Automobile Component Recognition Based on Deep Learning Network with Coarse-Fine-Grained Feature Fusion

Table 2

Performance comparison of classical networks.

Model	Training	Testing		Evaluation		Evaluation with Algorithm 1
	Dataset	Loss	Acc. (%)	Loss	Acc. (%)	Loss	Acc. (%)
VGG11	DATA_ORI	1.2754	82.03	7.4869	23.78	6.9467	36.21
	DATA_CON	1.0606	81.38	4.2180	33.51	3.933	51.35
VGG19	DATA_ORI	1.0462	83.66	7.5684	28.10	7.086	32.29
	DATA_CON	0.8374	66.67	2.3672	21.08	1.7511	34.59
ResNet18	DATA_ORI	0.3248	93.31	8.7534	31.35	7.2648	42.70
	DATA_CON	0.2903	91.19	13.6997	24.86	5.661	52.43
ResNet152	DATA_ORI	1.4845	71.71	7.8313	29.18	7.2735	30.01
	DATA_CON	1.5444	72.28	277.2992	19.45	142.59	28.11
DenseNet121	DATA_ORI	0.6082	85.33	3.9679	24.32	3.7571	30.19
	DATA_CON	0.568	85.48	14.5865	20.54	10.485	34.05
DenseNet201	DATA_ORI	0.5374	87.55	9.1101	32.43	8.1152	43.78
	DATA_CON	0.3379	77.69	11.0234	25.41	6.0151	47.03

The best number performed in these comparisons tested.