서브메뉴
검색
상세정보
Radio Tomographic Imaging with Deep Learning.- [electronic resources]
Radio Tomographic Imaging with Deep Learning.- [electronic resources]
상세정보
- 자료유형
- 학위논문(국외)
- 자관 청구기호
- 기본표목-개인명
- 표제와 책임표시사항
- Radio Tomographic Imaging with Deep Learning. - [electronic resources]
- 발행, 배포, 간사 사항
- 발행, 배포, 간사 사항
- 형태사항
- 141 p.
- 일반주기
- Source: Dissertations Abstracts International, Volume: 87-05, Section: A.
- 일반주기
- Advisor: Ma, Xiaoli.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2023.
- 요약 등 주기
- 요약Radio tomographic imaging (RTI) is a promising passive computational imaging technique for reconstructing attenuation images induced by objects in wireless networks. The attenuation images can be applied to realize passive object localization, environment monitoring, and even through-wall imaging in security systems. In RTI systems, the attenuation map is also referred to as the space loss field (SLF), which quantifies the attenuation rate of the radio frequency (RF) waves at each location in the imaging area covered by wireless networks. Any object on a signal propagation path causes signal attenuation on received signal strength (RSS) measurements between wireless transceiver pairs due to shadowing effects. The RSS attenuation can be modeled as the 2-dimensional (2D) integral of the SLF scaled by a weighting function. This principle underpins RTI techniques and enables the estimation of the SLF from the RSS measurements.This thesis aims to take advantage of the power of deep learning to develop effective and accurate RTI schemes to deal with one-shot and online imaging scenarios while considering diverse wireless environments. The RTI problem is generally ill-posed with insufficient observations, and traditional techniques that rely solely on optimization algorithms cannot achieve precise reconstruction caused by inaccurate prior assumptions on the SLF images and the noise. Compared to the conventional approaches, deep-learning-based RTI techniques are data-driven methods that can directly learn prior information from data to avoid making unreliable assumptions.Specifically, we design an attention-augmented optimization SLF estimation scheme using both deep learning and traditional optimization algorithms to refine the blurry SLF reconstruction in the one-shot estimation scenario. In addition to the one-shot RTI approach, we harness the temporal correlations across RSS observations at neighboring timesteps to develop an online SLF estimator that leverages past RSS measurements and SLF estimations. To further enhance the SLF imaging performance, we investigate the discriminability of the RTI scheme and exploit the state-of-the-art object detection and semantic segmentation models from the area of computer vision. This leads to the development of a fully deep-learning-based single-shot RTI approach, incorporating the U-Net structure to attain high-resolution SLF recovery. This RTI scheme is further extended to the online scenario. Finally, we address the challenge of model mismatch in the RTI system, where the mathematical model relating SLF and RSS may not perfectly align with real-world conditions. To mitigate this issue, we devise a deep unsupervised domain adaptation (UDA) module to adapt our high-resolution one-shot and online RTI schemes to actual environments while preserving imaging performance to the maximum extent possible. Collectively, these proposed strategies form a comprehensive investigation of the RTI problem from a deeplearning perspective.
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 주제명부출표목-일반주제명
- 부출표목-단체명
- 기본자료저록
- Dissertations Abstracts International. 87-05A.
- 전자적 위치 및 접속
- 원문정보보기
MARC
008260219s2023 us s 000c||eng d■001000017360605
■00520260202105554
■006m o d
■007cr#unu||||||||
■020 ▼a9798265401861
■035 ▼a(MiAaPQ)AAI32315852
■035 ▼a(MiAaPQ)GeorgiaTech76829
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620
■090 ▼a전자자료
■1001 ▼aHe, Ziyan.
■24510▼aRadio Tomographic Imaging with Deep Learning.▼h[electronic resources]
■260 ▼a[S.l.]▼bGeorgia Institute of Technology. ▼c2023
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2023
■300 ▼a141 p.
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: A.
■500 ▼aAdvisor: Ma, Xiaoli.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2023.
■520 ▼aRadio tomographic imaging (RTI) is a promising passive computational imaging technique for reconstructing attenuation images induced by objects in wireless networks. The attenuation images can be applied to realize passive object localization, environment monitoring, and even through-wall imaging in security systems. In RTI systems, the attenuation map is also referred to as the space loss field (SLF), which quantifies the attenuation rate of the radio frequency (RF) waves at each location in the imaging area covered by wireless networks. Any object on a signal propagation path causes signal attenuation on received signal strength (RSS) measurements between wireless transceiver pairs due to shadowing effects. The RSS attenuation can be modeled as the 2-dimensional (2D) integral of the SLF scaled by a weighting function. This principle underpins RTI techniques and enables the estimation of the SLF from the RSS measurements.This thesis aims to take advantage of the power of deep learning to develop effective and accurate RTI schemes to deal with one-shot and online imaging scenarios while considering diverse wireless environments. The RTI problem is generally ill-posed with insufficient observations, and traditional techniques that rely solely on optimization algorithms cannot achieve precise reconstruction caused by inaccurate prior assumptions on the SLF images and the noise. Compared to the conventional approaches, deep-learning-based RTI techniques are data-driven methods that can directly learn prior information from data to avoid making unreliable assumptions.Specifically, we design an attention-augmented optimization SLF estimation scheme using both deep learning and traditional optimization algorithms to refine the blurry SLF reconstruction in the one-shot estimation scenario. In addition to the one-shot RTI approach, we harness the temporal correlations across RSS observations at neighboring timesteps to develop an online SLF estimator that leverages past RSS measurements and SLF estimations. To further enhance the SLF imaging performance, we investigate the discriminability of the RTI scheme and exploit the state-of-the-art object detection and semantic segmentation models from the area of computer vision. This leads to the development of a fully deep-learning-based single-shot RTI approach, incorporating the U-Net structure to attain high-resolution SLF recovery. This RTI scheme is further extended to the online scenario. Finally, we address the challenge of model mismatch in the RTI system, where the mathematical model relating SLF and RSS may not perfectly align with real-world conditions. To mitigate this issue, we devise a deep unsupervised domain adaptation (UDA) module to adapt our high-resolution one-shot and online RTI schemes to actual environments while preserving imaging performance to the maximum extent possible. Collectively, these proposed strategies form a comprehensive investigation of the RTI problem from a deeplearning perspective.
■590 ▼aSchool code: 0078.
■650 4▼aWireless networks.
■650 4▼aTomography.
■650 4▼aDeep learning.
■650 4▼aComputer vision.
■650 4▼aMagnetic resonance imaging.
■650 4▼aSignal processing.
■650 4▼aNeural networks.
■650 4▼aAdaptation.
■650 4▼aDesign.
■650 4▼aOptimization algorithms.
■650 4▼aVisualization.
■650 4▼aComputer science.
■650 4▼aElectrical engineering.
■650 4▼aMedical imaging.
■690 ▼a0389
■690 ▼a0800
■690 ▼a0984
■690 ▼a0544
■690 ▼a0574
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05A.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2023
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360605▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
![Radio Tomographic Imaging with Deep Learning. - [electronic resources]](/Sponge/Images/bookDefaults/DDbookdefaultsmall.png)


한글
ENG
日本
中文
Việt Nam