High-Speed Surface Roughness Recognition by Scattering on Terahertz Waves

Recognizing surface roughness constitutes a crucial aspect of both channel modeling and environmental perception. Despite the availability of numerous scattering analysis methods in terahertz band, most of them are geared towards low-velocity environments and are primarily analytical, rendering them ill-suited for performing effective classification tasks in high-speed scenarios with limited data accessible and greater environmental complexity. This paper proposes using deep learning methods to address the problem of scattering-based surface roughness recognition in fast scanning scenarios by terahertz frequencies. To achieve this, we build a terahertz high-speed sampling platform operating at 140GHz, design seven surfaces with varying degrees of roughness, and apply deep learning data processing method and a Long Sort-Term Memory (LSTM) model. We conduct experimental verification on low-speed and high-speed scenarios, propose and verify a down-sampling method that applies low-speed data to high-speed scenarios. Through experiments, we find that the deep learning method can achieve good results even in scenarios involving high speed and few sampling points. This method provides a reference for surface roughness recognition and further channel modeling in various scenarios such as vehicle scanning, shipborne, and even airborne and spaceborne scanning.