DDN-SLAM is a real-time dense dynamic neural implicit SLAM system that integrates semantic features to address tracking drift and mapping errors in dynamic environments. The system introduces a feature point segmentation method combining semantic features with a mixed Gaussian distribution model to handle dynamic tracking interferences. It also proposes a mapping strategy based on sparse point cloud sampling and background restoration to avoid incorrect background removal. A dynamic semantic loss is introduced to eliminate dynamic occlusions. Experimental results show that DDN-SLAM achieves robust tracking and high-quality reconstructions in dynamic environments while preserving potential dynamic objects. Compared to existing neural implicit SLAM systems, DDN-SLAM shows an average 90% improvement in Average Trajectory Error (ATE) accuracy on dynamic datasets. The system uses a Gaussian mixture model to segment foreground and background, incorporating semantic information to distinguish dynamic and static objects. It also employs a mixed background restoration and rendering strategy, using feature-based optical flow to differentiate dynamic masks for background restoration. A rendering loss based on dynamic masks, including motion consistency loss, depth loss, and color loss, is introduced to constrain rendering artifacts of dynamic objects. The system is capable of real-time performance, low memory consumption, and high-quality geometric and texture details. It supports monocular, stereo, and RGB-D inputs and can stably track and reconstruct in dynamic and challenging scenarios at 20Hz. The contributions include the first dynamic SLAM system based on semantic features, a mixed background restoration and rendering strategy, and a rendering loss based on dynamic masks. The system is evaluated on multiple datasets, demonstrating superior performance in dynamic scenes compared to traditional and neural implicit SLAM methods.DDN-SLAM is a real-time dense dynamic neural implicit SLAM system that integrates semantic features to address tracking drift and mapping errors in dynamic environments. The system introduces a feature point segmentation method combining semantic features with a mixed Gaussian distribution model to handle dynamic tracking interferences. It also proposes a mapping strategy based on sparse point cloud sampling and background restoration to avoid incorrect background removal. A dynamic semantic loss is introduced to eliminate dynamic occlusions. Experimental results show that DDN-SLAM achieves robust tracking and high-quality reconstructions in dynamic environments while preserving potential dynamic objects. Compared to existing neural implicit SLAM systems, DDN-SLAM shows an average 90% improvement in Average Trajectory Error (ATE) accuracy on dynamic datasets. The system uses a Gaussian mixture model to segment foreground and background, incorporating semantic information to distinguish dynamic and static objects. It also employs a mixed background restoration and rendering strategy, using feature-based optical flow to differentiate dynamic masks for background restoration. A rendering loss based on dynamic masks, including motion consistency loss, depth loss, and color loss, is introduced to constrain rendering artifacts of dynamic objects. The system is capable of real-time performance, low memory consumption, and high-quality geometric and texture details. It supports monocular, stereo, and RGB-D inputs and can stably track and reconstruct in dynamic and challenging scenarios at 20Hz. The contributions include the first dynamic SLAM system based on semantic features, a mixed background restoration and rendering strategy, and a rendering loss based on dynamic masks. The system is evaluated on multiple datasets, demonstrating superior performance in dynamic scenes compared to traditional and neural implicit SLAM methods.