Abstract
In hyperspectral image (HSI) classification, there are challenges of the spatial variation in spectral features and the lack of labeled samples. In this paper, a novel spatial residual blocks combined parallel network (SRPNet) is proposed for HSI classification. Firstly, the spatial residual blocks extract spatial features from rich spatial contexts information, which can be used to deal with the spatial variation of spectral signatures. Especially, the skip connection in spatial residual blocks is conducive to the backpropagation of gradients and mitigates the declining-accuracy phenomenon in the deep network. Secondly, the parallel structure is employed to extract spectral features. Spectral feature learning on parallel branches contains fewer independent connection weighs through parameter sharing. Thus, fewer parameters of the network require a lesser number of training samples. Furthermore, the feature fusion is conducted on the multi-scale features from different layers in the spectral feature learning part. Extensive experiments of three representative HSI data sets illustrate the effectiveness of the proposed network.
Highlights
Hyperspectral image contains abundant spatial and spectral information, which can provide a lot of useful information for image classification and target detection [1]
A large number of spectral dimensions provide a great amount of useful information, there are generally close correlations between spectral bands, especially adjacent ones, which lead to information redundancy
Supervised learning algorithms including k-nearest neighbor (KNN), support vector machine (SVM), convolutional neural network (CNN) and so on
Summary
Hyperspectral image contains abundant spatial and spectral information, which can provide a lot of useful information for image classification and target detection [1]. Supervised learning algorithms including k-nearest neighbor (KNN), support vector machine (SVM), convolutional neural network (CNN) and so on These methods use a set of labeled data as input and aim to train a model that can get corresponding output according to the input. The two-channel network separates spectral feature learning from spatial feature learning, which may lead to the loss of some useful spatial-spectral correlative information during the fusion processing between spectral features and spatial features To tackle this problem, the 3-D CNN was proposed, which takes raw HSI cube data without any pre-processing or post-processing as input [31]. In order to mitigate the declining-accuracy phenomenon and gain ideal classification performance, this paper proposes a spatial residual blocks combined parallel network (SRPNet) for HSI classification.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
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 callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
