论文部分内容阅读
提出了一种基于多点地统计学理论的遥感影像分类后处理方法。此方法从训练图像中提取先验的空间结构信息,推断类别的空间分布模式和相关关系,训练图像中能够建立包含空间关系的模型,比传统变异函数模型所表达的点对之间的关系更为丰富。将此方法应用于从Landsat TM影像中提取湿地类别,与空间平滑法和基于马尔科夫随机场的分类方法相比,其总体分类精度有所提高,且对曲线分布的地物类别的处理具有明显优势。
A remote sensing image classification and post-processing method based on multi-point statistical theory is proposed. This method extracts a priori spatial structure information from the training images and deduces the spatial distribution patterns and correlations of the categories. The training images can be used to establish a model that contains the spatial relationships, which is more relevant than the point pairs expressed by the traditional variation function models To be rich. This method is applied to the classification of wetlands extracted from Landsat TM images. Compared with the spatial smoothing method and Markov random field-based classification methods, the proposed method has improved the overall classification accuracy and the processing of the distribution of feature classes with curved distribution has obvious advantage.