“The high temporal resolution event camera provides a new development approach for traditional motion image deblurring tasks. However, current event driven motion image deblurring methods suffer from insufficient cross modal compensation mechanisms, high computational complexity of depth features, and lack of attention to multi-scale spatiotemporal information, which limits the deblurring generalization performance in complex scenes. To address these challenges, a dual channel Mamba network (DCM Net) is proposed. The method uses a dual channel cross modal Mamba module (DCCM), which projects events and blurry images into a shared latent feature space through a linear complexity state space model (SSM) hidden state mapping. Then, through a nonlinear cross gating structure, low-noise blurry image information is used to suppress event noise, and clear edge features of events are extracted and embedded into image features to achieve complementary fusion of cross modal features of events and blurry images, achieving the effect of deblurring. In addition, a pyramid channel attention module (PyCA) is proposed to extract multi-scale spatiotemporal information of features, guide the network to focus on key time channels, enhance the reconstruction of local fuzzy details in space, and further improve the restoration accuracy of potential clear image sequences. The experiment was conducted on the synthesized REDS (real and diverse scenes) dataset and the semi synthesized HQF (high quality frames) dataset, and compared with 11 methods. Compared with the DeMo IVF method, our method improved the peak signal-to-noise ratio (PSNR) of reconstructed sequences on the REDS dataset by an average of 0.16 dB, and the structural similarity index (SSIM) by an average of 0.003; On the HQF dataset, PSNR and SSIM showed an average improvement of approximately 0.11 dB and 0.002 dB, respectively; The learned perceptual image patch similarity (LPIPS) of the sequence reconstruction results on two datasets is optimal. In the subjective comparative experiment compared with five of the more advanced methods, our method achieved the best score. Conclusion: The method proposed in this article can combine fuzzy images and event data to reconstruct clear latent image sequences, demonstrating the effectiveness of the proposed network framework”