{"title":"数字辐射体识别的迁移学习方法","authors":"Qi Wang, G. Xiao","doi":"10.1109/PIERS-Fall48861.2019.9021893","DOIUrl":null,"url":null,"abstract":"A radiator consisting of half-wave dipoles that can characterize number 0 to 9 is simulated. The 10 numbers can be identified from their radiation fields without decoding. A deep neural network (DNN) model is trained on a large far-field dataset. This source model can still recognize well based on transfer learning methods even if the target data are obtained under other external conditions. The transfer learning methods of fine-tuning or freezing several layers of the source DNN model are verified, and the results are different on various target data. Some explanations are provided from the perspective of hierarchical structures of the source DNN model. Based on the method of feature matching, the features are extracted from the source model and the target model to verify the effects of transferring knowledge from source model to the target model.","PeriodicalId":197451,"journal":{"name":"2019 Photonics & Electromagnetics Research Symposium - Fall (PIERS - Fall)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A Transfer Learning Approach for Recognizing the Digital Radiator\",\"authors\":\"Qi Wang, G. Xiao\",\"doi\":\"10.1109/PIERS-Fall48861.2019.9021893\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"A radiator consisting of half-wave dipoles that can characterize number 0 to 9 is simulated. The 10 numbers can be identified from their radiation fields without decoding. A deep neural network (DNN) model is trained on a large far-field dataset. This source model can still recognize well based on transfer learning methods even if the target data are obtained under other external conditions. The transfer learning methods of fine-tuning or freezing several layers of the source DNN model are verified, and the results are different on various target data. Some explanations are provided from the perspective of hierarchical structures of the source DNN model. Based on the method of feature matching, the features are extracted from the source model and the target model to verify the effects of transferring knowledge from source model to the target model.\",\"PeriodicalId\":197451,\"journal\":{\"name\":\"2019 Photonics & Electromagnetics Research Symposium - Fall (PIERS - Fall)\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2019-12-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2019 Photonics & Electromagnetics Research Symposium - Fall (PIERS - Fall)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/PIERS-Fall48861.2019.9021893\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2019 Photonics & Electromagnetics Research Symposium - Fall (PIERS - Fall)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/PIERS-Fall48861.2019.9021893","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
A Transfer Learning Approach for Recognizing the Digital Radiator
A radiator consisting of half-wave dipoles that can characterize number 0 to 9 is simulated. The 10 numbers can be identified from their radiation fields without decoding. A deep neural network (DNN) model is trained on a large far-field dataset. This source model can still recognize well based on transfer learning methods even if the target data are obtained under other external conditions. The transfer learning methods of fine-tuning or freezing several layers of the source DNN model are verified, and the results are different on various target data. Some explanations are provided from the perspective of hierarchical structures of the source DNN model. Based on the method of feature matching, the features are extracted from the source model and the target model to verify the effects of transferring knowledge from source model to the target model.