Enhanced Precision in Dam Crack Width Measurement: Leveraging Advanced Lightweight Network Identification for Pixel-Level Accuracy

Table 4

Ablation studies.

Model	MIoU (%)	IoU (%)	mRecall (%)	mPrecision (%)
DL	81.12	62.29	86.96	88.76
R + DL	81.14	62.83	87.34	89.63
C + DL	81.04	62.64	87.05	89.80
F + DL	80.59	61.69	87.71	89.35
L + DL	82.19	64.89	87.98	90.57
R + C + DL	81.26	63.07	87.87	89.24
R + L + DL	82.72	65.96	89.96	89.51
C + L + DL	82.78	66.07	89.13	90.20
R + C + L + DL	82.85	66.21	89.85	89.57
F + C + DL	80.59	61.69	86.25	89.96
F + L + DL	82.72	65.96	90.41	89.48
F + C + L + DL	83.23	66.96	90.47	89.51

The bold font is to highlight that the optimized algorithm in this paper performs better than any of the other optimizations.