基于深度学习的高光谱图像多标签分类算法

A multi-label classification algorithm for hyperspectral image based on deep learning

提出一种基于深度学习的高光谱图像多标签分类算法。采用深度学习算法中的堆叠降噪自动编码器方法对每个像素的深层特征进行抽取,该方法可以有效表现高维特征空间中的非线性混合像素。使用多标签逻辑回归方法为每个像素预测并分配多个类标签。通过对合成数据和实际高光谱数据的大量对比实验,实验结果表明:该算法能够有效地为高光谱图像的像素精确地分配多类标签。

A multi-label classification algorithm was proposed for hyperspectral images based on deep learning.The deep features of each pixel were extracted by the Stacked Denoising Auto-Encoder（ SDAE） method of deep learning,which could effectively represent the nonlinear mixed pixels in a high-dimensional feature space. In addition,a multi-label logical regression method was used to predict and assign multiple class labels for each pixel. The experimental results on the synthetic and actual hyperspectral image datasets show that the proposed algorithm can accurately assign multiple class labels to the pixels of hyperspectral images.