超谱遥感图像的模糊最大似然分类研究

Fuzzy maximum likelihood classification of hyperspectral remote sensing image

分类识别是超谱遥感图像的重要研究领域.由于超谱图像空间分辨率低,像元混合的概率大,因此采用单纯的聚类或者监督分类都不能取得好的效果.为了提高超谱图像分类的精度,提出了模糊最大似然分类算法.先用模糊C-均值法对图像进行聚类,再在聚类结果的基础上,参考真实地物图,选择训练样本,用最大似然法进行最终的分类.实验结果表明,提出的算法由于在聚类的基础上选择监督分类的样本,因而获得了关于图像的更准确的信息,最终分类结果比模糊C均值聚类高出34.38%,比最大似然分类高出10.46%.

Classification is an important factor of hyperspectral remote sensing images. However, the low spatial resolution of hyperspectral images makes it is easy for pixel mixing to occur. Perfect classification is also made difficult because of unsupervised clustering and supervised classification. In order to improve the classification accuracy of hyperspectral images, a fuzzy maximum likelihood classification method is proposed. First, clustering of a hyperspectral image was done by fuzzy C-means algorithm. Train samples were selected by true ground map and the clustering results. Finally, classifying image was supervised maximum likelihood algorithm. The results show that the proposed method obtains more exact information of images because train examples were selected on the basis of both ground truth map and unsupervised clustering results. The final classification result by the proposed method was 34.38% greater than fuzzy Cmeans clustering and 10.46% greater than maximum likelihood classification alone.