Year: 2025
Type: article
Abstract: With the rapid growth of the chemical industry, ensuring personnel safety in chemical parks is vital. Traditional monitoring systems relying on human surveillance are often inefficient and inaccurate. To address these challenges, we present YOLO8-ARR, a real-time detection and identification system based on an advanced AI algorithm. Key features include: (1) APIoU, optimizing bounding box regression to enhance localization; (2) RCPCA, a novel attention mechanism for better background extraction; (3) RFAConv, replacing traditional convolution to highlight important details; (4) a bidirectional feature pyramid network (BiFPN) for multi-scale feature fusion; and (5) an additional small object detection layer in YOLOv8 to improve small target detection. Experiments on a custom dataset show YOLO8-ARR increases mean Average Precision ([email protected]) by 5.475% over the standard model. This system accurately detects abnormal behaviors and risks, reduces reliance on human resources, minimizes errors, and enhances monitoring efficiency and safety.
Authors: Zhong Wang, Lanfang Lei, Tong Li, Peibei Shi
Institutions: Hefei Normal University, Hefei University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4410384407