利用BP算法进行新疆MODIS数据土地利用分类研究

土地利用信息获取已经成为遥感技术的重要应用领域之一。本研究以MODIS数据产品为信息源，用神经网络分类中的反向传播网络(BP)算法对新疆进行了土地利用分类研究。以新疆石河子为实验区，比较了BP算法与最大似然法的分类精度，前者的精度提高近10％。用BP算法对新疆全省进行了土地利用分类，统计分类结果各类别的面积和百分比，所得数据与相关部门公布数据非常接近。研究结果表明，250mMODIS数据对新疆地区的土地利用的类别信息获取比较实用：BP算法对中分辨率的MODIS数据能取得较高的分类精度。

Classification of Land Use in Xinjiang with BP Classifier Using MODIS Imagery

Land use in China has been in a continuous regulation since the 1990s.Currently,the speed of land use change is sped up,and the requirement about the information of land use situation and land use ways is also increased. Such information is of significance for rational management of land resources, planning of social and economic development,ecological conservation,achievement of sustainable utilization of natural resources.For remote sensing technology,land use is one of the most important application fields.BP neural network arithmetic was introduced for land use classification by using MODIS remote sensing data,on which a case study on land use in Xinjiang in carried out in this paper.In our experiment in Shihezi region,Xinjiang,the maximum likelihood classifier and BP neural network classifier are compared for their classification accuracy,the latter exceeds by about 10%.Moreover,the classification map of land use in Xinjiang is charted by using BP classifier.The areas and percent data of the classified types of land use are very close to the data publicized by a related environment department.The results confirm not only the practicability of 250m MODIS image in deriving the information of land use types,but also show that classifier a high classification accuracy can be achieved for 250m MODIS images by using the BP.