Abstract: To enhance the enhancement effect of low light images in complex areas and solve the problem of texture loss caused by insufficient utilization of local and global features, this paper proposes the Spatial Feature Gated Restoration Network (SFGR Net). Firstly, the texture intensity estimator dynamically segments texture regions and flat regions based on gradient features; Secondly, improve the non local attention mechanism in the feature gating module and model long-range dependencies across regions to enhance the accuracy of detail reconstruction; Further design multi-scale residual blocks, integrating novel partial convolution, deep convolution, and ordinary convolution to enhance local feature representation, significantly suppressing noise interference on image restoration. The experimental results show that the proposed method achieved PSNR of 23.04dB, 23.06dB, and 26.92dB on the three standard datasets of LOL, SID, and SDSD, respectively, effectively preserving image details and demonstrating good enhancement effects. Moreover, under complex conditions, the model also has higher robustness and can be better applied to nighttime scenes.