高光谱影像分类中纹理选择方法研究

Texture component selection method in hyperspectral image classification

用现有的纹理分析方法可提取多达几十种的纹理信息，但并非所有的纹理都适合用于遥感影像分类，不同纹理在遥感影像分类中的作用效果不同，目前缺少一种针对纹理信息参与分类所起作用的评价方法。提出一种纹理选择方法，将纹理添加到高光谱图像中，根据其所有像元隶属不同类别的概率熵的平均值，评价纹理参与高光谱图像分类的效果。实验证明，这种方法能够用于选择适合参与高光谱分类的纹理。

The current texture analysis methods can provide as many as dozens of kinds of texture information, but not all texture information is fit for remote sensing image classification, and there is not a quantitative method to evaluate the effect of different kinds of texture information. An effective method is proposed, which adds a texture component into the remote sensing image, and calculates the average entropy of the posterior probability of all the pixels belonging to different categories, to evaluate the effect of the texture component involved in image classification. The experiments prove that the method can be used to select suitable texture components to participate in image classification.