MS-YOLO: A Lightweight and High-Precision YOLO Model for Drowning Detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Marine Safety

Background:

• Drowning remains a significant cause of accidental death globally.
• Existing detection systems often lack the precision and efficiency required for real-time rescue operations.
• Intelligent hardware platforms require lightweight and accurate detection models.

Purpose of the Study:

• To develop a novel, lightweight, and high-precision detection model named MS-YOLO.
• To enhance the efficiency of drowning rescue operations through improved object detection.
• To ensure the model's applicability on intelligent hardware platforms.

Main Methods:

• Introduced the MD-C2F structure with dynamic convolution (DcConv) for capturing subtle aquatic movements.
• Incorporated the EMA mechanism for improved small object detection within the MD-C2F structure.
• Developed the MSI-SPPF module for multi-scale feature identification and complex background understanding.
• Replaced ConCat fusion with BiFPN weighted channel fusion for retaining relevant drowning features.

Main Results:

• Achieved an average detection accuracy of 86.4% on a self-built dataset.
• Operates at an ultra-low computational cost of 7.3 GFLOPs.
• Demonstrated superior performance compared to Faster R-CNN, SSD, YOLOv6, YOLOv9, and YOLOv10.

Conclusions:

• MS-YOLO significantly improves drowning detection efficiency and accuracy.
• The model's lightweight design and low computational cost make it suitable for intelligent hardware.
• This advancement holds potential for enhancing water safety and reducing drowning incidents.