用遥感手段对荒漠化进行监测是当前荒漠化研究的热点问题,传统的荒漠化遥感信息自动提取方法是基于光谱特征的图像分割,受多种因素的影响,分类精度的提高遇到瓶颈,因此基于知识的分类方法应运而生。CART是一种非参数化的分类与回归方法,在用于遥感影像自动分类时,可以方便地应用多源知识,提高分类精度。本文在分析了CART方法原理的基础上,针对荒漠化地区各种地物的特点,将包括地物光谱知识、纹理知识、植被盖度等在内的多种知识融入CART模型,克服了单纯利用光谱特征进行分类的不足,取得了85.94%的精度。

Remote sensing plays an important role in desertification monitoring. The traditional remote information of desertification automatic extraction method is image segmentation based on sensing spectral characteristics. Influenced by various factors, the improvement of class precision encounters barrier, and the knowledge-based classifying method comes into being. CART is a kind of non-parameter classification and regression method, which uses multi-source data to improve classification accuracy when used in automatic classification based on remote sensing images. After analyzing the principle of CART method, we put spectral knowledge, texture knowledge, and vegetation coverage together into CART model, the deficiency of simply using spectral characteristics is overcome, and an accuracy of 85.94% is obtained.