Compressed sensing for terahertz body scanners

点击翻译按钮获取中文摘要

Over the last 20 years the terahertz (THz) region of the electromagnetic spectrum has developed from once being known as the THz gap into a very promising region for many applications. The most prominent one is security imaging, where THz body scanners are a vivid research field. For this application high quality, real-time or video-rate imaging capabilities are necessary. However, large-scale detector arrays are not readily available for THz radiation. This challenge is addressed by using a single or just a few very sensitive detectors in combination with mechanical scanning. Therefore, techniques which combine high sensitivity, single pixel detectors with fast, non-mechanical scanning are crucial for not only for security imaging but for all THz imaging applications. Since in many applications THz images are quite smooth it is likely that they are sparse in the Fourier and/or wavelet domain. In fact, first research results demonstrate this sparsity. Utilizing Compressed Sensing (CS) techniques can reduce the number of measurement acquisitions and thus decreasing the imaging time, because if only a single pixel is available the imaging time is directly proportional to the number of acquisitions. Therefore CS is a very promising technique to improve THz imaging significantly. The objective of the project is to develop a measurement design and CS reconstruction algorithms for THz body scanners in order to improve the image quality and the imaging speed of these scanners. The algorithms will be developed for a phase-sensitive passive THz body scanner and a THz radar body scanner. Both are single pixel cameras and capable of imaging at 5 m stand-off distance. The peculiarities of the THz spectral region require tackling several challenges within the CS framework. These are, for example, multi-path propagation, diffraction problems, speckle and coherence effects of the radiation. Among all the issues to be account for, the exact modeling of these phenomena is extremely complex, and hence simplified models need to be introduced for diverse aspects. This may result in severe model errors, i.e. errors in the sensing matrix. During the project we intend to solve these specific challenges using a compressive deconvolution approach. Furthermore, we want to develop a model of the THz imaging process with both scanners. The model will be verified by measuring the accessible model components. The project is a joined research effort by teams of TU Berlin and TU München. TU Berlin contributes its expertise in THz imaging systems and application of CS to THz imaging while TU München contributes its expertise regarding the application of CS to radar techniques.