Abstract: To meet the demand for dynamic sensing of geological structures ahead of roadways under intelligent coal mining conditions, this study addresses the limited signal extraction capability and unstable imaging performance of traditional virtual shot methods in strong noise environments. A seismic signal intelligent extraction and quality discrimination approach based on a multi-channel convolutional neural network (Multi-Channel CNN) is proposed. By integrating temporal features of multi-trace seismic signals with inter-channel correlation characteristics, the method enables adaptive denoising and enhancement of raw signals before cross-correlation, as well as automatic screening of high-quality trace gathers after cross-correlation. Field experiments conducted at the 61203 working face in the Jungar Coalfield demonstrate that the proposed method significantly improves first-arrival identification accuracy and imaging resolution of geological anomalies, reducing the average positioning error from 7.2 m to 3.0 m. The method effectively enhances the data quality and imaging reliability of seismic-while-excavation monitoring, showing strong engineering applicability and promotional potential.