Railway Subgrade Defect Automatic Recognition Method Based on Improved Faster R-CNN

Table 6

The results of 10-fold cross-validation experiment.

Method	Dataset										Average
	1	2	3	4	5	6	7	8	9	10
HOG+SVM	37.7	38.6	38.9	39.5	36.9	37.9	39.3	38.3	38.8	39.4	38.5
Faster-RCNN	67.2	67.5	66.8	67.8	68.3	67.9	67.3	67.8	67.5	68.1	67.6
Proposed method	82.9	84.4	83.7	82.9	82.3	84.5	84.8	83.7	83.8	82.6	83.6