Improved YOLOv8n based helmet wearing inspection method.

Area of Science:

• Computer Vision
• Deep Learning
• Object Detection

Background:

• Contemporary helmet-wearing target recognition algorithms suffer from parameter redundancy, slow inference, and suboptimal precision.
• Existing models face challenges in efficiency and accuracy for real-world applications.

Purpose of the Study:

• To propose an optimized YOLOv8n_H method for enhanced helmet-wearing target recognition.
• To address parameter redundancy, inference speed, and detection precision issues in current algorithms.

Main Methods:

• Enhanced YOLOv8 C2f module with SC_Bottleneck and SCConv (SC_C2f) to reduce complexity.
• Replaced original Detect structure with PC-Head decoupling head for efficiency.
• Incorporated WIoU boundary loss function with dynamic non-monotonic focusing mechanism.

Main Results:

• Achieved a 46.63% reduction in model volume and 44.19% decrease in parameter count.
• Reduced computational load by 54.88% and improved mean Average Precision (mAP) to 93.8%.
• Demonstrated superior detection accuracy, reduced model size, and lower computational load compared to original YOLOv8.

Conclusions:

• The proposed YOLOv8n_H method significantly optimizes helmet detection performance.
• The enhancements lead to a more efficient and accurate model for recognizing helmet-wearing individuals.
• This approach offers a practical solution for real-time applications requiring high detection precision.