A Partial Hierarchical Model for Online Low-Resolution Wear Particle Images Classification

Table 6

The comparison results between SR-PHnet, AlexNet, ResNet50, and ShuffleNetV2.

Model	Class	TP	FP	FN	Precision	Recall	F1-score
Ours	Nonmetallic	55	6	2	0.902	0.965	0.932
	Sliding	78	18	20	0.813	0.796	0.804
	Cutting	80	0	14	1.000	0.851	0.920
	Fatigue	82	18	6	0.820	0.932	0.872
	Bubble	15	0	0	1.000	1.000	1.000
	Total sample	310	42	42	0.881	0.881	0.881
AlexNet	Nonmetallic	51	8	6	0.864	0.895	0.879
	Sliding	69	28	29	0.711	0.704	0.708
	Cutting	76	4	18	0.950	0.809	0.874
	Fatigue	69	30	19	0.697	0.784	0.738
	Bubble	15	2	0	0.882	1.000	0.938
	Total sample	280	72	72	0.795	0.795	0.795
ResNet50	Nonmetallic	54	6	4	0.900	0.931	0.915
	Sliding	75	20	23	0.789	0.765	0.777
	Cutting	81	1	13	0.988	0.862	0.920
	Fatigue	78	21	10	0.788	0.886	0.834
	Bubble	15	2	0	0.882	1.000	0.938
	Total sample	303	50	50	0.858	0.858	0.858
ShuffleNetV2	Nonmetallic	54	7	3	0.885	0.947	0.915
	Sliding	73	23	25	0.760	0.745	0.753
	Cutting	79	3	15	0.963	0.840	0.898
	Fatigue	75	22	13	0.773	0.852	0.811
	Bubble	15	1	0	0.938	1.000	0.968
	Total sample	296	56	56	0.841	0.841	0.841