多特征和多分类器组合的湿地遥感影像分类

摘要

为了适应湿地遥感影像分类,选择了湿地影像的典型特征,提出了一种组合多分类器的湿地遥感分类方法.提取湿地遥感影像的独立分量、纹理、湖泊透明度、归一化水体指数、绿度指数和湿度分量特征；选择样本对最小欧氏距离、光谱夹角填图、贝叶斯和支持向量机分类器进行训练学习.根据各分类器的混淆矩阵对其赋权值,检验样本是否满足正态分布；根据权值和假设检验结果构建组合分类器决策网络.实验表明该方法较传统湿地分类方法具有更好的性能和更高的精度.%Taking features of wetland's remote sensing image into account, typical feature selection is discussed. The independent component, texture, lake clarity, NDWI, GVI and WI of wetland image are extracted. The classifiers of minimum Euclidean distance, spectral angle mapper, Bayes and supporting vector machine are trained by sample respectively. Weights of every classifier are given by confusion matrices, and whether the sample meets normal distribution is tested. Multi-classifiers combination based on decision network is generated by weights and hypothesis test result. The experimental results show presented method has better performance and higher accuracy than traditional single-classifier method.