Abstract
Recently, convolutional neural networks (CNNs) have attracted enormous attention in pattern recognition and demonstrated excellent performance in hyperspectral image (HSI) classification. However, high-dimensional HSI dataset versus limited training samples is easy to cause the overfitting phenomenon in deep neural networks. Additionally, the intraclass distance of the embedding features extracted through the softmax-based CNNs may be greater than that of the interclass, which makes it difficult to further improve the classification accuracy. To address these issues, this article proposes a deep prototypical network with hybrid residual attention, which can effectively investigate the spectral–spatial information in the HSI. Specifically, in order to improve the generalization capability of the model, feature extraction with a hybrid residual attention module is presented to enhance the critical spectral–spatial features and suppress the useless ones in the classification task. Furthermore, a novel discriminant distance-based cross-entropy loss is proposed to increase the intraclass compactness, to obtain more superior results. Extensive experiments on three benchmark datasets are carried out to convincingly evaluate the proposed framework. With the generation of optimal prototypes representing each class and more discriminative embedding features, encouraging classification results are achieved compared with state-of-the-art methods.
Highlights
H YPERSPECTRAL image (HSI) classification, focusing on distinction of different materials through assigning an individual label to each pixel, has been widely applied in forest inventory, urban-area monitoring, and land-cover mapping
deep prototypical network with hybrid residual attention (DPN-hybrid residual attention (HRA)) is much faster than spectral–spatial residual network (SSRN) and fast dense spectral–spatial convolutional network (FDSSC); the reason relies on two respects: First, data volume is reduced by principal component analysis (PCA) preprocessing
A novel DPN-HRA model for HSI classification is proposed based on a deep prototypical network associated with HRA
Summary
H YPERSPECTRAL image (HSI) classification, focusing on distinction of different materials through assigning an individual label to each pixel, has been widely applied in forest inventory, urban-area monitoring, and land-cover mapping. To address this problem, a 3-D CNN was employed to simultaneously exploit the spectral–spatial features and achieved a better classification result [18], [19]. A dual-tunnel spectral–spatial attention network is proposed in [26] It consisted of a spectral attention long short-term memory branch and a spatial attention 2-D CNN branch to jointly extract critical spectral–spatial features, which is helpful to obtain competitive classification results. To cope with the aforementioned issues, in this article, we propose a novel deep prototypical network with hybrid residual attention (DPN-HRA) to achieve superior HSI classification.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call