基于多模态特征融合的无人机小目标检测方法研究OA

Objective To propose a UAV small target detection method based on multi-modal feature fusion to meet the requirements for high precision and light weight.Methods Firstly,an illumination-aware progressive infrared-visible image fusion network was employed to adaptively integrate multimodal information,generating fused images to provide comprehen-sive features of small targets for subsequent detection.Secondly,an improved model was constructed based on YOLOv11,and for Backbone a parallelized patch-aware attention(PPA)module was introduced into the C3k2 module to enhance small target feature extraction,a shallow detail fusion module(SDFM)was added into the Neck module to strengthen multi-scale feature integration and a dynamic head(DyHead)was involved in the Head module to improve the detection robustness of the model for multi-scale targets.Finally,the proposed method was validated using the Multi-Scenario Multi-Modality Dataset(M3FD)dataset,and the improved model had its small object detection performance on fused images versus single-modality images compared,whose detection capabilities for fused images were evaluated against mainstream small target detection models.Results The improved model achieved an mAP50(mean average precision at an intersection-over-union ratio of 0.5)of 0.811 for fused images,representing a 3.7%and 1.6%improvement over single-modal visible light images(mAP 0.782)and infrared images(mAP 0.798),respectively.It gained advantages over mainstream small target detection models for fused images.Conclusion The proposed method achieves a balance between detection accuracy and lightweight design,providing an efficient solution for small target detection in battlefield reconnaissance and casualty search-and-rescue missions conducted by unmanned aerial vehicles.[Chinese Medical Equipment Journal,2026,47(1):1-7].