This study proposes an evaluation technology for local battlefield situation reconstruction based on deep learning algorithms. By combining UE5 and Cesium Native, utilizing high-resolution cameras and sensors of unmanned aerial vehicles to collect actual geographical locations and environmental features, the proposed 3DGS-COLMAP deep learning method is employed for 3D reconstruction, constructing a virtual reality scene of the local battlefield. With VR devices and related operations, the system can dynamically adjust the status and behavior of objects in the virtual scene according to real-time data, providing immersive observation, analysis, evaluation, and command decision-making in the battlefield environment, thereby enhancing the interactivity and realism of combat training.