LE-YOLOv5: A Lightweight and Efficient Road Damage Detection Algorithm Based on Improved YOLOv5

Table 6

Multiple algorithms’ comparison.

Models	Params (M)	GFLOPS	Precision (%)	Recall (%)	F1-score (%)	FPS	mAP0.5 (%)
YOLOv5l	46.5	109.1	61.6	55.3	58.3	42	57.4
Faster R-CNN-ResNet50	41.48	94.3	57.0	52.2	54.5	9	50.3
DETR-ResNet50	36.74	100.9	48.7	38.3	42.8	36	39.7
SSD-VGG16	26.79	60.9	61.7	49.8	55.1	15	53.2
YOLOv5s	7.20	16.5	57.5	50.5	53.8	102	51.6
YOLOv7-tiny	6.02	13.2	51.4	52.3	51.8	111	49.1
LE-YOLOv5	3.41	7.0	63.9	53.2	58.1	71	56.9

The best results for each column in the table are bolded. The FPSs of all the abovementioned models are calculated on the basis of an input image resolution of 640 pixels × 640 pixels.