Adaptive millimeter-wave security imaging based on complex-valued neural networks

在这个项目中，我们取得了以下三个成果。(1)提出了一种新的（复值）自组织映射（CSOM）体系结构，即CSOM，该体系结构根据互信息调制SOM的动态，从而实现多模态信息（即空间和频域纹理信息）的有机集成。(2)构建了一种八元一维毫米波线性锥形槽天线（LTSA）阵列，直接连接并联前端电路实现包络相位检测（EPD），使系统前端实现空间平行。(3)我们评估了由(1)和(2)组成的系统，发现CSOM即使对看似几乎随机的原始观测数据也能产生成功的可视化输出。

In this project, we made the following three achievements.(1) We proposed a new(complex-valued) self-organizing map(CSOM) architecture, namely CSOM, in which the SOM dynamics is modulated in accordance with the mutual information, resulting in organic integration of multi-modal information, i. e., spatial-and frequency-domain texture information in the present millimeter-wave imaging system case.(2) We constructed an eight-element one-dimensional millimeter-wave linearly-tapered-slot antenna(LTSA) array, directly connected to parallel front-end circuit to realize the envelope phase detection(EPD), resulting in spatial parallelism of the system's front-end.(3) We evaluated the system consisting of(1) and(2), and found that the CSOM yields successful visualization output even for seemingly almost random raw observation data.