基于自编码器与多尺度空-谱特征编码的高光谱图像解混算法

Hyperspectral Image Unmixing Algorithm Based on Autoencoder and Multi-scale Spatial-Spectral Feature Encoding

摘要:

基于自编码器的高光谱解混方法或者过度关注高光谱图像中的空间信息或者过度关注光谱信息，忽略了空间信息和光谱信息的均衡提取。针对该问题，本文提出一种基于自编码器与多尺度空-谱特征编码的高光谱图像解混方法。该方法利用CNN编码器进行多尺度空间-光谱特征提取，Transformer编码器接收多尺度空间-光谱空间特征，利用子Transformer编码器和全局Transformer编码器解耦空间和光谱之间的依赖性。为了验证所提出方法的性能，在两个真实数据集上进行实验和对比。结果表明，所提出的解混算法可以提高高光谱图像解混精度.

Abstract:

Most hyperspectral unmixing methods based on autoencoders mainly focus either on the spatial information or on the spectral information of hyperspectral images, while neglect the balance of spatial information and spectral information. To address this issue, we propose a hyperspectral image unmixing method based on autoencoders and multi-scale spatial-spectral feature encoding. This method utilizes a CNN encoder for multi-scale spatial-spectral feature extraction. The Transformer encoder receives the multi-scale space-spectral features. It further utilizes sub-Transformer encoders and a global Transformer encoder to decouple the dependence between spatial and spectral information. Experimental analysis is conducted on two real datasets to validate the performance of the proposed method. The results demonstrate that the proposed unmixing algorithm can improve the accuracy of hyperspectral image unmixing.