2022年第83回応用物理学会秋季学術講演会

講演情報

一般セッション(口頭講演)

CS コードシェアセッション » 【CS.2】 3.2 情報フォトニクス・画像工学(旧3.3)、4.4 Information Photonicsのコードシェア

[21a-C205-1~6] CS.2 3.2 情報フォトニクス・画像工学(旧3.3)、4.4 Information Photonicsのコードシェア

2022年9月21日(水) 10:00 〜 12:00 C205 (C205)

堀崎 遼一(東大)

10:30 〜 10:45

[21a-C205-2] Complex Amplitude Demodulation Based on Deep Learning in Holographic Data Storage

〇(D)Jianying Hao1,2、Xiao Lin1、Ryushi Fujimura3、Yoshito Tanaka2、Soki Hirayama2、Xiaodi Tan1、Tsutomu Shimura2 (1.Fujian Normal Univ.、2.Tokyo Univ.、3.Utsunomiya Univ.)

キーワード:Holographic Data Storage, Computational Imaging, Deep Learning

Due to the characteristic of higher storage capacity, faster transmitting rate and longer storage life, holographic data storage has been becoming a powerful candidate in the era of big data. To continue improving the coding rate, complex amplitude encoding method is proposed. However, to retrieve the complex amplitude information from the reconstructed beam is a key technique, since phase cannot be detected directly. In this paper, a complex amplitude demodulation method based on deep learning is proposed. Both the amplitude and phase of light are used to encode the information data. The lensless near field diffraction intensity images corresponding to data pages were captured by simulation. Two convolutional neural networks were used to establish the relationship between the intensity image and the phase data page and the amplitude data page respectively. After training the CNNs, the encoded data pages can be directly retrieved through the trained CNNs. Compared with traditional holographic data storage method, the method proposed in this paper has multi dimensional modulation characteristics, which can greatly increase the data storage density of holographic storage.