Fig. 1. Detection flow chart of track damage image

Fig. 2. Sample images expanded by different operations. (a) Original images; (b) ground truth; (c) translation transformation; (d) rotation transformation; (e) scaling transformation

Fig. 3. Structure of improved U-Net convolution neural network

Fig. 4. Performance curves of proposed method in model training process

Fig. 5. ROC curve of proposed method

Fig. 6. Detection results of different methods

Fig. 7. Visualization results of different methods. (a) Defect images; (b) ground truth; (c) LN+DLBP; (d) MLC+PEME; (e) DWT; (f) CFE; (g) U-Net; (h) proposed method