and their usage; and 4) Thematic extraction accuracy. The purpose of this paper is to review the present development on how to apply remote sensing technique in geoscience fields based on the above four questions.Methods on how to select the best bands for combination to get more useful information from remotely sensed data are surveyed by the author. The Optimum Index Factor (OIF) is more subjective and applicable compared with visual separability and experiences. In practice, the bands should be adjusted when time, location, and ground characteristics are changed. Data from sources other than remotely sensed data are needed in order to suit the user's requirements. According to our own experiences, integrating data from diversified sources is indispensable in processing remotely sensed data. These multiple sources may include: data from remote sensors of different platform, geographical information, expert knowledge, etc. Many new classifiers have appeared in addition to the traditional statistical classification methods. Spatial structure analysis, GIS assisted classification, fuzzy classification and accuracy assessment and neural network approaches in classification are such classifiers briefly analyzed in the paper. Classification accuracy is the focal problem that users care about. Many work has been done on how to improve the classification accuracy such as taking suitable sampling method, using complicated registration equation, integrating GIS in classification and others. Accuracy expression is another question which needs further study.