Lightweight Single-Stage Ship Object Detection Algorithm for Unmanned Surface Vessels Based on Improved YOLOv5.

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Object detection is crucial across various industries, including maritime applications.
• Deploying object detection on resource-constrained devices like unmanned surface vessels necessitates efficient models.
• Existing models often struggle to balance accuracy with low computational load and parameter count.

Purpose of the Study:

• To develop an efficient object detection network for ship detection tasks on devices with limited computational resources.
• To improve the accuracy of small object detection in maritime environments.
• To reduce the parameter count and computational load of object detection models without compromising performance.

Main Methods:

• An enhanced ShuffleNetV2 network was used as the backbone for the YOLOv5-based object detection model.
• A novel split-DLKA module was integrated into the small object detection layer to boost accuracy.
• The WIOUv3 loss function was employed to mitigate the impact of low-quality data samples.

Main Results:

• The proposed method achieved a 71% reduction in parameters and a 58% decrease in computational load compared to YOLOv5s.
• Mean Average Precision (mAP) at IoU threshold 0.5 (mAP@0.5) increased by 3.9%.
• mAP at varying IoU thresholds from 0.5 to 0.95 (mAP@0.5:0.95) improved by 3.3%.

Conclusions:

• The developed object detection network offers a significant improvement in efficiency and accuracy for ship detection.
• The model is well-suited for real-time applications on edge devices with limited computational power.
• This research contributes to advancing efficient deep learning models for maritime surveillance and autonomous navigation.