Low-Light Target-Detection Algorithm Combined with Image Enhancement

Fig. 10. Visual comparison between YOLOv8n and LowLight-YOLOv8n. (a), (c), (e), (g) are the detection results of YOLOv8n, respectively; (b), (d), (f), (h) are the detection results of LowLight-YOLOv8n, respectively

Download full size

Table 1. Training parameters
View table
Table 1. Training parameters
Parameter Value
Image size 640 pixel×640 pixel
Learning rate 0.01
Epochs 200
Batch size 16
Weight decay 0.0005
Momentum 0.937

Table 2. Comparative experiment results of different algorithms

View table

Table 2. Comparative experiment results of different algorithms

Method	P /%	R /%	mAP@0.5 /%	mAP@0.5∶0.95 /%	Number of parameters /10⁶	FLOPs /10⁹	Model size /MB
Faster R-CNN	70.0	56.3	63.5		41.00	92.95
YOLOv5n	69.3	60.0	65.8	40.0	2.51	7.10	5.3
YOLOv6n	70.2	60.5	66.6	41.7	4.23	11.80	8.7
YOLOv7-Tiny	72.3	61.2	67.0	41.3	6.02	13.20	12.2
YOLOv8n	69.3	62.4	68.1	42.3	3.01	8.10	6.3
YOLOv9t	69.3	61.4	67.7	42.6	2.01	7.60	4.6
YOLOv10n	68.7	58.4	64.8	39.7	2.70	8.20	5.8
RTDETR-l	76.1	59.8	66.6	40.8	32.00	103.50	66.2
MAET^［11］			74.0		65.32	215.91
DK_YOLOv5^［26］	75.0	62.2	71.5	47.0
Dark-YOLO^［27］			74.7
DarkYOLOv8^［28］			70.1			8.53
Ours	78.7	66.3	74.9	48.1	2.92	18.10	6.3

Table 3. Comparison experiment results of low light enhancement networks
View table
Table 3. Comparison experiment results of low light enhancement networks
Method Number of parameters /10⁶ mAP@0.5 /%
YOLOv8n 3.01 68.1
YOLOv8n+Zero-DCE^［29］ 3.02 68.2
YOLOv8n+RUAS^［30］ 4.62 68.4
YOLOv8n+SCINet^［31］ 3.01 68.8
YOLOv8n+Retinexformer^［18］ 3.08 69.6

Table 4. Comparison of image quality metrics for enhancement networks
View table
Table 4. Comparison of image quality metrics for enhancement networks
Method PSNR /dB SSIM
Zero-DCE 23.13 0.73
RUAS 21.67 0.67
SCINet 24.51 0.74
Retinexformer 25.96 0.81

Table 5. Ablation experiment results

View table

Table 5. Ablation experiment results

Baseline model	Retinexformer	RFCAConv	C2f_UniRepLKNetBlock	Focaler-CIoU	mAP@0.5 /%	mAP@0.5∶0.95 /%	Number of parameters /10⁶	FLOPs /10⁹	Model size /MB
YOLOv8n					68.1	42.3	3.01	8.1	6.3
	√				69.6	43.5	3.08	18.3	6.5
		√			69.3	43.1	3.06	8.3	6.4
			√		68.7	42.2	2.79	7.7	5.9
				√	69.4	42.8	3.01	8.1	6.3
	√	√			71.8	44.3	3.13	18.5	6.6
	√	√	√		72.2	44.5	2.92	18.1	6.3
	√	√		√	73.5	46.7	3.13	18.5	6.6
	√	√	√	√	74.9	48.1	2.92	18.1	6.3

Tools

Get Citation

Copy Citation Text

Xiaodi Zhang, Shijie Jia. Low-Light Target-Detection Algorithm Combined with Image Enhancement[J]. Laser & Optoelectronics Progress, 2025, 62(12): 1237010

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites