Abstract: Shadow makes many computer vision tasks perform poorly, and the complex relationship between the shadow and scene makes it a research hotspot. Previous studies can be divided into two categories: 1) obtain shadow information by manual mask and remove it; 2) learn a direct mapping from shadow image to shadow-free image. The 1) method has a higher cost of manual annotation; the 2) method may interfere with non-shadow regions that should not be mapped. Therefore, the Separation and Reconstruction Network was proposed, a shadow removal method that uses an Independent Component Analysis-like method to separate the shadow features and the scene features from shadow images and realize shadow removal by differential mapping. This method focuses the primary mapping on shadow-related features while causing minimal interference in non-shadow areas, resulting in better shadow removal effects. Simulation results demonstrate that this paper's proposed shadow removal algorithm achieves competitive results compared to numerous state-of-the-art methods.