| Summary: | Active millimeter-wave imaging systems using complex-valued self-organizing map (CSOM) have potentially wide applications in moving-target imaging by acquiring and classifying complex textures. However, interference in the coherent observation causes harmful effects such as amplitude fluctuation and phase distortion, resulting in the deterioration of the visualization quality achieved by the CSOM. In this article, we propose the introduction of feature refinementZ using a complex-valued auto-encoder (AE) into the feature extraction process of our imaging system to suppress these effects. We show that the complex-valued AE is extremely useful for extracting features appropriately even under such adverse influences, and improves the clustering performance. We also experimentally investigate the influence of the number of hidden-layer neurons on the AE performance to discuss the robustness of our system.
|
|---|
