一种基于光谱知识库的TM影像地物识别方法

A New Algorithm of Object Recognition Based on Spectral Library for TM Images

地物的波谱特征是识别地物的主要标志之一,在实际应用中发现,由于遥感过程中存在着复杂的大气辐射影响,常常造成光谱知识库中的地物光谱曲线同由遥感图像上获得的光谱曲线不相匹配,进而引起地物识别误差。针对这一问题,从曲线形状分析入手,定义了曲线的形状因子,提出对形状因子构成的曲线进行匹配分析的算法。将该方法应用于TM影像的地物识别,由于形状因子曲线较好地反映了地物的特征信息,它能自动地挖掘遥感图像的细节信息,对地物有较高的识别精度,试验区内总体识别精度达到88.96%。

The characteristics of spectrum are very effective to recognize different kinds of objects. The figures of different objects＇ reflectance curves are quite different. Based on comparison of reflectance curves＇ figures, the kinds of different objects could be correctly distinguished. In practice, because of the complex influence of atmosphere in radioactive transfer, the reflectance curves computed from remote sensing image are different from those measured from the earth＇s surface directly. For the reason, errors of the object recognition would be caused. The paper analyses the differences between the two kinds of reflectance curves and defines their shape parameters to reflect their characteristics of the figure . Based on the shape parameters and spectral library, a new algorithm has been carried to reduce the difference caused by atmospheric radiation, and recognize objects automatically. The shape parameter curve well describes the figure characters of objects, and more detailed information could be acquired by the new method. The algorithm has been applied for the object recognition of TM images,and the total accuracy reaches 88.96% in the test region.