Table 4 Results for evaluation of processing methods for YOLO11 and RT-DETR based models.
From: Inspection of railway catenary systems using machine learning with domain knowledge integration
YOLO11 based methods | |||||
|---|---|---|---|---|---|
AP50 | Recall | Precision | F1 | Time | |
Basic elements | Full inference | ||||
1 | 88.95 (1.17) | 91.45 (1.12) | 81.06 (1.80) | 85.88 (0.95) | 0.507 (0.016) |
2 | 89.44 (1.64) | 92.12 (1.56) | 80.93 (2.28) | 86.11 (1.56) | 1.839 (0.037) |
3 | 89.44 (1.64) | 92.12 (1.56) | 80.93 (2.28) | 86.11 (1.56) | 0.703 (0.055) |
3+D | 89.33 (1.74) | 91.45 (1.59) | 85.60 (1.63) | 88.39 (1.26) | 0.823 (0.014) |
3+M+D | 89.33 (1.74) | 91.45 (1.59) | 85.60 (1.63) | 88.39 (1.26) | 0.894 (0.018) |
Small elements | |||||
1 | 35.97 (2.55) | 81.80 (4.12) | 22.44 (1.63) | 35.10 (2.09) | |
2 | 70.87 (2.21) | 90.84 (2.21) | 38.40 (2.65) | 53.33 (2.61) | |
3 | 68.32 (2.68) | 90.42 (1.07) | 51.85 (1.50) | 65.79 (1.29) | |
3+M | 69.50 (1.90) | 89.29 (1.41) | 55.87 (1.95) | 68.67 (1.65) | |
3+M+D | 68.91 (1.72) | 88.63 (1.64) | 56.58 (1.87) | 69.02 (1.66) | |
RT-DETR based methods | |||||
|---|---|---|---|---|---|
AP50 | Recall | Precision | F1 | Time | |
Basic elements | Full inference | ||||
1 | 89.53 (1.33) | 92.74 (1.12) | 67.83 (3.90) | 78.24 (2.83) | 0.745 (0.008) |
2 | 90.83 (2.18) | 94.30 (1.53) | 63.21 (5.24) | 75.37 (4.07) | 3.948 (0.021) |
3 | 90.83 (2.18) | 94.30 (1.53) | 63.21 (5.24) | 75.37 (4.07) | 1.632 (0.161) |
3+D | 90.73 (2.16) | 93.60 (1.59) | 72.26 (3.47) | 81.41 (2.56) | 1.839 (0.071) |
3+M+D | 90.73 (2.16) | 93.60 (1.59) | 72.26 (3.47) | 81.41 (2.56) | 1.731 (0.040) |
Small elements | |||||
1 | 42.43 (2.84) | 89.28 (3.48) | 18.21 (1.21) | 30.23 (1.82) | |
2 | 80.82 (2.29) | 92.95 (2.20) | 26.41 (4.02) | 40.39 (5.06) | |
3 | 80.14 (2.81) | 93.78 (1.10) | 40.61 (4.59) | 56.19 (4.71) | |
3+M | 77.88 (2.40) | 92.18 (1.76) | 43.61 (3.36) | 58.73 (3.36) | |
3+M+D | 77.50 (2.35) | 91.63 (1.79) | 45.73 (2.58) | 60.60 (2.56) | |
