Pavement Disease Detection through Improved YOLOv5s Neural Network

Table 3

Comparison of different model performance metrics.

Model	Recall/%	Accuracy/%	mAP/%
YOLOv3	51.43	82.16	73.93
YOLOv3 + Four scale layers	52.86	83.74	75.06 (73.93 + 1.13)
YOLOv5s	72.31	92.31	84.77
ShuffleNetv2	69.92	82.57	77.48
MobileNetv3	71.62	81.45	79.35
YOLOv5s + Mosaic	72.42	92.94	85.32 (84.77 + 0.55)
YOLOv5s + Ghost_backbone	73.35	92.79	85.93 (84.77 + 1.16)
YOLOv5s + Ghost_backbone + Mosaic	73.38	93.17	86.44 (84.77 + 1.67)
YOLOv5s + ShuffleNetv2	73.27	92.58	85.30 (84.77 + 0.53)
YOLOv5s + ShuffleNetv2 + Mosaic	73.44	93.21	85.71 (84.77 + 0.94)
YOLOv5s + Mobilenetv3	74.08	93.01	85.88 (84.77 + 1.11)
YOLOv5s + Mobilenetv3 + Mosaic	74.56	93.41	86.29 (84.77 + 1.52)
Our Ghost-YOLOv5s	76.39	95.18	87.69 (84.77 + 2.92)
Our Ghost-YOLOv5s + Mosaic	76.97	95.56	88.17 (84.77 + 3.40)