The absorption and attenuation of seismic waves in near-surface media(characterized by the quality factor Q) is recognized as a critical factor affecting seismic data quality. Traditional Q-value extraction methods based on uphole surveys are constrained by technical limitations including poor scalability, shallow detection depth, and high operational costs. To address these challenges, a comprehensive detection method integrating CPT(cone penetration testing) and Microtremor Survey was proposed, along with its field implementation workflow. A field implementation scheme was designed for multi-scale geophysical parameter fusion, through which spatially coordinated observations from uphole surveys, microtremor measurements, and CPT data were utilized to establish a multi-source heterogeneous near-surface parameter database. Concurrently, a time-series deep learning-based Q-field prediction algorithm was developed, employing multi-layer LSTM(long short-term memory) neural networks to construct nonlinear mapping relationships between uphole control points and regional geophysical parameters, thereby achieving Q-field reconstruction constrained by sparse uphole data. Practical applications demonstrated the method’s superior performance in seismic data processing within the Qinlan work area, where amplitude attenuation compensation based on the predicted Q-field resulted in effective bandwidth broadening, confirming the engineering value of this approach for near-surface attenuation compensation and excitation lithology optimization.