TSAS—YOLOv8: An Optimization Detection Model for Capturing Small Target Features and Processing Key Information
In object detection tasks, small targets are prone to losing critical information during feature extraction by traditional convolutional layers due to their tiny size and sparse features. This not only reduces the detection accuracy but also undermines the model’s generalization performan...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10976638/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In object detection tasks, small targets are prone to losing critical information during feature extraction by traditional convolutional layers due to their tiny size and sparse features. This not only reduces the detection accuracy but also undermines the model’s generalization performance. To address this issue, we propose the TSAS-YOLOv8 method. This method defines the T-CMUNeXt module and introduces a new Backbone structure based on this module, leveraging its multi-scale feature fusion advantage to capture the subtle features of small targets. Additionally, it employs the Shuffle Attention mechanism to focus on key features, enhancing the detection effect of small targets in all aspects. Furthermore, a specialized small target detection head is added to improve the localization capability for small targets. Validated on the VisDrone2019 dataset, TSAS-YOLOv8 achieves a 7.6% increase in mAP50 and a 4.8% increase in mAP50-95 compared to the original YOLOv8 model, demonstrating superiority even when compared to the latest models such as YOLOv10. On our self-made airport runway foreign object debris dataset FOD-x, TSAS-YOLOv8 shows a 1.8% increase in mAP50 and a 2.2% increase in mAP50-95, proving its good generalization in specific scenarios. In the experiment verifying generalization ability, the TSAS - YOLOv8 model performed outstandingly on the COCO128 dataset, demonstrating its good generalization.Ablation experiments verify that both individual and combined use of each module has a positive effect on performance improvement. In summary, TSAS-YOLOv8 exhibits efficiency and feasibility in small target detection tasks. |
|---|---|
| ISSN: | 2169-3536 |