FA-YOLO: An Improved YOLO Model for Infrared Occlusion Object Detection under Confusing Background

Table 1

Result comparisons between different models.

Model	Precision (%)	Recall (%)	(%)	F1 (%)	FPS
SSD-VGG16	78.11	74.06	76.95	76.03	59.24
YOLOv3	84.24	80.66	84.53	82.41	40.02
Faster R-CNN-Resnet	70.76	90.74	88.30	79.51	9.35
YOLOv4 (no transfer)	82.20	74.06	79.24	77.92	40.72
YOLOv4 (transfer learning)	89.22	85.85	90.34	87.50	41.86
YOLOv4+NSF	96.21	81.06	91.92	87.98	40.74
YOLOv4+NSF+dilated CBAM	98.11	80.00	92.95	88.13	35.61