Environmental Fault Diagnosis of Solar Panels Using Solar Thermal Images in Multiple Convolutional Neural Networks

Table 4

Intermediate results for SqueezeNet, AlexNet, and GoogleNet.

Type of network	Epoch	Iteration	Time elapsed (hh:mm:ss)	Validation accuracy	Training accuracy	Validation loss	Training loss
SqueezeNet	1	1	00 : 00 : 26	17.3585	18.95	3.3108	3.2987
	2	75	00 : 07 : 07	92.376	91.9345	0.1751	0.06371
	3	155	00 : 14 : 10	100	100	0.0436	0.00668
	3 (end)	231	00 : 20 : 44	100	100	0.0267	0.01164
AlexNet	1	1	00 : 00 : 18	13.5849	12.5	3.7124	4.5314
	2	75	00 : 04 : 11	94.5783	99.875	0.00021	4.575e-05
	3	155	00 : 09 : 35	100	100	6.9009e-05	2.644e-05
	3 (end)	231	00 : 14 : 16	100	100	9.0214e-05	0.0008
GoogleNet	1	1	00 : 00 : 28	18.4905	19.3834	2.63963	3.1751
	2	75	00 : 08 : 07	96.9811	92.4764	0.13759	0.0445
	3	155	00 : 14 : 50	100	99.1528	0.02365	0.005
	3 (end)	231	00 : 23 : 00	100	100	0.0124	0.0026