Semantic-Segmentation-Based Rail Fastener State Recognition Algorithm

Table 1

Detection and classification results of rail fasteners in five state types.

Fastener status		Experimental results					Evaluation index
	Type	Normal	Loosing	Separate	Missing	Dislocated	Precision (%)	Recall (%)	F-measure	Accuracy (%)
Real sample	Normal	780	11	4	0	5	94.43	97.50	0.9594	93.13
	Loosing	18	158	15	0	9	85.41	79.00	0.8208
	Separate	4	9	184	1	2	90.20	92.00	0.9109
	Missing	0	0	0	200	0	99.50	100.00	0.9975
	Dislocated	24	7	1	0	168	91.30	84.00	0.8750