Abstract
This is an introduction to the method of classifying high medium and law yield farmlands by remote sensing and GIS, which is the result of a key project of the Scientific and Industry Technology committee of National Defense. In the study, special information related to high medium and law yield farmland was compounded with TM data. The development of the method of compound hierarchy classification improved accuracy of remote sensing classification greatly.

1. Introduction
Conventional remote sensing classification methods by computer auto interpret are supervised classification and unsupervised classification, including maximum likelihood classification minim distance classification and multi-group distinguish classification Although these methods are valued and can be used to distinguish common object, they have obvious weaknesses. Many mistakes and loses may be made in conventional computer auto interpret. These are because of complexities of interpret objects: same spectra date, for different object, same object for different spectra date. In conventional methods, computer auto classification relies only spectra date of remote sensing, much important space and geographic information as well as their relationships are neglected. A lot of knowledge of geographic experts can not be used in conventional methods. We know that they are very important for farmland classification.

Along with the develop0emnt of Geographic Information System (GIS) technology, geographic information play an important role in overcoming same spectra for different object to improve results of classification . In this study, considering multi factors and complexity of high middle and low yield farmland in the loess region, on the basis of statistic interpret, classification method of composite hierarchy were used. Multiple remote sensing and non-remote sensing information related to farmland yield, such as landform, elevation, soil salinity and irrigation water supply were composted with Landsat TM data. Visual interpreter's scientific knowledge was introduction Experts' though and ability of logical analysis were used to judge every map spot finally.

2. Analysis of Formation Factors of High Medium and Low Yield Farmland and Their Classification Method

2.1 Analysis of Formation Factor of High medium and low yield lands
Study area of this project was Dingxiang county, which is located in the north part of Shanxi province with total land area of 865 km2. Mountains surround the country in the east, north and couth. Alluvial and floods plains are located in the west . According to formation factors, the main types of medium and low yield farmlands are as follow:

1. Drought, the county is belong to the semiarid with an annual precipitation of 400 programme, which is not enough for crop growth. Irrigation or dry, and water supply play important roles to crop yields.
2. Soil salinity, soil containing much salinity which limits crop growth, mostly located in low plain.
3. Water loss and soil erosion, mountains and hills cover 55% of the total land of the Dixining county, water loss and soil erosion are main limits to crop yields in mountains and hills.

2.2 Discrimination of classification Factories
High- medium- low- yield farmland on remotely sensed images is based on using computer for distinguishing by auto model discriminate technology . Geographic information system has an important role in over-coming same spectra for different target, to improve result of classification.