YOLO-HPSD: A high-precision ship target detection model based on YOLOv10.
Ship target detection is crucial in maritime traffic management, smart ports, autonomous ship systems, environmental monitoring, and ship scheduling. Accurate detection of various ships on the water can significantly enhance maritime traffic safety, reduce accidents, and improve the efficiency of po...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Public Library of Science (PLoS)
2025-01-01
|
| Series: | PLoS ONE |
| Online Access: | https://doi.org/10.1371/journal.pone.0321863 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Ship target detection is crucial in maritime traffic management, smart ports, autonomous ship systems, environmental monitoring, and ship scheduling. Accurate detection of various ships on the water can significantly enhance maritime traffic safety, reduce accidents, and improve the efficiency of port and waterway management. This study proposes a high-precision ship target detection algorithm based on YOLOv10, named YOLO-HPSD (High-precision Ship Target Detection). To meet the high-precision requirements in practical applications, several precision-enhancement strategies are introduced based on YOLOv10. To optimize the feature fusion process, the Iterative Attentional Feature Fusion (iAFF) is integrated with the C2F module in the backbone, resulting in the development of a novel C2F_iAFF module that utilizes a multi-scale channel attention mechanism. Meanwhile, the Mixed Local Channel Attention (MLCA) is introduced after the C2F module at the network neck, which improves the model's ability to integrate both local and global information. Additionally, the BiFPN module is incorporated after the connection operation at the network neck, utilizing learnable weights to optimize the importance of different input features, thereby further enhancing multi-scale feature fusion. The experimental results demonstrate that YOLO-HPSD achieves excellent detection performance on the ship dataset, with an F1-score of 97.88% and mAP@0.5 of 98.86%. Compared to YOLOv10n, the F1-score, and mAP@0.5 have improved by 1.22% and 0.31%, respectively. Furthermore, the detection time for a single image is only 20.6 ms. These results indicate that the model not only ensures high detection speed but also delivers high-accuracy ship target detection. This study provides technical support for real-time ship target detection and the development of edge computing devices. |
|---|---|
| ISSN: | 1932-6203 |