Attention mechanism in deep learning is similar to information selection mechanism, and the goal of attention is to select critical information for the current task. In hyperspectral classification, the distinction of some categories depends on the subtle differences, however, most of the classification methods have the problem of insufficient expression ability to discriminate the fine differences of categories. In this paper, a classification method based on group attention is proposed to enhance the difference of hyperspectral data between categories. Firstly, we slice the hyperspectral sample into several groups on spectral channels, and extract the group CNN features. Then we use the attention module to obtain the attention weights for each spectral group. Finally, the "feature recalibration" strategy is used to recalibrate the spectral group CNN features. The experiment show that the proposed approach can improve the classification accuracy of categories with subtle differences.
Attention can be interpreted as a method which allocates available computing power to the most informative part of the signal. In deep learning, attention mechanism also helps us to dig out the subtle information. In hyperspectral classification, the discrimination of some land cover types depends on the fine differences of hyperspectral, but most classification methods do not focus on the fine differences between hyperspectral categories. In this paper, a hierarchical group attention classification method is proposed to focus on the differences of categories from coarse to fine, therefore, the fine differences between categories can be obtained to achieve more accurate classification. For comparison and validation, we test the proposed approach with three other classification approaches on Salinas and Indian datasets, and the experiments demonstrate that our proposed approach can distinguish the spectral subtle differences of similar categories more accurately.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.