Fig 1.
Structure of the YOLOv8 model.
Fig 2.
MFDA-YOLO network structure.
Fig 3.
Structural diagram of the AIFI module.
Fig 4.
The SPD-Conv specific process when scale = 2.
Fig 5.
(a): C-OKM. (b): Omni-Kernel module. (c): DCAM. (d): FASM.
Fig 6.
The structure of the DADH.
Fig 7.
The principle of the task decomposition.
Table 1.
Ablation study on hyperparameters of WIoUv3.
Table 2.
Ablation experiment results of modules on the VisDrone2019-DET-Test.
Table 3.
Results of different models on the VisDrone2019-DET-Test.
Fig 8.
Confusion matrix of YOLOv8n.
Fig 9.
Confusion matrix of MFDA-YOLO.
Table 4.
Results of different models on the HIT-UAV.
Table 5.
Results of different models on the NWPU VHR-10.
Fig 10.
Comparison of detection results across different models on the Visdrone2019 dataset. (The black box demonstrates the MFDA-YOLO’s ability to reduce missed and false detections).
Fig 11.
Heat map comparison among different models on the HIT-UAV dataset. (The black bounding box highlights that MFDA-YOLO produces markedly more concentrated heat-maps on small objects).