PSD-YOLO: An Enhanced Real-Time Framework for Robust Worker Detection in Complex Offshore Oil Platform Environments

Area of Science:

• Artificial Intelligence
• Computer Vision
• Industrial Safety Engineering

Background:

• Offshore drilling platforms present complex safety challenges due to hazardous environments.
• Existing monitoring systems often struggle with real-time detection of personnel in cluttered or occluded conditions.

Purpose of the Study:

• To develop an enhanced, real-time object detection framework for improving personnel safety on offshore platforms.
• To address limitations in detecting small, occluded, or dense target scenarios.

Main Methods:

• Integration of a Channel Attention-Aware (CAA) mechanism to reduce background noise.
• Introduction of a C2fCIB_Conv2Former module for improved multi-scale feature fusion.
• Utilization of the Soft-Non-Maximum Suppression (Soft-NMS) algorithm to minimize missed detections.

Main Results:

• The Platform Safety Detection YOLO (PSD-YOLO) framework achieved a mean Average Precision (mAP@0.5) of 82.5%.
• The system demonstrated a high inference speed of 232.56 Frames Per Second (FPS).
• Significant reduction in missed detections in dense scenes was observed compared to traditional methods.

Conclusions:

• PSD-YOLO offers an efficient and accurate solution for automated safety monitoring in offshore environments.
• The framework provides critical technical support for real-time hazard warnings, enhancing personnel safety.
• This advancement contributes to intelligent sensor monitoring and overall safety system improvements for offshore platforms.