Tunnel Lining Defect Identification Method Based on Small Sample Learning

Table 2

Identification accuracy table of tunnel defect data (excluding generated pictures).

Defect types	Val.mAP@.5	Test.mAP@.5	Label
Not dense	98.5%	90.4%	0
Void beneath	98.9%	86.8%	1
Cavity	98.5%	88.7%	2
Seepage	99.5%	77.6%	3
Seriously not dense	98.0%	70.3%	4
All	98.7%	82.8%	All