Multidimensional Tank Battle Damage Assessment Using YOLO-MobileNet with META Labeling on Russia-Ukraine Battlefield Imagery

Battle Damage Assessment (BDA) plays a vital role in maintaining combat effectiveness, yet existing approaches still depend on manual interpretation of UAV imagery, leading to delays and inconsistent results. This study presents a multidimensional tank damage assessment framework that integrates YOLOv8 and MobileNetV2 with META labeling to simultaneously evaluate damage type and severity. YOLOv8 performs real-time tank detection and region-of-interest extraction, while MobileNetV2, using a multitask classification structure, predicts both damage attributes through shared feature representations. Experiments using 1,596 UAV images from the Russia-Ukraine battlefield achieved mAP@0.5 = 0.921 for detection and 71.6 % classification accuracy (Macro F1 = 0.67) for damage assessment, running at 118 FPS. The proposed system demonstrates high efficiency and accuracy in real battlefield conditions and can be extended to explainable LLM-based BDA for maintenance reasoning and operational decision support.