Labor protection article detection method and system based on improved YOLOv5s

By improving the SLSKA-POOL, CAKConv, and EMA attention mechanisms of the YOLOv5s framework, the problems of deformation, occlusion, and complex backgrounds in the inspection of personal protective equipment were solved, achieving higher accuracy in inspection.

CN122391944APending Publication Date: 2026-07-14PETROCHINA CO LTD +1
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
PETROCHINA CO LTD
Filing Date
2026-03-24
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing technologies for detecting personal protective equipment in dynamic and complex industrial scenarios suffer from insufficient robustness to target deformation and occlusion, sensitivity to small targets and complex backgrounds, low efficiency in utilizing model features, and limited convergence of loss functions and positioning accuracy, resulting in insufficient detection accuracy.

Method used

An improved YOLOv5s framework is adopted, which enhances the ability to extract small target shape features through the SLSKA-POOL module, adaptively extracts target deformation features through the CAKConv module, aggregates multi-scale spatial structure information through the EMA attention mechanism, and optimizes bounding box regression using the EIOU loss function.

Benefits of technology

It improved the testing accuracy of personal protective equipment in dynamic and complex environments, significantly reduced missed detections and false detections, increased the testing accuracy by 2.2%, and achieved more efficient testing of personal protective equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122391944A_ABST
    Figure CN122391944A_ABST
Patent Text Reader

Abstract

The application provides a labor protection article detection method and system based on improved YOLOv5. The method comprises the following steps: acquiring a production monitoring video stream and decoding; using a pre-constructed construction operator detection model to perform target detection on each frame of image to obtain at least one target detection frame including a construction operator, and assigning a unique ID to the target in the detection frame; continuously tracking the same construction operator in the continuous image frames based on the target detection frame and the corresponding ID; for the tracked construction operator, using a pre-constructed head detection model to locate the head region in the corresponding target detection frame to generate a head detection frame; classifying the image region in the head detection frame by using a pre-constructed labor protection article classification model to determine whether the construction operator wears labor protection articles in the current frame; and counting the wearing judgment result of the labor protection articles of the construction operator with the same ID in the continuous N frames, and selecting whether to trigger an alarm according to the result.
Need to check novelty before this filing date? Find Prior Art