A UAV object detection and spatial mapping framework integrating enhanced YOLOv7 with graph attention networks

Abstract

This study proposes a unified UAV perception framework that combines enhanced YOLOv7-based small-object detection with graph attention reasoning and spatial projection. The framework augments conventional single-stage detection with relation-aware context modeling while preserving real-time throughput at an input resolution of 640 × 640 on an NVIDIA RTX 3090 GPU. On VisDrone2019, the proposed method achieves 42.6% mAP@0.5 and 25.8% mAP@0.5:0.95, corresponding to gains of 4.8 and 4.2 percentage points over the YOLOv7 baseline. The corrected projection branch also consistently reduces RMSE across multiple flight-altitude scenarios, indicating that the integrated framework improves both detection reliability and mapping stability under the reported experimental conditions.