自适应模糊规则分类方法及在TM土地覆盖分类中的应用研究

根据自组织网络和模糊逻辑推理,实现土地覆盖自适应模糊规则分类方法。该方法通过网络的节点和权值提取出模糊规则,调整网络中节点个数(即相应增加规则节点数)和权值向量,使模糊规则自动生成,并利用模糊逻辑推理,完成TM土地覆盖分类。对拒分类的像元,自适应增加K值使其可分。该方法所得分类精度及Kapp系数与最大似然分类方法结果相比分别提高了2.7%和2.9%;与自组织网络相比,总精度相差不大,而Kapp系数低1%。实验证明,如何提取和表示非光谱知识,从而解决类别混淆等问题,是提高自适应模糊规则分类性能的关键

AN ADAPTIVE FUZZY RULE CLASSIFIER APPLIED TO LANDCOVER CLASSIFICATION OF TM

Based on self-organizing network and fuzzy logic reasoning, this paper discusses an adaptive fuzzy rule classifier for landcover classification. The fuzzy rules can be extracted from the nodes and weight vector of network which can adjust the node numbers (rule number accordingly) and weight vector. This classifier finished TMlandcover by fuzzy logic reasoning, and the unclassified pixels increase K adaptively to be classified; It improved 2.7% and 2.9% in overall accuracy and Kapp coefficient compared with MLC, deceased 1% in Kapp coefficient and no change in overall accuracy compared with self-organizing network. How to extract and express the non-spectral knowledge dissolved class confusion, is the key step to improve the classification.