结合RBF网络与光谱指数的遥感分类应用研究

Application Study on Remote Sensing Classification Based on RBF Network and Spectral Indexes

本文基于光谱波段信息,提取NDWI、NDVI、NDBI三种归一化指数,作为城市地区土地利用分类的关键辅助信息。在此基础上,提出基于RBF网络与归一化指数的城市遥感分类模型。最后,以四川南充市为研究区域,以TM影像为数据源,对本文提出的城市地区分类模型进行了分类实验。实验结果表明,RBF在融合地学参数方面具有一定的优势,基于RBF神经网络与地表指数进行分类,能获得95.02%的较为理想的总体分类精度,比单纯利用波段信息进行分类其精度提高了7.05个百分点。

In this paper, spectral indexes NDWI, NDVI, NDBI were inversed from TM images as the key aux-iliary information in the classification of city land use. On this basis, the city remote sensing classi-fication model was put forward based on RBF network and the normalized indexes. Finally, taking the Sichuan Nanchong city as the study area, using TM image as data source, the city classification model proposed in this paper was experimented. The experimental results show that RBF network has a certain advantage in the integration of learning parameters. The overall accuracy using RBF neural network and the surface indexes can reach to 95.02%, which is improved by 7.05 percentage points than only using the band information.