GMS-YOLO: An Algorithm for Multi-Scale Object Detection in Complex Environments in Confined Compartments.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Confined compartments present safety hazards (e.g., loose fasteners, object intrusion) due to limited space, complicating manual inspections.
• Existing inspection methods struggle with complex environments, diverse targets, and varying object scales within these confined areas.

Purpose of the Study:

• To develop an advanced object detection network, GMS-YOLO, for enhanced safety inspections in challenging confined compartment environments.
• To improve accuracy and efficiency in identifying potential hazards within confined spaces.

Main Methods:

• Proposed a novel GMS-YOLO network based on the YOLOv8 framework, incorporating GhostHGNetv2 for superior feature extraction.
• Integrated Multi-Scale Convolutional Attention (MSCA) to address varying target scales and a Shared Convolutional Detection Head (SCDH) for lightweight, high-accuracy detection.

Main Results:

• The GMS-YOLO network demonstrated a 37.8% reduction in parameters and a 27.7% decrease in GFLOPs compared to the original model.
• Achieved an increase in average accuracy from 82.7% to 85.0% in object detection tasks within confined compartments.

Conclusions:

• The GMS-YOLO network offers a lightweight and accurate solution for object detection in complex, confined environments.
• The proposed method significantly enhances safety inspection capabilities by improving hazard identification accuracy and efficiency.