基于特征构造的影像跨尺度分类研究

A scale-span classification approach for multispectral images based on feature construction

本文提出一种跨尺度分类方法,该方法立足于多尺度像斑模型,应用特征构造来实现跨尺度特征的构建,从而将最佳尺度选择问题隐含在特征构造中,不直接进行最佳尺度选择,避免了主观选择尺度的弊端。实验结果证明跨尺度分类方法一方面能减少特征维数空间,另一方面能充分利用尺度之间的纵向信息,较单一尺度分类能更准确地区分地物,提高分类精度。

The paper proposed a scale-span classification method based on multi-scale homogeneous-re- gion model. The method uses the feature construction of the data mining to fulfill the construction of scale- span features, and the best scale choice is implicit in the new constructive features, rather than directly carrying on the best scale choice with subjective errors. The experimental result proved that the new classi- fication method could not only reduce the dimension of the feature space, but also fully use the longitude information between different scales, so that it would distinguish objects more accurately than sole-scale classification, thus improve the classification precision.