提出了一种新的基于自相关的高光谱遥感图像纹理特征提取算法,该算法通过引入核函数技术,将单波段纹理窗口的空间自相关函数,扩展到多波段遥感图像的纹理描述。然后对特征矢量进行无监督C均值聚类实验和有监督RBF神经网络分类实验,在分类实验中确定了最佳窗口尺寸。实验结果表明,该文提出的自相关特征可以有效地描述高光谱遥感图像的纹理。 A new feature extraction algorithm based on autocorrelation for hyperspectral remote sensing image texture is proposed. This algorithm extends the spatial autocorrelation function of single-band texture window to the texture description of multi-band remote sensing image by introducing kernel function. Then, the unsupervised C-means clustering experiments and the supervised RBF neural network classification experiments were carried out on the feature vector to determine the optimal window size in the classification experiment. The experimental results show that the autocorrelation features proposed in this paper can effectively describe the texture of hyperspectral remote sensing images.