Transfer Learning for CNN-Based Damage Detection in Civil Structures with Insufficient Data

Table 2

The number of the utilized samples and the results of the training process for the three structures (stage 1).

Structure	Number of training samples	Number of validation samples	Number of test samples	Test accuracy	Test loss	Total training time
The BHM benchmark model	2457	615	1024	1	0.0008143	0 : 00 : 59
QU grandstand simulator	628	158	262	1	7.41E-05	0:00:15
The Tianjin Yonghe bridge	10398	2592	4320	0.99166	0.02361	0 : 04 : 03