Table 1.

Effects of CNN depth and data augmentation according to 12 measurement conditions and results of the A-Test for evaluating structural risks.

		No Aug				Aug				EN
		VGG-11	VGG-13	VGG-16	VGG-19	VGG-11	VGG-13	VGG-16	VGG-19
Train	All	97.46	97.33	97.46	97.19	95.32	95.08	95.27	95.33	-
Test	Sit (R)	97.20	97.14	97.57	97.41	93.65	94.22	93.45	95.16	96.37
	Stand (R)	97.67	97.73	98.06	97.94	93.86	94.34	93.83	94.06	96.06
	Sup (R)	97.30	97.31	97.79	97.83	93.29	93.07	92.77	94.25	96.02
	Sit (A)	97.75	97.67	97.92	97.95	93.89	94.54	94.95	92.74	95.33
	Stand (A)	97.55	97.53	98.01	97.71	94.11	94.98	95.12	93.11	95.56
	Sup (A)	98.16	98.09	98.49	98.21	95.15	95.34	95.82	95.55	97.02
	Walk (3.2)	95.19	95.20	95.11	94.79	92.45	91.83	92.07	91.90	93.51
	Walk (4.5)	94.49	94.52	94.72	94.54	94.37	92.78	94.05	94.88	94.80
	Walk (5.8)	94.54	94.34	94.21	94.15	95.4	94.26	95.01	95.60	94.91
	Run (6.4)	93.54	93.31	93.29	93.36	94.69	93.61	94.47	95.03	94.16
	Run (8.5)	92.83	92.35	92.57	92.17	95.09	94.98	94.80	95.11	93.84
	Run (10.3)	92.48	92.20	92.05	91.70	94.62	95.18	94.85	94.93	93.49
	All	95.72	95.62	95.82	95.65	94.21	94.09	94.27	94.36	95.09
A-Test	$\widehat{{\mathbf{\tau }}_{\mathit{n}}}$	7.95	7.83	7.61	7.57	8.40	8.06	7.88	7.81	6.98
	$\mathbf{min}{\tau }_{\mathit{n}}$	3.11	2.97	2.43	2.44	5.21	5.10	4.67	4.82	2.85

No Aug = No Augmentation; Aug = Augmentation; EN = Ensemble Network. Bolded numbers are the highest accuracy according to 12 measurement conditions in No Aug and Aug models, respectively. Red numbers are the highest accuracy according to 12 measurement conditions in all models. Optimal architecture is the ensemble network of VGG-16 without data augmentation and VGG-19 with data augmentation.