DySC-YOLOv8: Pollutant Identification Algorithm Designed for Building Facade

Fig. 8. Comparison of detection results between YOLOv8n and DySC-YOLO. (a) Original images; (b) detection results by YOLOv8n;(c) detection results by DySC-YOLOv8

Download full size

Table 1. Set of hyperparameters
View table
Table 1. Set of hyperparameters
Hyperparameter Value
Learning rate 0.01
Momentum 0.937
Weight decay 0.0005
Batch size
Epoch
8
300

Table 2. Results of ablation experiments

View table

Table 2. Results of ablation experiments

Model	mAP@0.5 /%	Params /10⁶	FLOPs /10⁹	Detection speed /（frame·s^-1）
YOLOv8n	88.4	3.01	8.1	92
YOLOv8n+DySample	91.2	3.01	8.1	90
YOLOv8n+ SCAM	88.9	2.78	7.6	93
YOLOv8n+SIoU	89.1	3.01	8.1	101
YOLOv8n+DySample+SIoU	93.5	3.01	8.1	89
YOLOv8n+DySample+SCAM	89.7	2.78	7.4	95
YOLOv8n+SCAM+SIoU	94.4	2.76	7.6	94
YOLOv8n+DySample+SCAM+SIoU	94.5	2.71	7.3	93

Table 3. Results of different detection algorithms

View table

Table 3. Results of different detection algorithms

Algorithm	P /%	mAP@0.5 /%	Params /10⁶	FLOPs /10⁹	Detection speed /（frame·s^-1）
SSD FasterR-CNN YOLOv5s	86.2 83.1 88.9	72.7 78.6 86.3	30.30 41.30 7.21	61.5 271.2 15.4	48 19 75
YOLOv8n YOLOv8-FN-T0	89.3 94.3	91.9 92.7	3.15 2.91	105.1 4.4	70 95
YOLOv8s	92.4	95.5	11.10	26.7	84
DySC-YOLOv8	96.4	94.9	3.42	7.2	96

Table 4. Experimental results based on UAV-DT dataset
View table
Table 4. Experimental results based on UAV-DT dataset
Algorithm P R mAP@0.5 mAP@0.5∶0.95
YOLOv5s 90.4 82.3 89.9 55.7
YOLOv7 89.9 83.0 89.4 56.2
YOLOv8-FN-T0 93.1 87.5 94.4 63.1
YOLOv8n 92.8 87.6 94.2 62.2
DySC-YOLOv8 93.2 89.5 95.7 65.3

Table 5. Experimental results based on DIOR dataset
View table
Table 5. Experimental results based on DIOR dataset
Algorithm P R mAP@0.5 mAP@0.5∶0.95
YOLOv5s 79.1 72.3 77.1 46.3
YOLOv7 82.8 67.2 73.6 50.8
YOLOv8-FN-T0 83.9 71.1 78.0 53.2
YOLOv8n 84.5 73.1 79.9 54.4
DySC-YOLOv8 85.1 74.4 81.2 56.1

Tools

Get Citation

Copy Citation Text

Zhijun Gao, Kexun Li, Zhenbo Li. DySC-YOLOv8: Pollutant Identification Algorithm Designed for Building Facade[J]. Laser & Optoelectronics Progress, 2025, 62(14): 1415007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites