<p>Terahertz (THz) communication systems are anticipated to play a pivotal role in enabling various emerging technologies such as the Internet of Things and cloud computing, offering wide bandwidth and low latency for efficient and reliable data transmission. However, the sub-millimeter wavelength of THz waves leads to considerable scattering effects when interacting with rough surfaces. Therefore, an accurate scattering model for rough surfaces is crucial for the reliable characterization of THz wireless channels. In this paper, we first analyze the properties of the equivalent currents (electric and magnetic currents) on the dielectric rough surface with different conditions in the THz band; then a deep learning architecture proposed in previous work is extended to generate equivalent electric and magnetic currents. Additionally, a scattering modeling method is presented for rough dielectric surfaces combined with the physical optics approximation. In the future, the proposed model will be incorporated into the ray-tracing-based THz channel modeling framework to assist the development and deployment of THz communication devices.</p>