Multimodal deep learning for solar radio burst classification. (January 2017)

• Record Type: Journal Article
• Authors: Ma, Lin; Chen, Zhuo; Xu, Long; Yan, Yihua
• Is Part Of: Pattern recognition. Volume 61(2017:Jan.)
• Page Start: 573
• Page End: 582
• Publication Date: 2017-01
• Subjects: Multimodal learning -- Solar radio spectrum -- Classification; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.04.013
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: In this paper, multimodal deep learning for solar radio burst classification is proposed. We make the first attempt to build multimodal learning network to learn the joint representation of the solar radio spectrums captured from different frequency channels, which are treated as different modalities. In order to learn the representation of each modality and the correlation and interaction between different modalities, autoencoder together with the structured regularization is used to enforce and learn the modality-specific sparsity and density of each modality, respectively. Fully connected layers are further employed to exploit the relationships between different modalities for the joint representation generation of the solar radio spectrums. Based on the learned joint representation, solar radio burst classification is performed. With the validation on the constructed solar radio spectrum database, experimental results have demonstrated that the proposed multimodal learning network can effectively learn the representation of the solar radio spectrum, and improve the classification accuracy. Abstract : Highlights: Multimodal deep learning for solar radio burst classification is proposed. AE together with the structured regularization is used to enforce and learn the modality-specific sparsity and density of each modality. The proposed network can effectively learn the representation of the solar radio spectrum.