A High-Efficiency Deep-Learning-Based Antivibration Hammer Defect Detection Model for Energy-Efficient Transmission Line Inspection Systems

Table 3

Performance of models on the same data set under the SC scheme.

Model	IoU = 0.5			IoU = 0.75			Time (ms)
	Nondefective (%)	Defective (%)	mAP (%)	Nondefective (%)	Defective (%)	mAP (%)
YOLOv4	98.64	98.32	98.48	86.96	92.65	89.81	48.9
RetinaNet	93.65	98.54	96.09	80.04	93.16	86.60	49.3
FCOS	94.06	96.15	95.10	81.75	90.38	86.06	44.9
CenterNet	93.39	96.61	95.00	80.96	86.18	83.57	40.7
Faster RCNN	93.33	97.96	95.64	81.92	94.97	88.32	62.0
Cascade RCNN	95.36	97.91	96.64	92.72	95.89	94.31	72.5
PI-Cascade RCNN	95.34	98.59	96.96	90.57	96.51	93.54	54.4