Multimodal Feature Fusion-Based Airborne-to-Ground Small Target Detection Network

In response to the difficulties of small target detection caused by differences in imaging principles between heterogeneous remote sensing images from different sensors, and the discrepancies in air-to-ground scenes, this paper proposes a small target detection method based on multimodal feature fusion for air-to-ground scenarios. The method extracts features from visible light and infrared images, uses a balanced multimodal matrix for feature fusion, and performs target detection using the fused feature map. A detection head specifically designed for small target detection is added to the YOLOv8s target detection network, enhancing the robustness of the detection network. Experimental results show that the proposed method improves the classification accuracy by 4.71% compared to the original YOLOv8s, effectively validating that the fusion of deep and shallow features can enhance the network’s perception ability for small air-to-ground targets.