基于面向对象的无监督分类的遥感影像自动分类方法

为了实现无任何先验知识的高分辨率遥感数据的自动分类,并进一步提高自动分类精度和效率,提出了一种基于面向对象的无监督分类方法(Object Oriented Unsupervised Classification).具体步骤如下:首先对遥感影像进行分割,得到一系列空间上相邻、同质性较好的分割单元,然后对分割单元进行特征提取,得到分割单元的对象特征(光谱特征、纹理特征等多特征信息),进而对分割单元进行基于对象特征马氏距离聚类.最后,通过分类后处理(类别合并、错分类别调整等)得到最终的分类结果.通过实验表明:本文提出的方法不仅能够利用影像中更多的特征信息进行聚类而且还可以有效地减少聚类对象的个数,从而使自动分类的精度和效率都得到较大的提升.

A method of auto classification based on object oriented unsupervised classification

In order to achieve auto-classification for high resolution remote sensing image without any prior knowledge and further improve the efficiency and accuracy of automated classification,a new method of object-oriented unsupervised classification was presented in this paper.The detailed steps are as follows: First,the image was segmented into a series of segments which were composed of a cluster of contiguous and homogeneous pixels.Second,the object features of the segments were obtained(including spectral features,texture features and so on) through extracting features from the segments.And then by using the object feature we can cluster the segments using the Mahalanobis distance.Finally,through post-classification processing(including merging classes;adjusting misclassification),we can get the final classification results.The method was tested with an experiment and the result shows that the new proposed method could make use of more feature of image and reduce the number of cluster objects,so that both accuracy and efficiency of the auto classification have been much improved.