基于空间模糊纹理光谱的多光谱遥感图像分类方法

A Method for Classification of Multi-spectral Remotely Sensed Image Based on Spatial Fuzzy Texture Spectrum

为了充分利用各波段的纹理信息,针对遥感图像不同波段之间具有较大相关性的特点,提出了一种用空间模糊纹理光谱描述多光谱遥感图像纹理特征的方法。根据纹理特征具有多尺度的特性,对原始图像进行二次模糊纹理滤波,一次滤波采用平面三角隶属度函数,二次滤波采用空间距离代替平面距离形成滤波隶属度函数,其模糊滤波图像的隶属度分布称之为空间模糊纹理光谱。用FasART神经网络分类验证,实验结果表明,该方法具有较高的分类精度,尤其对纹理特征较为复杂的区域的分类效果更为明显。

In order to make full use of texture information of all spectral bands, and take into account the characteristic of significant correlation between the different bands, this paper presents a method to describe texture of multi-spectral remotely sensed image with spatial fuzzy texture spectrum. Based on muhi-scale character of texture, the method implements two fuzzy texture filtering algorithm, the first filtering uses triangle membership function, while the second uses spatial distance in place of plane distance. The spatial membership distribution of the fuzzy image is denoted as spatial fuzzy texture spectrum. With classification by FasART（fuzzy adaptive system. ART-based） networks, experimental results show that the proposed method has higher classification precision, especially to the complex texture area.