论文部分内容阅读
提出一种基于分层稀疏表示特征学习的方法对高光谱图像进行分类,首先在空间邻域内采用判别式字典学习得到样本的稀疏编码.然后用空间金字塔平方和根池化方法得到每层的空间特征.最后将该空间特征与中心像素的光谱特征相结合形成最后的判别特征,用极限学习机做分类.在两个高光谱数据集上的实验结果表明该方法能够得到较好的分类精度
This paper proposes a method of feature extraction based on hierarchical sparse representation to classify hyperspectral images.First, the discriminant dictionary is used to learn the sparse coding of the samples in the spatial neighborhood.Finally, the spatial pyramid root summation method is used to get the space of each layer Finally, this spatial feature is combined with the spectral features of the central pixel to form the final discriminant feature, which is classified by the extreme learning machine.Experimental results on two hyperspectral data sets show that this method can obtain better classification accuracy