An Improved YOLOv9 Based Object Detection with Attention Mechanism for Personal Protective Equipment

Area of Science:

• Computer Vision
• Industrial Safety Engineering

Background:

• Industrial environments present significant safety hazards, necessitating robust worker protection through personal protective equipment (PPE).
• Manual verification of PPE compliance (safety helmets, shoes, gloves) is impractical in large-scale industrial operations.
• Automated detection systems are crucial for real-time monitoring and enforcement of safety protocols.

Purpose of the Study:

• To develop an efficient and accurate automated system for detecting the usage of essential PPE on industrial sites.
• To enhance the YOLOv9 deep learning architecture for improved performance in industrial safety applications.

Main Methods:

• Integration of the Convolutional Block Attention Module (CBAM) into the training-only auxiliary branch of YOLOv9's Programmable Gradient Information (PGI) architecture, creating the CBAMLinear module.
• The CBAMLinear module enhances training gradients without impacting inference-time computational overhead, as the auxiliary branch is removed post-training.
• Evaluation of the proposed method on industrial safety datasets, focusing on the detection of safety helmets, shoes, and gloves.

Main Results:

• The proposed YOLOv9 with CBAMLinear achieved consistent mean Average Precision (mAP@0.5:0.95) gains of 0.005-0.007 over the baseline for larger model variants.
• The method maintained identical inference-time parameters and FLOPs compared to the baseline YOLOv9 architecture.
• Demonstrated effectiveness in improving detection accuracy for critical PPE items.

Conclusions:

• The CBAMLinear-enhanced YOLOv9 offers a computationally efficient solution for automated PPE detection in industrial safety.
• Even marginal improvements in detection accuracy can significantly reduce industrial accidents by minimizing false positives and negatives.
• The approach is well-suited for real-time safety management systems, contributing to safer industrial work environments.