Tunnel Lining Defect Identification Method Based on Small Sample Learning

Table 3

Identification accuracy table of tunnel defect data (include generated images).

Defect types	Val.mAP@.5	Test.mAP@.5	Label
Not dense	98.0%	91.4%	0
Void beneath	99.0%	88.2%	1
Cavity	98.9%	89.3%	2
Seepage	97.6%	79.4%	3
Seriously not dense	97.8%	87.6%	4
All	98.3%	87.2%	All