高光谱半监督分类的标签约束弹性网图算法

LABEL CONSTRAINED ELASTIC NET GRAPH ALGORITHM FOR HYPERSPECTRAL IMAGES SEMI-SUPERVISED CLASSIFICATION

图半监督算法联合利用少量的标定样本与大量的无标签数据进行学习,可缓解高光谱图像的维数灾难问题,被广泛应用于高光谱图像分类,其核心在于如何确定图模型中各样本的近邻样本。提出一种高光谱图像半监督分类的标签约束弹性网图算法。充分利用有限的样本标签信息,通过顶点间的约束传递形成标签约束矩阵,进而针对每一顶点自适应选取符合标签约束的像素作为表示字典。通过在该字典上的弹性网表示来选择与其最为关联的近邻样本,构建高光谱图像的图表示模型,并基于该图模型实现高光谱图像的半监督分类。实验结果验证了该算法的有效性,相比多个半监督算法,分类准确性更高。

The graph semi-supervised algorithm combines a small number of labeled samples with a large amount of unlabeled data to learn,which can alleviate the dimensionality disaster of hyperspectral images.It is widely used in hyperspectral images classification,and its core is how to determine the neighbor samples of each sample in the graph model.This paper proposes a label-constrained elastic network graph algorithm for hyperspectral images semi-supervised classification.The given label information was fully utilized,and the label constraint matrix was formed by the constraint propagation between the vertices.Then,for each vertex,the pixels that meet the label constraint were adaptively selected as the representation dictionary of the underlying vertex.The nearest neighbors of each vertex could be found by selecting the mostly related vertices in its elastic net representation upon the associated dictionary,and the semi-supervised classification of hyperspectral images was realized based on the constructed graph model.The experimental results verify the effectiveness of the proposed algorithm,and it has higher classification accuracy than multiple semi-supervised algorithms.