Abstract: Objectives: Shale gas extraction activities in the southern Sichuan region have led to frequent induced seismic events. Accurate and timely monitoring of these seismic activities is essential to understand the triggering mechanisms and mitigate associated hazards. This study aims to develop and validate a deep learning-based InSAR method to enhance deformation extraction and noise suppression for seismic monitoring induced by shale gas extraction in the southern Sichuan region. Methods: We propose a novel InSAR denoising and deformation recovery framework based on deep learning. The method adopts an encoder-decoder architecture that integrates three-dimensional convolutional neural networks (3D CNN), two-dimensional convolutional neural networks (2D CNN), residual connections, and channel attention mechanisms. The model is trained using a simulated dataset, and its performance is evaluated under different signal-to-noise ratios. Results: The performance of the proposed model was evaluated using Root Mean Square Error (RMSE) and the Structural Similarity Index (SSIM). On the test dataset, the trained model achieved an RMSE of 1.27 mm and an SSIM of 0.74, indicating strong reconstruction capability. Experiments under varying signal-to-noise ratios (SNRs) demonstrated that the model maintains robust performance when SNR exceeds 0.1. On real-noise datasets, the proposed approach outperformed the traditional Stacking method, achieving approximately 40% improvement in deformation recovery accuracy. Furthermore, when SNR>10^-0.84 (~0.14), the model successfully reconstructed coseismic deformation fields with peak displacements greater than 10 mm. Applying the method to the 2021 Luxian Mw 5.4 earthquake and the 2019 Zizhong Mw 4.9 earthquake further validated its effectiveness, yielding high-precision deformation fields superior to those obtained by the conventional Stacking method. ^Conclusions: The proposed deep learningbased InSAR framework provides an effective solution for monitoring small to moderate induced earthquakes in shale gas regions. Its ability to extract accurate coseismic deformation fields under high-noise environments offers valuable technical support for hazard assessment, seismic mechanism analysis, and rapid post-event emergency response. This study also provides a conceptual framework and practical insights for applying deep learning to InSAR-based earthquake monitoring.