Dynamic objects in the scenes may cause blurring and bokeh of reconstruction. This Paper proposed a hybrid objects recognition for neural simultaneous localization and mapping (ORCO-SLAM) based on object recognition algorithm and Neural Radiance Fields (NeRF). Firstly, the YOLOv8 algorithm is employed to achieve object recognition in dynamic scenes, followed by the determination of the dynamic or static properties of detected objects. The model is fed an image devoid of the dynamic object, which has been removed from the view. Then the multi-resolution hash encoding for fast convergence and the multi-resolution compressed encoding are employed in ORCO-SLAM to fill the holes caused by removing dynamic objects in the images. Finally this article proposed an improved method for calculating the smooth loss based on the Co-SLAM method, which enhances the smoothness of scene representation and can predict scene information that is occluded by dynamic objects. The experiments show that ORCO-SLAM can construct dynamic scenes accurately avoiding the blurring and reconstructing virtualization which easily occurs with traditional SLAM methods in dynamic scenes. Besides, ORCO-SLAM runs accurate reconstruction and tracking performance in various datasets. The accuracy of camera pose estimation has improved campared to other methods.