Real-time early warning of water inrush via machine-vision-enabled two-stage electrical monitoring
SUN Wenbin
DING Mingkang
LIANG Lijuan
YANG Hui
ZHANG Xiaobo
CHU Yixin
Abstract：The deep mining working face is characterized by high in-situ stress and high water pressure,making water in-rush hazards increasingly severe.With the rapid development of artificial intelligence and the continuous advancement of smart mining construction,deep learning and computer vision have been widely applied in mine safety monitoring,provid-ing new approaches for mitigating water inrush risks.Aiming at the need for real-time water inrush early warning in deep mines,this study innovatively integrates deep learning,machine vision,and electrical monitoring,proposing a real-time two-stage electrical monitoring method based on machine vision.The monitoring process consists of two consecutive stages:the visual recognition stage and the analysis and decision-making stage.In the visual recognition stage,the light-weight deep learning model YOLO11s is used to rapidly and accurately identify low-resistivity regions in electrical im-ages and obtain their spatial coordinates.These coordinates are then passed into the analysis and decision-making stage,where an OpenCV-HSV color filtering algorithm analyzes the identified low-resistivity regions from three aspects—quant-ity,area ratio,and color depth—to comprehensively determine whether the regions pose potential water inrush risks,trig-gering multi-level warning responses through an automatic decision mechanism.Results show that the YOLO11s model can accurately assess whether water sources may cause water inrush hazards and automatically trigger multi-level warn-ings,overcoming the limitations of other monitoring methods in real-time detection,disaster evolution analysis,and early warning.The YOLO11s model achieved a mean Average Precision(mAP)of 90.2%,with an average inference time of 34.6 ms per frame and an average processing rate of 28 Frames Per Second(FPS),outperforming YOLOv8,YOLOv5,and Fast R-CNN in detection accuracy and speed.The OpenCV-HSV color filtering algorithm required only 7 ms per image for analysis and decision-making,realizing real-time early warning of water inrush hazards.
Keywords：electrical monitoringtwo-stagedeep learningmachine visionreal-time water-inrush early warningYOLO11sOpenCV-HSV color filtering algorithm
Publication Date：2025-12-31
Online Publishing Date：2026-01-16(First online date of this platform, not the publication date of the document)
Pages：14( 5286-5299 )