To solve the problem that the traditional simultaneous localization and mapping (SLAM) algorithm will reduce the pose estimation accuracy and system robustness in dynamic environments, this paper proposes a visual-inertial SLAM system suitable for dynamic environments. This paper designs a plane segmentation network called scSE-PlaneNet, which is an encoder-decoder structure combined with an attention mechanism. scSEPlaneNet is used to provide static plane features for subsequent SLAM systems. Then, the error calculated according to the principle of plane homography is used to accelerate the pose optimization. We conduct experiments on multiple sequences of the ADVIO dataset, and the results show that our proposed method can obtain higher accuracy in camera pose estimation compared to the current state-of-the-art VINS-MONO system.