Intelligent monitoring of small target detection using YOLOv8
In complex scenes, small target face detection is crucial but often hampered by detection accuracy and efficiency limitations. Our method addresses these challenges by incorporating Gaussian noise, which is key in improving model robustness and generalization. By simulating real-world imperfections,...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-01-01
|
| Series: | Alexandria Engineering Journal |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1110016824012791 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In complex scenes, small target face detection is crucial but often hampered by detection accuracy and efficiency limitations. Our method addresses these challenges by incorporating Gaussian noise, which is key in improving model robustness and generalization. By simulating real-world imperfections, Gaussian noise acts as a regularizer and makes the model more resistant to variations in lighting and texture. Traditional methods often face difficulties when dealing with small targets and complex backgrounds due to inadequate feature extraction, suboptimal loss function design, and vulnerability to noise. To overcome these issues, we propose an improved YOLOv8 model based on multi-scale feature fusion and an optimized loss function. By leveraging Gaussian noise during training, our approach enhances both detection accuracy and operating efficiency. Experiments on the FDDB and WIDER FACE datasets demonstrate that our method performs better in various complex scenarios. Our method achieved 0.780 on the WIDER FACE validation set, outperforming existing mainstream techniques. |
|---|---|
| ISSN: | 1110-0168 |