论文部分内容阅读
多类别识别对于遥感图像分类的实用化具有重大意义。本文提出一种由多层神经网络与无监督分类相结合的复合分类方法。第一步用多层网络对几个大类进行有监督分类，第二步将网络输出作为无监督分类的输入，对遥感图像进行细分，使得可识别的类别数从原来的１０类提高到３０类。对ＳＰＯＴ遥感图像识别的结果表明，该算法能适应多类别识别任务的要求。
Multi-category recognition is of great significance for the practical application of remote sensing image classification. This paper presents a multi-layer neural network and unsupervised classification combined with the composite classification method. The first step is to supervise and classify several broad categories with multi-layer networks. The second step is to classify the output of the network as an unsupervised classification and subdivide the remote sensing images so that the number of identifiable categories can be increased from 10 30 categories. The results of SPOT remote sensing image recognition show that the algorithm can meet the requirements of multi-category recognition task.