{"title":"基于深度学习的全息存储相位检索抗噪声研究","authors":"Jianying Hao, Yongkun Lin, Mingyong Chen, Xiao Lin, X. Tan, Yuhong Ren","doi":"10.1117/12.2592674","DOIUrl":null,"url":null,"abstract":"Phase retrieval is the key technique in phase-modulated holographic storage. In this paper, a deep convolutional neural network is proposed to directly retrieve phase data. Compared with the traditional non-interferometric phase retrieval method, this method has the advantages of fast retrieval speed and high reconstruction accuracy. In this paper, the influence of intensity image noise on retrieval results under different retrieved conditions is researched and analyzed. By establishing a simulation system that is in strict agreement with real experiments, the lensless spatial diffraction images are generated. By adding different proportions of random noise into the intensity images we get the training dataset. The convolutional neural network is trained by a training dataset and tested by a new noisy test dataset. Experimental results show that the phase retrieval method based on deep learning has a high tolerance for systematic errors and strong anti-noise performance.","PeriodicalId":19391,"journal":{"name":"ODS 2021: Industrial Optical Devices and Systems","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2021-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Research on anti-noise of deep learning-based phase retrieval in holographic storage\",\"authors\":\"Jianying Hao, Yongkun Lin, Mingyong Chen, Xiao Lin, X. Tan, Yuhong Ren\",\"doi\":\"10.1117/12.2592674\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Phase retrieval is the key technique in phase-modulated holographic storage. In this paper, a deep convolutional neural network is proposed to directly retrieve phase data. Compared with the traditional non-interferometric phase retrieval method, this method has the advantages of fast retrieval speed and high reconstruction accuracy. In this paper, the influence of intensity image noise on retrieval results under different retrieved conditions is researched and analyzed. By establishing a simulation system that is in strict agreement with real experiments, the lensless spatial diffraction images are generated. By adding different proportions of random noise into the intensity images we get the training dataset. The convolutional neural network is trained by a training dataset and tested by a new noisy test dataset. Experimental results show that the phase retrieval method based on deep learning has a high tolerance for systematic errors and strong anti-noise performance.\",\"PeriodicalId\":19391,\"journal\":{\"name\":\"ODS 2021: Industrial Optical Devices and Systems\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2021-08-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"ODS 2021: Industrial Optical Devices and Systems\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1117/12.2592674\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"ODS 2021: Industrial Optical Devices and Systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1117/12.2592674","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Research on anti-noise of deep learning-based phase retrieval in holographic storage
Phase retrieval is the key technique in phase-modulated holographic storage. In this paper, a deep convolutional neural network is proposed to directly retrieve phase data. Compared with the traditional non-interferometric phase retrieval method, this method has the advantages of fast retrieval speed and high reconstruction accuracy. In this paper, the influence of intensity image noise on retrieval results under different retrieved conditions is researched and analyzed. By establishing a simulation system that is in strict agreement with real experiments, the lensless spatial diffraction images are generated. By adding different proportions of random noise into the intensity images we get the training dataset. The convolutional neural network is trained by a training dataset and tested by a new noisy test dataset. Experimental results show that the phase retrieval method based on deep learning has a high tolerance for systematic errors and strong anti-noise performance.