基于非线性神经网络的高清晰高光谱遥感图像分类器设计与应用

对于高光谱和空间分辨率的遥感图像而言，它具有较为复杂的地物判读特性，应用常规的监督或非监督分类方法难以达到理想的结果。为此设计了一种非线性BP网络分类器，它将纹理结构特征与地物光谱特征相结合．针对上海市某地区的卫星遥感图像．在ENVI／IDL平台上与K-means的非监督分类和最小距离的监督分类方法进行了分类的对比应用。试验结果表明．该分法较好地考虑了图像的光谱特征，能有效地提高类别辨识的精度。

Study on the Design of Remote Image Classifier with HighResolution and Multi-spectrum Based on Neural Net

As High resolution and multi-spectrum remote image is of complex objects judgement characters, it will be of no ideal results with general supervised or unsupervised classifiers. A BP neural net classifiers has been put forward, which combines the texture and spectral features of object. This method is applied on the classification of some area image of Shanghai. In addition, it is also compared with K-means unsupervised and minimum distance classifier on ENVI/IDL platform. Results show that the method take the image spectral features into consideration completely and can improve the precision of image classification effectively.