*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Shadowing Image Estimation and Object Detection in Radio Tomographic Imaging with Deep Learning
Committee:
Dr. Ma, Advisor
Dr. Stuber, Chair
Dr. Ying Zhang
Abstract: The objective of the proposed research is to take advantage of the power of deep learning to develop effective and accurate Radio Tomography Imaging (RTI) approaches to jointly reconstruct spatial loss field (SLF) images, detect objects, and handle the model mismatch problem. RTI problem is an ill-posed inverse problem with insufficient observations, and conventional RTI techniques with prior assumptions on SLF cannot achieve precise estimation. Moreover, traditional approaches cannot fully exploit the correlation of observations in the time domain and obtain satisfied performance for online estimation scenarios. Therefore, deep learning is considered in the RTI problem. However, deep-learning-based approaches require a large amount of data for the training stage and it is difficult to establish the true SLF in real-world experiments precisely, and thus to collect the labeled dataset for RTI in practice. Instead, we propose to design and train the deep-learning-based RTI scheme with the simulated labeled dataset generated from a pre-defined RTI system model. Since the actual RTI system model of the practical environment is infeasible to attain, the RTI model mismatch problem hinders the practical use of deep-learning-based RTI models. To mitigate the model mismatch effects, unsupervised transfer learning methods will be investigated to adjust the deep-learning-based RTI techniques in the testing environment. In the preliminary research, we find that deep-learning-based RTI schemes are more potent than conventional techniques when evaluating with simulated data, which indicates deep learning has great potential for the RTI task. With high-precise SLF estimates from deep learning approaches, we propose to stretch applications of estimated SLF, namely realizing object detection of shadowing images in addition to estimating SLF.
